Critical Theory in the Age of Digintelligenceデジタルインテリジェンス時代の批判理論数智时代的批判理论

Introduction to the Critical Theory in the Age of Digintelligence Summer Reading Group
At the height of Fordism and industrial capitalism—the early or even "classical" phase of modernity familiar to us—large-scale machine production rested on rigorous spatial enclosure and the linear division of time. Capital forcibly concentrated labour in standardized factory production lines, physically disciplining the proletariat in the dual dimensions of body and time. This mode of exploitation, based on dense spatial concentration, objectively furnished an indispensable organizational precondition for classical class solidarity, trade-union movements and the awakening of collective political consciousness. Workers' physical co-presence under one roof transformed the visibility of exploitation into the cohesion of resistance. Yet with the advent of the post-Fordist phase dominated by cybernetics, the topology of exploitation underwent a fundamental ontological transformation. The logic of capital accumulation no longer relied solely on the direct extraction of absolute surplus value in physical space, but turned toward the continuous extraction of immaterial labour, cognitive surplus and the "behavioural surplus" persistently captured by digital platforms.
Within today's digintelligent platforms and algorithmic capitalism, the physical factory in its traditional sense has been thoroughly dismantled and reorganized. Labour is dispersed by APIs, crowdsourcing platforms, outsourced data-labelling networks and algorithms for allocating gig work, taking on a highly distributed, black-boxed and decentralized form. The medium of control has shifted from visible masonry walls and the supervisor's gaze into invisible TCP/IP protocols, deep packet inspection (DPI) and covertly operating predictive machine-learning models. Against this backdrop, traditional struggles over physical space, exemplified by Occupy Wall Street, face a serious structural mismatch at the instrumental and epistemological levels, and cannot avoid losing their efficacy before the underlying network architecture in which "code is law." If exploitation no longer takes the form of direct violent coercion, but becomes interface interactions and personalized recommendations seamlessly embedded in everyday life, do the physical points of purchase for resistance dissolve along with it?
This summer reading group seeks to suspend conventional humanist moral criticism and overly simple doctrines of technological refusal, directing theoretical inquiry into protocols, cryptographic infrastructures, the planetary computational Stack and the micropolitical networks through which heterogeneous subjects operate. The "age of digintelligence" is not a convenient label for technological progress, but a field of problems that must be historicized and theorized. Digital platforms, machine learning, cloud computing, data centres, automated management and generative AI are rearranging relations among capital accumulation, state governance, mediation, labour organization, knowledge production and political subjects. These changes can neither be described simply as a replacement for industrial capitalism nor reduced to old institutions acquiring a technological interface. More pressing questions concern which institutional relations have had their scale, speed and visibility changed by computation; which forms of power have merely gained more fine-grained operating conditions through new media; and which phenomena labelled "ruptures" actually continue the long histories of management science, advertising, colonial governance, infrastructure and the division of labour. Digitization and intelligentization thus constitute new objects of research while also forcing critical theory to reconsider the limits of its own concepts.
The first line of inquiry concerns forms of control and the production of computational subjects. From Wiener's feedback and communication to Foucault's analysis of disciplinary apparatuses, and then to Deleuze's modulation, Galloway's protocol and Rouvroy's algorithmic governmentality, these theories do not describe successive upgrades of a single mechanism of power. Discipline depends on spatial partition, timetabling, examination and normalization; protocols organize interoperability through technical standards; algorithmic systems may shape environments of action through classification, scoring, prediction, ranking and recommendation. These mechanisms may overlap, but may also generate conflicts among state jurisdiction, corporate governance and technical architectures. The question is therefore not whether "the algorithm has become the new sovereign," but how computational results enter institutional procedures; which organizations translate probabilistic outputs into eligibility, risk or punishment; who holds the power to interpret, modify and appeal; and how uncertainty within models is solidified into real consequences through organizational processes.
The second line of inquiry addresses surveillance, extraction, coloniality and infrastructuralization. Data are not naturally existing resources awaiting collection; they are continually produced through sensing, logging, labelling, identity systems, platform interfaces, business models and demands for governance. Zuboff's surveillance capitalism, Couldry and Mejias's data colonialism, Browne's historical analysis of racialized surveillance, and Mbembe's political reflections on coloniality, enmity and the exposure of life offer explanatory frameworks that are not fully compatible with one another. Concepts such as "colonial," "feudal" and "imperial" acquire analytical significance only when their corresponding mechanisms can be specified: resource appropriation, legal asymmetry, epistemic authority, compulsory dependence, spatial partition and differentiated risk are not interchangeable rhetorical figures. Likewise, the "cloud" cannot remain an interface metaphor. Data centres, submarine cables, chips, energy, minerals, land, spectrum, standards, logistics and maintenance labour together constitute the material conditions of computation. Platform power must therefore be discussed alongside infrastructure, territory, rent and the costs of exit.
The third line of inquiry concerns the political economy of platforms, labour and new relations of dependence. Varoufakis's and Durand's theses of technofeudalism are examined here in relation to platform capitalism, monopoly, theories of rent and infrastructure studies. What needs to be distinguished is not the rhetorical question of whether "capitalism is dead," but whether cloud gateways, network effects, intangible assets and platform governance produce structures of dependence that traditional theories of capital cannot fully explain, and in what respects platform rent and feudal ground rent are or are not comparable. Research on platform labour brings these macro-level questions back to orders, ratings, dispatching, microtasks, data labelling, content moderation, entrepreneurial subjects and global labour chains. The theory of affect and temporality offered by Berlant's Cruel Optimism is especially important here: why can precarious "freedom," "flexibility" and "self-enterprise" remain objects of attachment that subjects find hard to relinquish, and how might institutional change address ownership, the distribution of risk, lived time and expectations of the future together?
The fourth line of inquiry turns to mediation, publics and organization. Digital politics is often described in terms of falling connection costs, faster information circulation and weakened gatekeeping, yet publicness is not equivalent to traffic, nor organization to instantaneous assembly. Kelty's recursive public asks whether publics can maintain and modify the technical and institutional conditions of their own existence; Warner and Fraser discuss circuits of circulation, counterpublics and unequal participation; Kornbluh's Immediacy, or The Style of Too Late Capitalism critiques the fascination with transparency, immediacy, immersion and disintermediation through an analysis of cultural forms. Political action often depends on procedures, representation, memory, delay, interpretation, chains of responsibility and sustainable organization, while platform media may elevate "immediate visibility, immediate response, immediate circulation" into values in themselves. Decentralization is therefore not inherently democratic, and mediation is not inherently oppressive; the more important question is whether mediation can be held accountable, modified and collectively shaped by participants.
The fifth line of inquiry centres on general intellect, the concentration of computing power and technological futures. In the "Fragment on Machines" in the Grundrisse, Marx explores the tensions among social knowledge, fixed capital, labour time and the value-form; Virno further stresses that general intellect is not only crystallized in machines, but also exists in language, judgement, memory, cooperation and shared capacities. Generative AI brings this problem into view in a new form: model training depends on public texts, code, images, feedback, labelling labour and enormous physical computing resources, while these social conditions can be repackaged as private models, interfaces and cloud services. What requires analysis is therefore not whether "the LLM is general intellect," but through which mechanisms of property and infrastructure the cognitive traces of collective social production are concentrated; how rights of access, training, deployment and returns are distributed; and which layer of relations can be changed by open weights, public computing resources, interoperability and cooperative governance respectively.
The concluding discussion of AI, automation and posthuman futures does not aim to predict a timetable for AGI. Turing, Russell, Bostrom, Brassier, Land, Negarestani, Haraway, Hayles and Braidotti understand "intelligence," "the human," "machines," "control," "emancipation" and "the future" in profoundly different ways; Crawford, Bender and Hanna, Narayanan and Kapoor, and Hao continually bring these philosophical questions back to actual technologies, organizations and resource relations. Philosophical speculation, engineering judgement, industrial forecasting, political economy and empirical social research have different standards of evidence and cannot substitute for one another. Descriptive judgements, causal explanations and normative judgements should likewise remain distinct: exposing a mechanism of power does not automatically entail a single political programme, and the technical feasibility of an institutional arrangement does not establish its legitimacy. In this reading group, we seek to build testable connections among conceptual history, empirical mechanisms and normative questions, so that classical theories are neither revered as universal keys nor deprived, by differences between historical periods, of their capacity to reinterpret the present.
| Session | Core Question |
| Session 1: Cybernetics, Control and the Birth of Digital Governance | How do cybernetics, discipline, societies of control and protocols constitute the historical preconditions of digital governance? |
| Session 2: Algorithmic Governmentality and Biopolitics | How do classification, prediction, scoring and ranking produce computational subjects and acquire institutional efficacy? |
| Session 3: Surveillance, Extraction and Data Coloniality | How can we establish rigorous, rather than merely rhetorical, connections among data extraction, surveillance, racialized classification and coloniality? |
| Session 4: Planetary Computation and Infrastructure Space | How do the cloud, platforms, data centres and infrastructure reorganize space, authority, rent and exit costs? |
| Session 5: Recursive Publics, Mediation and Digital Politics | How can publics possess and modify their own media conditions? Why might mediation and delay be conditions of political organization? |
| Session 6: Platform Labour, Precarity and New Class Relations | How do platform labour, algorithmic management and precarious lives reorganize labour relations and collective action? |
| Session 7: General Intellect and Resistance to Compute Concentration | How is collectively produced social intelligence turned into fixed capital and infrastructure, and how might it be made public again? |
| Session 8: Artificial Intelligence, Automation and Posthuman Futures | How can we distinguish AI's actual capabilities, industrial hype, arguments about risk and visions of posthuman futures? |
Part I: The Topology of Control (Weeks 1–2)
Session 1: Cybernetics, Control and the Birth of Digital Governance
This session begins with a prehistory often overlooked in theories of digital politics: before "algorithm" became an everyday word, the twentieth century had already developed an intellectual vocabulary that redescribed machines, organizations and societies through communication, feedback, information and control. Wiener's cybernetics, Cold War command-and-control, industrial management's demand for information processing, distributed communication networks and software protocols together constitute the historical conditions of digital governance. These conditions do not, however, form a linear chain necessarily leading to contemporary algorithmic governance. Foucault's disciplinary power depends on spatial partition, timetabling, examination and normalization; Deleuze's modulation points to continuous and variable forms of control; Galloway's protocol addresses how distributed networks stipulate operability while permitting connections. Their differences are, first of all, differences of object, scale and mechanism.
Within this genealogy, control cannot be reduced to whether a "centre" exists. Standards, protocols, interfaces, identities, permissions, hardware and law impose constraints at different levels; execution can be highly distributed while standard-setting, resource ownership and rights of modification remain concentrated. Baran's distributed communication is first an engineering design; May's crypto-anarchism is a political manifesto; Berners-Lee's Web proposal originates in an information-management problem; DeNardis shows that technical standards always occupy the intersection of transnational institutions, corporations and state power. The same network topology can serve military resilience, open collaboration, commercial platforms or surveillance systems. Its political character therefore cannot be deduced directly from architectural labels such as "distributed," "open" or "encrypted."
This session thus establishes a multilayered concept of control: technical systems may delimit operability by encoding rules, but may also depend on market concentration, organizational procedures and state jurisdiction to acquire practical efficacy. "Protocol politics" does not replace legal politics; it results from the nesting of standards, software, institutions and sovereignty. This distinction provides the basis for the next session on algorithmic governance. Only after specifying whether rules operate at the level of architecture, protocol, organization or law can we analyse how scoring, prediction and recommendation systems transform behaviour into computational objects.
Required Readings:
- Wiener, Norbert. The Human Use of Human Beings: Cybernetics and Society. Houghton Mifflin, 1950. (Read: Chapters 1–2, pp. 15–47).
[Note: Wiener redescribes relations among humans, machines and social organizations through communication, feedback and control. The aim here is not to treat cybernetics directly as a theory of algorithmic governance, but to examine how a language of knowledge for "managing systems through information and feedback" took shape. It provides a historical background for the protocols and algorithms that follow, while reminding us that some basic imaginaries of what is now called "data-driven governance" did not begin with the Internet but have a longer cybernetic genealogy. More importantly, we should first establish the original text's object and technical level, and only then look for structural homologies; whenever the legal environment, organizational actors or material conditions differ, analogy must be treated as inspiration rather than evidence. In applying this framework to contemporary digital objects, the key is to identify the level at which a similarity occurs; changes to the legal environment, organizational actors or material conditions also alter the concept's explanatory scope. Wiener's The Human Use of Human Beings: Cybernetics and Society further grounds the "historical differences among cybernetics, discipline, societies of control and protocol governance" in institutional relations. The key to digital governance is not whether a centre disappears, but how command, feedback, standards and rights of maintenance are recombined across organizations. Execution can be highly distributed while standard-setting and resource ownership remain concentrated; network topology and political decentralization must therefore be assessed separately. Wiener's significance also lies in his refusal to treat feedback merely as a technical process inside machines: by connecting communication, entropy and organization, he formed a language capable of crossing engineering and social theory. Precisely because this language was highly transferable, it later exerted influence in management, military systems and computational culture far beyond its original engineering context. Rereading cybernetics today is better suited to examining when feedback becomes a principle of management, and when it is mythologized as a universal model of social self-regulation.]
- Foucault, Michel. Discipline and Punish: The Birth of the Prison. Translated by Alan Sheridan. Vintage Books, 1977. Originally published 1975. (Read: Part III, Chapter 3, "Panopticism," pp. 195–228).
[Note: This chapter provides a comparative theoretical starting point. Foucault discusses not "surveillance" as such, but how discipline produces governable bodies through spatial partition, timetabling, examination, record-keeping and the institutionalization of visibility. Panopticism depends not on a watcher constantly staring at someone, but on placing the observed person in a condition of always being potentially visible, thereby transforming external coercion into the subject's self-adjustment. We must maintain some distance when transferring this framework to digital platforms: data systems do not simply reproduce the positioning of bodies in architectural space. Instead, they may use logs, permissions, scores, recommendations and identity systems to turn action into data objects that can be computed and called up again. We can therefore ask: when surveillance shifts from "who is watching me?" to "what systems can record and process my actions?", how must the theory of panopticism be rewritten? The relation between platform datafication and disciplinary apparatuses can only be established through comparison of specific mechanisms, not deduced directly from their shared vocabulary of "visibility." What is genuinely comparable between digital platforms and panopticism is not the appearance of "seeing," but how visibility is institutionalized and enters the formation of subjects. Platform records often serve commercial prediction, risk control and organizational management simultaneously, and their data visibility is not symmetrically available to all actors. Digital surveillance is therefore better understood as a recombination of disciplinary techniques and data infrastructures than as the mere electronic conversion of Bentham's architectural model. Furthermore, digital environments change the temporal structure of disciplinary power. Conventional examinations often occur at definite moments, whereas platform records allow continuous accumulation and retrospective recalculation; a past act that once seemed irrelevant may acquire new meaning under a new model or policy. This retraceable, recomputable visibility means that digital governance both inherits discipline's logic of record-keeping and develops a politics of time different from that of prison architecture.]
- Deleuze, Gilles. "Postscript on the Societies of Control" October 59 (1992): 3–7. Originally published 1990.
[Note: Deleuze's "societies of control" is not an empirical prediction of the Internet, but a highly compressed model of changing forms of power. Its core is the shift from the closed disciplinary mold to continuous modulation; the "dividual" means that an individual can enter different systems as multiple data fragments, credit, passwords and forms of identity. The text should be treated as a generator of questions, rather than translating the dividual directly into a "data profile." To apply it to digital governance today, what most needs to be added is not more technical terminology but the chain of implementation: where rules are encoded, who deploys and maintains them, and which legal and market relations make them effective. Thus, in Deleuze's "Postscript on the Societies of Control," the claim that "distributed topology and political decentralization are not the same thing" is more than a conceptual issue. The key to digital governance is not whether a centre disappears, but how command, feedback, standards and rights of maintenance are recombined across organizations. Execution can be highly distributed while standard-setting and resource ownership remain concentrated; network topology and political decentralization must therefore be assessed separately. Deleuze's distinction between mold and modulation is especially useful for describing power's shift from periodic certification to continuous updating: account status, credit limits, prices, permissions and recommendations can all be adjusted in real time as data flow. Yet "societies of control" is not an umbrella term for every digital phenomenon. Modulation has analytical force precisely because it identifies continuously changing conditions of access and the partitioning of identities, while state law, organizational rules and physical coercion may continue to coexist with them. What is most worth retaining from "societies of control" is not a label for a historical stage, but its account of power's temporality: through passwords, limits, qualifications and constantly updated statuses, control continually adjusts the space of action. Control and discipline are not mutually exclusive. Enclosed institutions such as schools, prisons and corporations remain, but increasingly combine with continuous assessment and data flows across settings.]
- Galloway, Alexander R.. Protocol: How Control Exists after Decentralization. MIT Press, 2004. (Read: Introduction; Chapter 1, "Physical Media"; and Chapter 2, "Form," pp. 1–92).
[Note: Galloway's central contribution is to move the question of control away from a simple "centre-periphery" model. He asks how protocols establish rule relations in decentralized networks: systems such as TCP/IP and DNS allow large numbers of nodes to communicate, while specifying which actions can occur through standardized addresses, formats, permissions and modes of conversion. It is particularly important to distinguish "protocol is control" from "protocol is law": Galloway discusses the political form of technical systems, rather than claiming that state law has been replaced by network protocols. Theoretically, his work takes up Foucault's problem of discipline while providing a conceptual bridge toward algorithmic governance, blockchains and cloud platforms. We should further ask: when platforms write registration, identity, interfaces, ranking and access permissions into technical structures, does control become less dependent on explicit commands? Extending this text to platforms, the cloud or AI requires preserving both historical continuities and differences of mechanism: the former explain why a problem recurs, while the latter set boundaries that concepts cannot cross. Protocol politics thus exhibits an apparently paradoxical structure: the more communication depends on shared standards, the more a network may be distributed at the level of execution while generating new concentrations in standard-setting, naming systems or key resources. Galloway's value lies precisely in restoring "decentralization" from a normative compliment to an analysable technical and political relation, so that topology, governing authority and ownership are no longer conflated. Protocols also shift political conflict into versions, compatibility and the evolution of standards. Who may propose modifications, which changes lead to forks, and whether older nodes can still connect are all questions of governance within technical communities. Protocol is thus not a static rule, but a continually maintained institutional process; control occurs not only during execution, but also when standards are formulated and updated.]
Recommended Readings:
- Lessig, Lawrence. Code and Other Laws of Cyberspace. Basic Books, 1999. (Read: Chapter 7, "What Things Regulate," pp. 85–99).
[Note: The most noteworthy feature of Lessig's account of "code" is precisely that it does not draw a sharp divide between technical systems and legal institutions. He argues that behaviour is shaped jointly by four regulatory forces: law, social norms, markets and architecture. Code on networks can therefore regulate by changing the conditions of action, but its efficacy remains embedded in legal, economic and social relations. Rather than turning "Code is Law" into a slogan, we should analyse how a platform uses passwords, interfaces, authentication, defaults, payment rules and content ranking to transform an act from being "unlawful" into being "impossible to perform at all." This question is fundamental to understanding "architectural power." The political efficacy of technical structures always passes through organizations and institutions. Protocols prescribe interoperability; hardware determines the range of implementation; market concentration shapes exit costs; and state jurisdiction determines when rules carry coercive force. These layers together constitute the actual form of digital control. Lessig's four regulatory forces—law, norms, market and architecture—prevent digital governance from being reduced to "code determines everything." Architecture may make certain actions technically impossible or increase their costs, but the design and execution of code are themselves shaped by corporate business models, legal liability and social norms. What most deserves study in platform governance is how these four forces overlap: defaults, pricing mechanisms, community rules and state regulation often operate together within the same interface. Lessig's framework also suggests that the most stable form of platform power is often not a single mode of regulation, but multiple forces reinforcing one another. Commercial incentives may, for example, drive a particular interface design; the interface then changes users' norms; and law subsequently takes the platform's established technical conditions as its object of governance. The politics of digital architecture arises from this circuit, not from code possessing an independent sovereignty above society.]
- Beniger, James R.. The Control Revolution: Technological and Economic Origins of the Information Society. Harvard University Press, 1986. (Read: Introduction and Chapter 1).
[Note: Beniger situates modern information society within the "crisis of control" brought about by industrialization: as production, transport, distribution and consumption accelerated, established organizational forms became increasingly unable to handle the resulting complexity, prompting the continuing development of information-processing and management technologies. This avoids a narrative in which "the Internet suddenly invented control" and places platforms, logistics and algorithmic scheduling within a longer history of management. The focus is on the economic and organizational demands that control serves. The political efficacy of technical structures always passes through organizations and institutions. This long-term perspective is particularly useful for explaining platform logistics, real-time pricing and algorithmic scheduling: these are not forms of "intelligence" appearing out of nowhere, but managerial responses to continued increases in speed, scale and complexity. Beniger thus reconnects the history of information technology with the organizational history of capitalism. Studying the so-called digital transformation must also involve examining how accelerating production and circulation generate new pressures for coordination, and who has the capacity to undertake that coordination. From the perspective of organizational history, the "crisis of control" also reveals the symbiosis between information processing and expansion of scale. Firms can sustain coordination over a wider area only when they can track inventory, transport, consumers and labour processes; new information technologies both respond to complexity and make further expansion possible. The growth of digital platforms can likewise be placed within this cycle, rather than described as a purely technological breakthrough. Information-processing capacity can therefore itself be understood as a productive condition of organizational expansion: it resolves existing complexity while opening space for centralized coordination on a larger scale. This reciprocal reinforcement between organization and information also makes Beniger an important mediator between the history of industrial management and that of platform governance. It places management technology and the capitalist problem of scale within a single historical framework. The information revolution is thus recast as a result of the co-evolution of organizational capacities and economic structures. This perspective also helps explain why "real-time" capability has become a persistent goal of modern management.]
- Baran, Paul. "On Distributed Communications Networks" IEEE Transactions on Communications Systems 12, no. 1 (1964): 1–9.
[Note: Baran's article should be read as a text in engineering history. Its central problem is how communication systems can improve redundancy and survivability in high-risk environments, not how to establish "decentralization" in a political sense. For precisely this reason, it is well suited to testing later analogies in digital politics: why can the same distributed topology be interpreted as military resilience, network freedom, platform elasticity or even surveillance infrastructure? A technical structure does not carry a single political direction within itself. Through Baran's "On Distributed Communications Networks," the proposition that "distributed topology and political decentralization are not the same thing" enters the organizational and institutional levels. The political efficacy of technical structures always passes through organizations and institutions. At the scale of Baran's "On Distributed Communications Networks," this proposition requires attention to both continuity and transformation: old institutions may gain new modes of operation through new technologies, while new technologies may be reabsorbed into existing institutions. Baran's engineering argument also reminds us that a distributed network is first a solution for reliability, not a political institution. Node redundancy can improve a communication system's capacity to survive destruction, but cannot answer who owns the network, who sets its protocols or who can cut off access. Later interpretations that directly translate "distributed" into freedom, autonomy or democracy effectively turn engineering topology into an unproven normative judgement. Engineering redundancy also reveals the difference between "resilience" and "autonomy." Multipath design may make a network harder to destroy, while a small number of institutions still control equipment, spectrum and access licences. Baran's value lies in preserving this distinction within technical history, requiring subsequent political imaginaries of distributed networks to address ownership and governance rather than infer power relations from topology alone.]
- DeNardis, Laura. The Global War for Internet Governance. Yale University Press, 2014. (Read: Chapter 1, pp. 1–24).
[Note: DeNardis places the Internet's technical governance directly within global politics. IP addresses, domain names, standards, routing and security are not purely "technical issues," but neither can they simply be reduced to state sovereignty. Read alongside Galloway, this work shows how protocols, corporations, international organizations and state agencies jointly produce network order. The "connection among technical design, maintenance labour and the legitimacy of governance" revealed by DeNardis's The Global War for Internet Governance extends beyond conceptual debates internal to the text. The political efficacy of technical structures always passes through organizations and institutions. DeNardis's The Global War for Internet Governance allows this connection to be delimited to specific objects and causal chains; analogies beyond that scope require new evidence rather than relying on terminology alone. DeNardis brings Internet governance to a more precise level: domain names, address allocation, routing, security certificates and technical standards have engineering properties while also constituting global power relations. Network order is not a choice between the state and the Internet; it is sustained jointly by governments, standards organizations, operators, platform firms and technical communities. Infrastructure governance therefore often takes place at technical layers that are difficult for publics to perceive, yet profoundly alter conditions of access and expression. The complexity of Internet governance also shows that the technical layer is not a stage prior to politics. How standards organizations reach consensus, how operators exchange traffic, and how certificate authorities establish trust are seemingly backstage arrangements that are themselves institutions. DeNardis thus breaks network sovereignty into identifiable points of governance, exposing both blanket claims that "the state controls the Internet" and that "the Internet transcends the state" as excessively crude.]
- Turner, Fred. From Counterculture to Cyberculture: Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianism. University of Chicago Press, 2006. (Read: Chapter 1, pp. 11–39).
[Note: Turner traces the historical translation among American counterculture, the Whole Earth network and personal computing culture. The book is useful for testing a familiar account of Silicon Valley's origins: the communities, individual autonomy, technological media and network imaginaries of the 1960s did not naturally lead to platform capitalism, but were continually recomposed through different organizational, commercial and cultural forces. At the analytical scale of Turner's From Counterculture to Cyberculture: Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianism, the "historical differences among cybernetics, discipline, societies of control and protocol governance" first appear as a set of concrete relations. More importantly, the account makes feedback, information and adaptation part of organizational thinking, while later discipline, protocols and platform governance reorganize these relations in different ways. Turner's history shows that the relation between countercultural communities and later digital capitalism is not simply one of betrayal. Individual autonomy, the rejection of bureaucracy, networked collaboration and entrepreneurial culture were recombined in particular organizational environments, gradually forming a technological culture that combined communal imaginaries with market logic. Understanding this transformation helps explain why openness, participation and innovation can simultaneously serve anti-establishment politics and corporate management. This history also explains a persistent paradox in Silicon Valley culture: anti-bureaucratic discourse frequently coexists with highly centralized corporations. Individual freedom is emphasized in user interfaces and entrepreneurial myths, while backstage infrastructures, capital control and labour management may become more concentrated. Countercultural legacies are therefore not simply the opposite of platform capitalism, but one of the sources of its cultural legitimacy.]
- Kittler, Friedrich. Optical Media. Translated by Anthony Enns. Polity, 2010. Originally published 2002. (Read: "Theoretical Presuppositions," pp. 29–46).
[Note: Kittler emphasizes how the physical conditions of media and their technical capacities for storage and transmission shape cultural forms. Read before protocol politics, his work helps prevent us from understanding code as purely abstract symbols: protocols must run on hardware, spectrum, storage and computing capacity. His media determinism requires critical reading, but this extreme position is precisely what forces digital media studies to take the material foundation of digital power seriously. Kittler's Optical Media, Turner's From Counterculture to Cyberculture: Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianism and Barbrook's "The Californian Ideology" place the "layered relations among standards, protocols, software, hardware and institutional power" within different intellectual traditions. Kittler's Optical Media frees these layered relations from overly abstract judgements about historical eras. The contribution of Kittler's Optical Media to this problem lies precisely in keeping theory open to refutation. Emphasis on media materiality also changes the direction of analysis in software politics. The "cloud," "real time" and "seamless experience" all depend on concrete devices for storage, transmission, display and computation; these devices determine what information can be recorded, how fast it circulates and where distortions occur. Although Kittler's forceful media theory cannot replace institutional analysis, it prevents critique from remaining at the level of interfaces and discourse.]
- Barbrook, Richard, and Andy Cameron. "The Californian Ideology" Science as Culture 6, no. 1 (1996): 44–72.
[Note: This article critiques the ideological form in which digital technological culture combines liberal individualism, anti-bureaucracy and technological progress. Together with Turner, it forms a historical and ideological pairing: this article examines how cultural imaginaries are reorganized, while Turner provides the historical genealogy. "The Californian Ideology" should not be treated as an all-purpose explanation of Silicon Valley; we should examine the real tensions it reveals. Through Barbrook's "The Californian Ideology," the claim that "distributed topology and political decentralization are not the same thing" can be broken down into several interrelated components. From this perspective, the claim cannot stand apart from institutional authorization and resource allocation; technical forms produce stable social consequences only when combined with organizational power. What most deserves retention is not a fixed label of "Silicon Valley ideology," but the political mixture the article reveals: anti-bureaucracy, individual emancipation and market freedom can intertwine in technological utopias. Open networks and personal computers therefore do not naturally correspond to either left- or right-wing politics; their political direction depends on property structures, labour relations, and how public investment is narrated and distributed. The critique of "The Californian Ideology" also concerns how public investment is erased from technological myths. The Internet, computers and research systems have long depended on state and university resources, yet commercial success is often retold as a victory for individual entrepreneurship and market spontaneity. This politics of memory helps shape Silicon Valley's distinctive attitudes toward regulation, taxation and public responsibility. This ideology has remained effective partly because it can repackage conflicting values as a coherent narrative: individual autonomy and venture capital, anti-bureaucracy and corporate expansion, open culture and intellectual property can coexist under the name of "innovation." The task of critique is not to seek a pure origin, but to analyse how these contradictions are provisionally stitched together by a temporal narrative of technological progress.]
- May, Timothy C.. "The Crypto Anarchist Manifesto" In High Noon on the Electronic Frontier, edited by Peter Ludlow, 41–44. MIT Press, 1996. Originally written in 1988 and circulated in the early cypherpunk milieu; read at the September 1992 founding meeting. (Read: entire manifesto, pp. 41–44).
[Note: This text is first a document of its historical context, rather than a theoretical authority capable of independently grounding normative judgements. May imagines strong encryption, anonymous communication, digital cash and electronic contracts as technical conditions for weakening state capacities for regulation, taxation and information control, while also revealing marked libertarian and technological-determinist tendencies in cypherpunk culture. Its significance here is that it shows the political imaginary in which terms such as "decentralization," "privacy," "anonymity" and "code autonomy," subsequently reused by different camps, were initially connected. It should especially be compared with Galloway and Lessig: if technical architectures do possess regulatory capacities, why can encryption be understood both as a tool against state surveillance and as an infrastructure of new private power? May's "The Crypto Anarchist Manifesto," Barbrook's "The Californian Ideology" and Berners-Lee's "Information Management: A Proposal" together bring the "connection among technical design, maintenance labour and the legitimacy of governance" down from macro-level judgements to mechanisms. May's contribution to this connection lies in forcing broad judgements back onto identifiable institutional mechanisms. His text also shows that cryptography's political significance has never been singular. Privacy technologies can limit the visibility available to states or corporations, but may also weaken the conditions for implementing taxation, labour protections and public regulation. Debate about encryption therefore cannot remain confined to a binary opposition between "state and individual"; platform capital, financial intermediaries, infrastructure operators and public responsibility must all enter the analysis.]
- Berners-Lee, Tim. "Information Management: A Proposal" CERN, 1989. (Read: entire proposal).
[Note: This is not a mature manifesto of Web utopianism, but an engineering proposal to solve CERN's information-management problems. It matters as an original design document because it reveals the initial intentions behind hypertext, links, network sharing and open information. Comparison with later walled gardens, platform economies and cloud services allows us to discuss how design visions shift through commercialization, standardization and capital accumulation. Starting from Berners-Lee's "Information Management: A Proposal," the "historical differences among cybernetics, discipline, societies of control and protocol governance" acquire clearer causal levels. Engineering design carries no fixed political direction. Openness, encryption, distribution and modularity can have opposite consequences in different institutions; assessing their political implications still requires returning to rights of standard-setting and modification, resource ownership and authority over fault resolution. Looking back from this original proposal to the platformized Web also reveals a historical divergence between "linking" and "controlling gateways." Open hypertext emphasizes extensible connections among heterogeneous information, whereas platformization increasingly organizes visibility through account systems, recommendation logics and proprietary interfaces. This difference reminds us that network technologies' political effects often arise from subsequent commercial and institutional structures, rather than being predetermined by their initial engineering designs. Berners-Lee's proposal also retains an institutional problem that later becomes increasingly significant: open protocols can guarantee interconnection, but cannot ensure that the information ecosystem will not be reconcentrated around a few gateways. The Web's basic standards remain open, yet search, social media, identity and cloud services can form new bottlenecks on top of them. The coexistence of an open lower layer and a concentrated upper layer is a structure that cannot be ignored in understanding contemporary Internet politics. It makes the relation between open standards and platform concentration an important thread in Web history: interconnection at the base does not automatically prevent gateways at the upper level from forming monopolies again.]
- Edwards, Paul N.. The Closed World: Computers and the Politics of Discourse in Cold War America. MIT Press, 1996. (Read: Introduction; selected chapters on command-control systems and Cold War computing).
[Note: Edwards places Cold War computing, military command-and-control systems and a broader political imaginary of the "closed world" within a single historical framework. His value lies not in attributing all digital technology to the military, but in showing how computational architectures, simulations, chains of command and political narratives once shaped one another. For this session, the book supplies an indispensable mediation in institutional history between Wiener's cybernetics and later network governance, while reminding us not to treat "control" as merely an abstract form at the software level. Engineering design carries no fixed political direction. Edwards's closed world situates Cold War computers within military command, simulation and imaginaries of closed systems. Computation is not just a tool; it also participates in shaping a political epistemology of enemies, environments, decisions and controllable worlds. The history of digital governance therefore cannot be traced solely through the civilian Internet: military organizations' demands for real-time information, command-and-control and systems integration were also deeply involved in forming computational infrastructures. The closure of the closed world also manifests in modes of knowledge production: complex realities are transformed into systems that can be simulated, computed and commanded, giving decision-makers an imaginary of "seeing the whole from the centre." Today's data dashboards and real-time surveillance continue part of this epistemological impulse, but their organizational structures differ from Cold War military systems. Comparison should focus on how system visualization shapes decision-makers' beliefs in controllability. From this perspective, dashboards, simulations and real-time situational awareness do more than improve decision-making efficiency: they shape standards for "complete information" and "rational control." The more systems emphasize visualizing the whole, the more they may treat local knowledge, ambiguities and political disputes that cannot enter the model as noise. The epistemological boundaries of computational governance are thus directly connected to organizational power. This systems perspective also affects how responsibility is allocated: when failures are attributed to "insufficient information" or "incomplete models," organizations often pursue further data expansion rather than reflect on the setting of objectives itself. The closed world is therefore also a logic of governance that continually demands greater computability.]
- Pickering, Andrew. The Cybernetic Brain: Sketches of Another Future. University of Chicago Press, 2010. (Read: Introduction; selected chapters on British cybernetics).
[Note: Through particular figures and experiments in the British cybernetic tradition, Pickering stresses that cybernetics did not have only one historical path toward centralized management, prediction and optimization. From a performative and adaptive perspective, he presents another conception of machines continually coupled with unknown environments. Following Wiener and Beniger, the book importantly puts the brakes on the linear genealogy that "cybernetics necessarily leads to societies of control": the same concepts of feedback, adaptation and systems may acquire very different political meanings in different institutions and practices. With Pickering's The Cybernetic Brain: Sketches of Another Future, the claim that "distributed topology and political decentralization are not the same thing" can be transformed from a rhetorical proposition into institutional analysis. Engineering design carries no fixed political direction. In Pickering's The Cybernetic Brain: Sketches of Another Future, this claim does not lead to a single conclusion, but requires comparison of configurations of power in different organizational settings, so as to establish concepts' explanatory scope and the conditions of their failure. Through the heterodox tradition of British cybernetics, Pickering shows that cybernetics does not necessarily lead to centralized control. Experiments by Ashby, Pask and others attend more closely to adaptation, emergence and dancing with complex environments, developing a technological imaginary distinct from command-and-control. This difference prevents us from treating "cybernetics" as a single political genealogy, and demonstrates that the same concepts of feedback can lead to different engineering practices and social philosophies. Pickering's "performative" cybernetics also offers another politics of technology: rather than presupposing that an environment can be fully represented, it allows a system continually to adapt to unknown environments through feedback. This stance is not inherently democratic, but provides an important counterexample to the dataist assumption that "more accurate prediction means better governance." Complexity need not always be addressed through more comprehensive models; it can also be handled through open relations of adaptation.]
- Chun, Wendy Hui Kyong. Programmed Visions: Software and Memory. MIT Press, 2011. (Read: Introduction; Chapter 1, "On Software, or the Persistence of Visual Knowledge").
[Note: Chun asks why software is experienced as something both invisible and apparently capable of directly commanding machines, placing its executability, interfaces, memory and discourses of control within media history. This corrects the tendency to treat protocol or code as purely logical rules: code's effects of power depend on compilation, hardware, interfaces, institutions and user practices, and "programmability" is not a frictionless sovereign command. Engineering design carries no fixed political direction. Chun's Programmed Visions: Software and Memory ultimately turns the "connection among technical design, maintenance labour and the legitimacy of governance" into a problem that empirical research can continue to dissect, rather than a diagnosis of the era completed in advance. Chun particularly reveals software's duality: it is presented as transparent, executable command, yet can operate only through numerous invisible layers of abstraction and material processes. Software acquires its distinctive governing efficacy precisely between this "visible sense of control" and "invisible technical mediation." For digital politics, analysing code means asking not only what it says, but how it is compiled, deployed, updated and embedded in organizations. Software's "executability" can also produce an illusion of immediacy, as if writing code were enough to realize an intention without friction. In fact, execution always passes through operating systems, hardware, libraries, networks and user environments; errors, compatibility and updates are normal features of systems. Chun thus returns software from abstract command to a medium with historical and material conditions. Software should therefore be understood as an institution of mediation: it turns complex hardware into operable interfaces while concealing numerous conditions of execution through abstraction layers. Users' experience that "a click makes it happen" comes from a continually maintained technical chain, not from the absence of distance between command and result. It is through this hidden mediation that software acquires powerful effects of everyday governance.]
Session 2: Algorithmic Governmentality and Biopolitics
This session moves from "how rules prescribe action" to "how computation organizes the possibilities of action." Rouvroy's algorithmic governmentality does not presuppose a reconstruction of subjects' worlds of meaning. Instead, through data collection, correlation analysis and probabilistic prediction, it produces structures of classification, risk and opportunity before action occurs. Cheney-Lippold's algorithmic identity further shows how subjects can be divided into continually updated data categories. Bucher rejects the image of "the algorithm" as a single, mysterious black box, stressing that code, training data, platform objectives, engineering practices, user behaviour and institutional environments together constitute algorithmic power. The computational subject is therefore not a person completely "known" by a machine, but an actor recoded, grouped and ranked in different institutional settings.
The political efficacy of algorithms cannot be explained solely by model performance. A probabilistic output becomes part of governance only when it is given consequences for action within organizational procedures such as loan approval, welfare eligibility, border screening, recruitment, content distribution or insurance pricing. Prediction, explanation, judgement and justification must therefore be distinguished: statistical correlation is not causal explanation, and a model's risk score cannot automatically serve as a legal or ethical reason. Amoore's analysis of uncertainty and probabilistic judgement, Eubanks's empirical research on welfare automation and Pasquale's discussion of black boxes and accountability all point to the same question: how are technical outputs translated within institutions into decisions that impose real constraints?
This session also returns "algorithmic bias" from a general value judgement to a chain of institutional processes. Training data, labelling systems, target variables, evaluation metrics, threshold settings, deployment environments, human review and channels of appeal can each generate or amplify inequality. Classification does not merely describe the world: once adopted by organizations, it helps shape structures of opportunity and social identities. Algorithmic governmentality therefore concerns not only knowledge but also administrative procedures, organizational responsibility and the protection of rights. The expansion of computability has not abolished political judgement; in many settings, it has changed how judgement is concealed, distributed and held accountable.
Required Readings:
- Rouvroy, Antoinette. "The End(s) of Critique: Data Behaviourism versus Due Process" In Privacy, Due Process and the Computational Turn, edited by Mireille Hildebrandt and Katja de Vries, 143–167. Routledge, 2013.
[Note: This chapter is the conceptual core of the session. Rouvroy's algorithmic governmentality describes a form of governance that need not first understand subjects' meanings, motivations or intentions, but intervenes before action through data collection, correlation analysis and prediction. The key is to distinguish correlation from causation, prediction from justification, and profiling from judgement: algorithms do not need to "truly know who you are"; they only need institutions to believe, probabilistically, that certain classifications suffice to support action. Only then can we understand the real tension between algorithmic governance and traditional legal procedures. More specifically, the conflict with due process is not that "machines lack humanity," but that data behaviourism tends to transform subjects from legal and political persons capable of giving reasons and answering accusations into probabilistic behavioural profiles. Correlation analysis can support pre-emptive intervention without establishing causal explanations, making critique itself difficult: if a system no longer asks who you are or why you are as you are, but merely adjusts the environment you will encounter on the basis of past data, is a traditional critique centred on exposing ideology or correcting false representations still sufficient? This chapter should therefore be read alongside Yeung, Eubanks and Amoore to test, through regulation, welfare administration and security judgements respectively, whether algorithmic governmentality holds in the same way across institutions. Rouvroy's data behaviourism forms the core of algorithmic governmentality because it attempts to bypass the intentions, narratives and normative judgements of the traditional subject, discovering behavioural patterns directly in correlations across vast datasets. Governance can thus probabilistically shape subjects' spaces of opportunity before they have expressed themselves. A fundamental jurisprudential problem then emerges: correlation may support prediction, but cannot automatically supply reasons; procedural justice requires not merely accurate outcomes but contestable normative grounds.]
- Cheney-Lippold, John. "A New Algorithmic Identity: Soft Biopolitics and the Modulation of Control" Theory, Culture & Society 28, no. 6 (2011): 164–181.
[Note: Cheney-Lippold introduces algorithmic identity to explain how digital systems produce dynamic categories from behavioural data. Read alongside Deleuze, this shows that identity is no longer determined solely by subjects' self-declarations or fixed demographic attributes, but can become a probabilistic object continually recalculated from behaviour. The limits are equally clear: algorithmic classification still draws on existing social categories, rather than being purely mathematical classification detached from history. The continuity between algorithmic identity and conventional population classifications must be specified through the origins, institutional uses and updating mechanisms of categories, rather than remain a correspondence between terms. The computational subject is not an individual fully known by a model, but an object divided into scores, probabilities, categories and behavioural traces in different institutional settings. Once classification enters welfare, credit, recruitment or content distribution, it in turn shapes opportunity structures and subjects' actions. Algorithmic identity turns identity from a stable attribute into a continually updated probabilistic classification. Systems need not establish who "the real you" is; they only need to place individuals provisionally into operable categories on the basis of behavioural traces, and use these to decide advertising, prices, risk or visibility. Cheney-Lippold thereby reveals a new dimension of identity politics: subjects may not know which category they have been assigned to, and have even greater difficulty participating directly in its definition and revision. Dynamic classification has another important consequence: identity is no longer confirmed only at registration, but can be inferred anew after every click, movement and purchase. Subjects confront not one fixed dossier but a series of continually changing and potentially contradictory probabilistic identities. Appeals are therefore especially difficult, since those affected may not even know which inference triggered differential treatment, or when. The political difficulty of algorithmic identity thus involves not only "being misjudged," but also subjects' difficulty in knowing which categories describe them and how those categories travel across settings.]
- Bucher, Taina. If...Then: Algorithmic Power and Politics. Oxford University Press, 2018. (Read: Chapter 2, "The Multiplicity of Algorithms," pp. 19–40; Chapter 3, "Neither Black nor Box," pp. 41–65).
[Note: Bucher's important contribution is to reject the image of algorithmic power as a mysterious, unified, invisible "black box." Algorithmic power is distributed across code, data, platform objectives, user behaviour, engineering practices and institutional arrangements. This helps ground Rouvroy's broad concept of governance in concrete systems: we do not face "the algorithm" as a single actor, but a set of different software, institutions and practices. "Algorithmic transparency" must therefore also be redefined: transparent code does not automatically mean transparent governance. Predictive systems often need only construct sufficiently operable proxies, without developing a complete theory of identity; this is precisely where algorithmic identity and traditional identity politics are related yet cannot be equated. Bucher's If...Then: Algorithmic Power and Politics thus offers a limited but effective explanatory scale for the "institutional chain linking quantification, data production and algorithmic systems": it can identify mechanisms, but does not replace further analysis of organizations, law and material conditions. Bucher rejects the algorithm as a single black-box subject, emphasizing that algorithms are jointly constituted by code, data, commercial goals, user practices and organizational institutions. Algorithmic power is therefore relational power: it resides not only in ranking logics, but also in engineering teams' choices, platform metrics and users' adaptations to systems. Transparency understood merely as releasing source code cannot encompass this complex sociotechnical assemblage. Bucher's attention to algorithmic imaginaries also shows that algorithmic power resides not only in what systems actually do, but in what people believe they can do. Users adjust their expression to accommodate imagined recommendation rules, while institutions may expand automated deployments in the belief that "algorithms are more objective." Algorithms' social efficacy thus arises partly from the expectations and uncertainties that form around them.]
- Amoore, Louise. Cloud Ethics: Algorithms and the Attributes of Ourselves and Others. Duke University Press, 2020. (Read: Chapter 1, pp. 1–24).
[Note: Amoore examines how probabilistic judgements acquire the capacity to produce action in border, security and risk governance. The important question is not whether algorithms can "predict accurately," but when institutions turn probabilistic outputs into actual decisions, and how that translation changes the politics of responsibility and uncertainty. Transparency is not a single solution either. This session places explainability within a broader understanding of due process and institutional accountability. Placed in the session's overall context, Amoore's Cloud Ethics: Algorithms and the Attributes of Ourselves and Others gives more concrete structure to the question of "how transparency, explanation, appeal and responsibility are redistributed in automated decision-making." The computational subject is not an individual fully known by a model, but an object divided into scores, probabilities, categories and behavioural traces in different institutional settings. Once classification enters welfare, credit, recruitment or content distribution, it in turn shapes opportunity structures and subjects' actions. Amoore shifts algorithmic ethics from "whether a model is correct" toward how probabilistic judgements are put into action. Border, security and risk systems often operate under uncertainty; the key is not to eliminate uncertainty, but to examine how institutions choose thresholds, give certain attributes greater weight and allow partial correlations to have real consequences. Responsibility therefore cannot be transferred to the model: the institutional allocation of uncertainty is itself a political decision. Amoore thereby advances the ethical question into attribution itself: systems infer the potential properties of a person or event from partial traces and pre-allocate attention, suspicion and resources accordingly. Probability is not a neutral waiting state, but can become a reason for action. The politics of risk governance lies in who is made to bear the costs of uncertainty, and which errors institutions deem acceptable.]
Recommended Readings:
- Gillespie, Tarleton. "The Relevance of Algorithms" In Media Technologies, 167–194. MIT Press, 2014.
[Note: Gillespie discusses how "relevance" itself becomes one of platforms' most important public decisions. Search, recommendation and ranking do not merely compute information; they decide what deserves to be seen and considered important. This adds a dimension of media and public culture to algorithmic governmentality. Theoretical similarity does not automatically imply institutional identity. Predictive accuracy, statistical correlation and administrative legitimacy belong to different levels; a risk score truly becomes governance only when it acquires consequences through thresholds, human review, appeals and enforcement procedures. "Relevance" is therefore normative rather than purely technical. What platforms call relevance commonly compresses click probabilities, dwell time, commercial objectives and content value into one ranking process. Algorithms do not directly adjudicate truth, yet can alter the conditions under which public issues emerge by allocating attention. Gillespie thus moves algorithm studies from inside models to platform institutions: what requires explanation is how criteria of relevance are set, adjusted and legitimized. Recommendation systems also have a public dimension because they alter the visible structure of a "common world." When users receive highly personalized rankings, public issues no longer have identical gateways, while platforms retain the authority to adjust recommendation goals and content boundaries. Research on algorithmic relevance must therefore connect media institutions to business models, rather than merely measure ranking accuracy. Platform ranking is both information processing and the allocation of attention; when relevance becomes a function of commercial objectives, media power enters the core of algorithmic design. Gillespie further reminds us that the normativity of algorithmic ranking often appears in the name of "user demand" or "relevance," thereby turning value judgements into technical matters. What deserves recommendation, what counts as spam and which content should be downranked all presuppose some public order. If algorithm studies ignore the institutional objectives behind these classifications, they will misrepresent platforms' most important political judgements as purely engineering optimization.]
- Eubanks, Virginia. Automating Inequality. St. Martin's Press, 2018. (Read: Introduction, pp. 1–14; Chapter 1, pp. 17–36).
[Note: This book brings abstract algorithmic governance into welfare, poverty and administrative systems. Its core is not to prove that algorithms "are biased," but to observe how technical systems reorganize the allocation of social resources and how historical inequalities enter risk classifications. It shifts analysis from "platform experience" toward state administration and social policy. Eubanks's Automating Inequality does not provide a final name for "how organizations translate probabilistic outputs into judgements with real consequences," but a theoretical point of entry open to testing against competing explanations; this openness is itself part of its research value. Eubanks's Automating Inequality also shows that automated governance first falls on populations with the least capacity to exit. When technical systems repackage resource scarcity, welfare screening and risk management as questions of efficiency, poverty itself can readily become suspicion and computable risk. Eubanks's cases show that automated welfare systems do not simply substitute machine for human judgement; they reorganize screening within existing frameworks of resource austerity, administrative reform and the governance of poverty. Technology can speed up assessment, but can also fix historically entrenched suspicion and punishment of the poor into data rules. Algorithmic fairness therefore cannot be separated from the institutional objectives of the welfare state: a system that accurately enforces harsh policies can still produce profound injustice. Eubanks's empirical material also reveals a common political asymmetry in technological reform: administrative agencies deploy systems to reduce costs and prevent fraud, while those undergoing assessment bear the temporal costs of data errors, complex processes and automatic refusals. Efficiency calculated only from the institution's side excludes the extra labour borne by vulnerable groups from performance measures. The criteria for evaluating automated governance must themselves therefore be politicized. These cases also show that automation in poverty governance produces more than isolated errors; it can change the relationship between the state and welfare applicants. When face-to-face explanation and case-specific discretion are replaced by database verification and risk scoring, applicants must shoulder more responsibility for proving their eligibility. Digitizing procedures can thus transfer administrative burdens from inside institutions to those with the fewest resources.]
- Pasquale, Frank. The Black Box Society. Harvard University Press, 2015. (Read: Introduction; Chapter 2, "The Digital Poorhouse," pp. 35–58).
[Note: Pasquale focuses on information asymmetry and accountability in automated decision-making. His work can test whether "open-source code" is a sufficient answer: even a system with open code may remain unaccountable because of its data, model complexity, commercial secrecy or organizational permissions. Through Pasquale's The Black Box Society, the "institutional chain linking quantification, data production and algorithmic systems" enters the organizational and institutional levels. At the scale of Pasquale's The Black Box Society, this chain must be approached through both continuity and transformation: old institutions may gain new ways of operating through new technologies, and new technologies may be reabsorbed into existing institutions. Pasquale's The Black Box Society also suggests that analysing this institutional chain requires preserving differences among objects and institutional boundaries, rather than substituting concepts' rhetorical force for empirical explanation. Pasquale's The Black Box Society extends invisibility from the technical to the institutional black box: commercial secrecy, control of data, professional barriers and organizational permissions together determine who can explain a system. Explainability can therefore become real accountability only when joined to auditing, appeals and the allocation of responsibility. Pasquale's black box encompasses technical complexity, trade secrets and institutional information asymmetry simultaneously. The opacity of financial scoring, search ranking and platform decisions cannot all be solved by "explaining the model"; those affected often also lack access to data, channels of appeal and bargaining power. Accountability must therefore extend from disclosure to organizational responsibility, specifying who has the authority to decide, who must give reasons and who can correct mistakes.]
- Harcourt, Bernard E.. Exposed: Desire and Disobedience in the Digital Age. Harvard University Press, 2015. (Read: Chapter 1, "The Expository Society," pp. 1–28).
[Note: Harcourt discusses exposure, desire and surveillance together. Subjects' self-display should not simply be explained as "voluntarily submitting to surveillance"; we should analyse how platform design, social norms, the desire for visibility and commercial data extraction reinforce one another. Transparency is not a single solution either. The question of "how transparency, explanation, appeal and responsibility are redistributed in automated decision-making," revealed by Harcourt's Exposed: Desire and Disobedience in the Digital Age, extends beyond conceptual debate within the text. Harcourt's Exposed: Desire and Disobedience in the Digital Age allows this question to be delimited to specific objects and causal chains; analogies beyond that scope need new evidence and cannot rest on terminology alone. Harcourt's "exposure" also reveals an important turn in surveillance studies: subjects do not always exist as passive objects, but actively produce visibility through social interaction, consumption and self-display. The question is not whether this activity cancels coercion, but how commercial architectures capture desire and turn it into resources for prediction and governance under asymmetric data-access rights. Harcourt makes "exposure" more than a loss of privacy: it becomes a social relation. Through sharing, searching, location-tracking and consumption, subjects continually produce themselves, while platforms retranslate these expressions into computational resources. Desire and surveillance are therefore not opposing forces: precisely because digital environments satisfy needs to be seen, connected and expressive, data extraction can penetrate everyday life in the forms of participation and convenience. The paradox of the expository society is that subjects gain freedom of expression and connection while handing ever more traces of their lives to platforms they cannot observe reciprocally. Freedom and surveillance do not simply grow at one another's expense; they may increase together within the same architecture of interaction.]
- Beer, David. The Data Gaze: Capitalism, Power and Perception. Sage, 2018. (Read: Introduction and Chapters 1–2).
[Note: Beer uses the concept of the data gaze to discuss how datafication changes relations of "seeing" and "being seen." This connects algorithmic classification with social perception: as life becomes increasingly datafied, institutions are not alone in "seeing" us; people increasingly understand themselves through metrics as well. At the analytical scale of Beer's The Data Gaze: Capitalism, Power and Perception, "how classification, prediction, scoring and ranking constitute computational subjects" first appears as a set of concrete relations. So-called algorithmic bias cannot remain a value judgement. Training data, labelling systems, target variables, evaluation metrics, deployment environments and existing institutional inequalities may each produce different problems; only tracing them layer by layer reveals where bias arises. The data gaze is not merely an institutional gaze upon people, but a perceptual mechanism through which subjects reconceive themselves in terms of metrics. Step counts, credit scores, popularity, rankings and performance data translate life into comparable quantities that in turn shape goals for action. Beer thus connects algorithmic governance to quantified subjectivation: data describe reality while also helping determine what kind of reality deserves pursuit. The metricized self does not arise entirely from coercion. People actively organize their lives with rankings, health data and social feedback, but this voluntary use occurs within the measurable dimensions that platforms have predefined. Subjectivity therefore includes both the desire to quantify oneself and guidance by metric architectures. Beer's analysis moves power beyond an external gaze into the technical forms of everyday self-evaluation. Such self-quantification further shows that data power is often realized through subjects' active participation, rather than solely through external surveillance or compulsory collection. The power of metrics thus comes both from measurement and from people reorganizing their actions around metrics. It is precisely in this self-adjustment that the data gaze enters subject formation.]
- Yeung, Karen. "Algorithmic Regulation: A Critical Interrogation" Regulation & Governance 12, no. 4 (2018): 505–523.
[Note: This article offers a corrective from the perspectives of institutions and the rule of law. Algorithmic regulation does not simply improve technical efficiency; it concerns how machines execute, monitor and dynamically adjust rules. It helps build a clearer theoretical bridge between Rouvroy and Lessig. Yeung's "Algorithmic Regulation: A Critical Interrogation," Beer's The Data Gaze: Capitalism, Power and Perception and Seaver's "Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems" place "how organizations translate probabilistic outputs into judgements with real consequences" within different intellectual traditions. Yeung's "Algorithmic Regulation: A Critical Interrogation" frees this problem from overly abstract diagnoses of the era. So-called algorithmic bias cannot remain a value judgement. Yeung's contribution to this question lies precisely in keeping theory open to refutation. Her institutional perspective further shows that the key to algorithmic regulation is not merely its degree of automation, but how feedback loops change rule enforcement. Continuous monitoring can shift governance from retrospective punishment toward real-time adjustment, but dynamism may also compress procedural review and stable expectations. There is consequently a real institutional tension between efficiency and adaptability, on one hand, and the foreseeability and explainability demanded by the rule of law, on the other.]
- Seaver, Nick. "Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems" Big Data & Society 4, no. 2 (2017). (Read: entire article).
[Note: Seaver opposes treating algorithms merely as technical objects that can be read directly from code, arguing instead that algorithmic systems are cultural objects jointly constituted by engineers' classifications, organizational cultures, models of users, data practices and maintenance work. This is crucial for this session: if algorithmic governmentality involves prediction and classification, researchers must ask not only how models work mathematically, but how categories such as "relevance," "user," "anomaly" and "success" are defined within organizations. Through Seaver's "Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems," the "institutional chain linking quantification, data production and algorithmic systems" can be broken down into several interrelated components. So-called algorithmic bias cannot remain a value judgement. From Seaver's perspective, this institutional chain cannot stand apart from institutional authorization and resource allocation; technical forms produce stable social consequences only when combined with organizational power. Seaver's cultural-anthropological approach returns algorithms from "independently operating code" to organizational life. How engineers understand users, which metrics count as success and how teams handle anomalies are all cultural practices that help constitute algorithmic systems. The significance of algorithmic ethnography thus lies not in entering secret code repositories, but in observing how technical objects are continually produced through everyday work, narratives and institutional routines.]
- Fourcade, Marion, and Kieran Healy. "Classification Situations: Life-Chances in the Neoliberal Era" Accounting, Organizations and Society 38, no. 8 (2013): 559–572. (Read: entire article).
[Note: Fourcade and Healy return scoring, ranking and classification to research on social stratification, emphasizing that modern classification systems do not neutrally describe, but change the opportunities, prices, resources and institutional treatment available to individuals. Their concept of the "classification situation" is especially useful alongside algorithmic identity: computational subjects are not only representations in data, but occupy real opportunity structures shaped by classification outcomes. Fourcade's "Classification Situations: Life-Chances in the Neoliberal Era," Seaver's "Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems" and Espeland's "A Sociology of Quantification" together move the question of "how transparency, explanation, appeal and responsibility are redistributed in automated decision-making" from broad judgements to mechanisms. Transparency is not a single solution either. Fourcade's contribution to this question lies in forcing macro-level judgements back onto identifiable institutional mechanisms. So-called algorithmic bias cannot remain a value judgement. "Classification Situations: Life-Chances in the Neoliberal Era" places this redistribution of transparency, explanation, appeal and responsibility back within historical conditions. Similar digital appearances need not arise from the same mechanisms, and the validity of concepts therefore depends on preserving differences. Fourcade and Healy's classification situations directly connect classification to life chances. Classifications in credit, insurance, education and consumption do not simply describe social differences: they determine interest rates, opportunities, treatment and visibility, thereby producing feedback effects. What digitization intensifies is not classification itself, but its capacity to enter everyday institutions at finer granularity, higher frequency and lower visibility.]
- Espeland, Wendy Nelson, and Mitchell L. Stevens. "A Sociology of Quantification" European Journal of Sociology 49, no. 3 (2008): 401–436. (Read: entire article).
[Note: Espeland and Stevens provide a methodological baseline for studying quantification: numbers, metrics and rankings compress complexity, but also use commensuration to place objects that were not directly comparable on a common scale. Bringing this article into algorithmic governance is not to claim that all algorithms are merely "scoring," but to see the social engineering that often occurs before predictive models can enter institutional decisions: objects must become quantifiable, metrics comparable and results rankable. Starting from Espeland's "A Sociology of Quantification," the question of "how classification, prediction, scoring and ranking constitute computational subjects" acquires clearer causal levels. Transparency is not the single solution to automated governance. Once commensuration is accomplished, political disputes are often rewritten as problems of metric design. University rankings, credit scores and risk models wield power not because numbers are inherently coercive, but because organizations begin allocating opportunities and resources according to them. The sociology of quantification therefore offers an important path: trace how classification standards change behaviour after entering institutions, while in turn stabilizing the classifications themselves. Espeland and Stevens also emphasize reactivity: those who are ranked and measured adjust their behaviour to fit evaluation systems. Once a metric becomes a target, it may change the object it was intended to measure. This mechanism is especially important in algorithmic governance, because deploying predictive models changes people's strategies and consequently changes future data. Quantification is thus not passive recording, but a social process capable of reshaping institutional settings. When metrics become goals of action, systems of quantification change the quantified objects themselves; this is the fundamental reason rankings and scores can generate institutional feedback. Commensuration and reactivity together constitute the basic mechanism through which quantification regimes continually reinforce themselves.]
- O'Neil, Cathy. Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. Crown, 2016. (Read: Introduction; selected case studies on education, employment, credit, and policing).
[Note: Through a series of institutional cases, O'Neil shows how large-scale scoring models amplify inequality when opacity, closed feedback loops and high-impact settings coincide. This is not the endpoint of algorithmic theory, but an excellent stress test for concepts: only by placing models back within admissions, employment, credit or policing procedures can we see how errors are distributed, whether channels of appeal exist and how quantitative judgements acquire de facto coercive force. Transparency is not the single solution to automated governance. O'Neil's WMD concept stresses how models' scale, opacity and feedback properties jointly amplify harm. This accessible concept needs to be paired with more rigorous algorithm research, but it captures the core of institutional deployment: once a model enters high-stakes fields such as education, employment or justice, errors are no longer merely statistical deviations. They change people's opportunities and write their effects back into future data. Model evaluation must therefore include social feedback, not just offline accuracy. This feedback property connects WMDs with performativity in the social sciences. Model-generated judgements change educational resources, credit opportunities or policing attention, and those outcomes may then be treated as "real-world data" for the next round of training. Unless this cycle is interrupted, historical inequalities acquire renewed empirical legitimacy in the form of model performance. Evaluating models therefore requires not only asking whether predictions are accurate, but also whether deployed predictions generate new data that continually validate existing differences. The WMD framework also raises an accountability problem: when models are treated as objective tools, institutions can explain unpopular decisions as "results of the data," obscuring policy choices. Which variables are selected, which errors are acceptable and whether human overrides are permitted remain institutional decisions. Automation has not eliminated judgement; it has written some judgements into models and procedures in advance. Institutions therefore cannot treat responsibility for models as a one-off technical duty fulfilled by developers. Deployment, updating, threshold adjustments and human overrides continually change systems' consequences; accountability should cover their whole life cycle.]
- Kitchin, Rob. The Data Revolution: Big Data, Open Data, Data Infrastructures and Their Consequences. Sage, 2014. (Read: Introduction; chapters on data infrastructures and data assemblages).
[Note: Kitchin restores big data from the slogan of "more data" to a data assemblage: collection devices, standards, databases, analytical tools, organizational objectives and institutional practices together determine how data are produced. This supplies the infrastructural component that can easily become abstract in Rouvroy's theory: data behaviourism does not confront naturally existing data, but objects that have been selected, cleaned, formatted and institutionalized. Through Kitchin's The Data Revolution: Big Data, Open Data, Data Infrastructures and Their Consequences, the "institutional chain linking quantification, data production and algorithmic systems" can be transformed from a rhetorical proposition into institutional analysis. Transparency is not the single solution to automated governance. Kitchin's analysis of data infrastructure prevents "big data" from being imagined as self-existing, like a natural resource. Data become usable objects only through collection protocols, database structures, cleaning standards, metadata and institutional objectives. "Data-driven" decisions therefore involve selection and exclusion from the outset; understanding model outputs requires tracing backward how data are produced, classified and maintained. Kitchin also distinguishes data openness from data power. Data can be technically open yet genuinely usable by only a few actors because of formats, computing resources, expertise and institutional permissions. The politics of open data thus cannot end with publishing files: it must address who has the capacity to analyse, link and turn data into resources for decision-making. Data infrastructure is itself a mechanism for distributing unequal capacities. Its governance also includes compatibility between standards and long-term maintenance. Whether different institutions can understand and reuse data often depends on metadata, formats and continual updating; this low-visibility work determines whether so-called "data assets" can exist at all.]
- Rieder, Bernhard. Engines of Order: A Mechanology of Algorithmic Techniques. Amsterdam University Press, 2020. (Read: Introduction; selected chapters on ranking, recommendation, and algorithmic techniques).
[Note: Rieder's mechanological approach shifts attention from the excessively broad noun "algorithm" to specific technical mechanisms—ranking, recommendation, classification—and their histories. Read after Bucher, it trains a finer-grained mode of critique: different algorithmic techniques have different inputs, objective functions, evaluation methods and organizational uses, and cannot all be described as a single omnipresent "algorithmic power." Transparency is not a single solution either. Transparency is not the single solution to automated governance. Rieder's fine-grained method is particularly useful for resisting the overgeneralization of "algorithm." Ranking, classification and recommendation have distinct problem formulations, technical traditions and evaluation metrics, and their power effects within organizations differ as well. Only by matching model types with specific institutional uses can algorithmic critique move from grand diagnosis to verifiable analysis of mechanisms. Mechanology also requires returning algorithmic techniques to genealogies of tools, rather than following only the latest models. Ranking, matching, clustering and recommendation all have longer histories and have produced different bodies of engineering common sense. Understanding technical genealogies helps distinguish governance effects arising from machine learning's new capabilities from those that actually derive from earlier information retrieval, statistical classification or database management. Such technical disaggregation also supplies more precise objects for governance: different mechanisms of ranking, matching and classification should face different requirements for transparency, auditing and responsibility. Rieder's emphasis on algorithmic techniques can also change auditing methods. Instead of requiring an abstract "algorithm" to be transparent as a whole, we can separately examine input data, ranking functions, feedback signals and organizational objectives. This disaggregation links responsibility to specific stages and frees critique from a binary choice between "black boxes are unknowable" and "releasing the code solves everything."]
Part II: Extraction, Coloniality and Infrastructure (Weeks 3–4)
Session 3: Surveillance, Extraction and Data Coloniality
This session further situates algorithmic governance within data production, resource extraction, and historical inequality. Zuboff's surveillance capitalism emphasizes how behavioral data become predictive assets; Couldry and Mejias's data colonialism places data relations within a longer history of colonial appropriation. Starting from the historical continuity of slavery, racialized identification, and surveillance technologies, Browne reveals how systems of classification embed differentiated visibility and exposure. These accounts are not equivalent: capital accumulation, colonial appropriation, racialized governance, and biopolitics have distinct historical mechanisms. Juxtaposing these concepts means testing what each can explain about the data economy and when an overextended analogy obscures differences.
Data extraction is not simply an exchange occurring after "user consent," either. Through identity systems, defaults, tracking infrastructure, advertising markets, data brokers, and network effects, platforms continually convert everyday actions into records that can be stored, linked, and used for prediction. Groups bear unequal risks when they are recorded, identified, or excluded; classifications in biometrics, credit scoring, targeted advertising, and public administration bring existing social differences into technical processes. Noble, Benjamin, Magnet, Dencik, and others approach these problems through information retrieval, racial technologies, identification, and data justice. "Data bias" consequently becomes more than model error: it is part of social classification, resource allocation, and relations of power.
Mbembe is not used here to produce facile analogies such as "account suspension equals digital death." He supplies a political context for understanding coloniality, enmity, borders, and differentiated exposure. An analysis of data coloniality holds only where compulsory dependence, relations of appropriation, epistemic authority, legal asymmetry, and exit costs can be specified. This session therefore does not seek the most dramatic historical name for digital society. It clarifies how datafication connects observation, extraction, classification, and governance, and how these mechanisms inherit or reorganize earlier structures of power in different institutional settings.
Required Readings:
- Zuboff, Shoshana. The Age of Surveillance Capitalism. PublicAffairs, 2019. (Read: Chapter 3, "The Discovery of Behavioral Surplus," pp. 63–96).
[Note: Zuboff's concept of "behavioral surplus" connects data governance with capital accumulation. Focus on how she distinguishes user behavior, data extraction, prediction products, and commercialization. We must therefore test whether this model applies to the entire digital economy or chiefly explains the extractive mechanisms of Google and advertising platforms. Chapter 3 requires unpacking the whole argumentative chain of behavioral surplus: users leave behavioral traces in obtaining services; firms turn data exceeding the immediate needs of service improvement into additional inputs; computation then produces products predicting future behavior. What needs testing is whether this chain generalizes beyond advertising platforms and whether surveillance capitalism flattens differences among states, finance, logistics, and nonadvertising platforms. Read Zuboff as a strong model rather than a general theory of the digital economy: in which industries is it most explanatory, and where must platform capitalism, data colonialism, or administrative automation supplement it? A valid conceptual transfer must specify mechanisms such as coercion, ownership, spatial control, hierarchies of knowledge, or resource flows, while acknowledging important differences between classical colonial rule and platform data relations. Zuboff's behavioral surplus brings user activity, data extraction, and prediction markets into one process of capital accumulation. Its theoretical ambition extends far beyond online advertising, generating debate over whether surveillance capitalism has been overgeneralized. The most productive reading distinguishes businesses genuinely dependent on behavioral prediction from platforms whose revenues come chiefly from infrastructure rents, transaction commissions, or monopoly pricing, rather than fitting the entire digital economy into the concept.]
- Couldry, Nick, and Ulises A. Mejias. The Costs of Connection. Stanford University Press, 2019. (Read: Chapter 2, "Cloud Empire," pp. 37–68; Chapter 3, "The Coloniality of Data Relations"; Chapter 5, pp. 167–208).
[Note: Couldry and Mejias use data colonialism to describe the ongoing conversion of social life into extractable data. "Cloud Empire" particularly suits this session because it connects clouds, platforms, data extraction, and social relations. The central questions are whether data colonialism can provide a historical framework for cloud infrastructure and where this framework may overextend the analogy with classical colonialism. The authors place data relations within a longer history of colonial resource appropriation, stressing the continuous translation of social life itself into commercially usable data flows. "Cloud empire" names the infrastructure and corporate power supporting this relation. We cannot merely seize on the forceful word "colonialism": we must closely read how the analogy is built through the appropriation of social life, the normalization of data extraction, and a new social order. Alongside Browne, ask whether a generalized colonial model centered solely on extraction overlooks historical differences, given coloniality's enduring entanglement with racialization, territorial appropriation, and violent rule. Surveillance has no single central agent. National security, advertising, employer management, financial risk control, and social platforms may share data and technologies, yet have different legal mandates and interests; one concept of surveillance capitalism must not erase these institutional differences. Describing data relations as a new form of colonial appropriation emphasizes how everyday life itself continually becomes an extractable resource. The concept's strength is connecting capital accumulation and relations of subjectivity; its risk is flattening historical colonialism's differences of violence, territory, and race. A valid comparison therefore specifies mechanisms of appropriation, resource flows, and structures of dependence rather than borrowing only the moral force of "colonial."]
- Browne, Simone. Dark Matters: On the Surveillance of Blackness. Duke University Press, 2015. (Read: Introduction; Chapters 1–2).
[Note: Browne extends the historical depth of surveillance studies into slavery, racialized identification, and the continuous management of Black bodies, resisting accounts of digital surveillance as an entirely new problem that suddenly arrived with the internet. Her claim is not that all surveillance is alike, but that modern identification, border management, and regimes of visibility inherit and reorganize earlier technologies of racialized rule. Between Zuboff and data colonialism, this raises the question of whether explanations centered on commercial extraction adequately account for historical differences in surveillance power. Being seen and being datafied are also distinct. Many governing systems derive their power precisely from invisible backend classification and inference; critique must therefore investigate classificatory opportunities, risk distribution, and epistemic authority as well as privacy exposure. In Browne's Dark Matters: On the Surveillance of Blackness, the analytical gains and nonnegotiable historical limits of colonial analogy are consequently more than conceptual questions. The politics of extraction arise from asymmetry: some actors control collection, linkage, inference, and reuse, while others appear chiefly as those recorded, classified, and predicted. Privacy is only one layer; property rights, epistemic authority, and conditions of exit are equally important. Browne also forces surveillance history to revisit the historical continuity of race and the body. Biometrics, border checks, and database classifications seem to rely on new technologies, but who must be identified, who counts as a risk, and who must prove their identity have never been technically neutral arrangements. The history of digital surveillance must therefore trace the longer institutional genealogy of visibility and racialized rule rather than begin with internet firms. Browne's dark sousveillance also preserves resistance: monitored people do not merely accept surveillance but develop practices of counteridentification, counterrecording, and recoding visibility. Surveillance history is thus not only the history of expanding power but also that of classified subjects continually disrupting classification—a crucial point for research on digital identity and biometrics.]
- Mbembe, Achille. Necropolitics. Translated by Steven Corcoran. Duke University Press, 2019. Originally published in French as Politiques de l'inimitié in 2016. (Read: Introduction; "Necropolitics"; selected chapters on enmity, borders, and exposure).
[Note: Mbembe's necropolitics concerns sovereignty, war, coloniality, and the political conditions under which lives are differentially exposed to death and injury. Applied to digital and intelligent technologies, it must never simply rename account bans, downranking, or platform exclusion "digital death": such analogies weaken the concept's historical force and obscure actual relations of violence. A more rigorous transfer asks how datafied borders, automated risk judgments, militarized surveillance, and resource allocation help produce different degrees of exposure, vulnerability, and freedom of action. Further questions concern who determines data standards, servers, models, and research agendas, and whether researched communities can exit, refuse, revise, and govern jointly. In the context of this session, Mbembe's Necropolitics gives a more concrete structure to the question of bringing data justice, epistemic authority, and differentiated exposure into one research framework. Extraction is political because relations are asymmetric: some actors control collection, linkage, inference, and reuse, while others appear chiefly as recorded, classified, and predicted subjects. Privacy is one layer; property rights, epistemic authority, and exit conditions matter equally. Necropolitics examines how sovereignty organizes life through exposure to death, warlike spaces, and differentiated conditions of survival; its historical background far exceeds digital governance. Within data coloniality, the theory should not inflate account bans or downranking into "digital death," but investigate when digital infrastructures enter borders, policing, war, or resource allocation and connect with actual biopolitics and necropolitics. Mbembe is therefore most useful here for calibrating scale. Datafied governance enters a necropolitical problem space only when linked to closed borders, military strikes, policing risks, or unequal distributions of basic resources. Retaining the concept in this intensive biopolitical context prevents digital critique from borrowing the rhetoric of death at the cost of historical weight.]
Recommended Readings:
- Benjamin, Ruha. Race After Technology: Abolitionist Tools for the New Jim Code. Polity, 2019. (Read: Introduction and Chapter 1).
[Note: This connects algorithms with racialized power, supplementing Noble's approach through information retrieval. It reminds us that general intellect is not an undifferentiated "common human resource": whose knowledge, languages, and embodied experience most readily become training data is itself political. Theoretical resemblance does not automatically establish institutional identity. Benjamin's Race After Technology: Abolitionist Tools for the New Jim Code therefore explains the historical relations among extraction, surveillance, racialized classification, and coloniality only insofar as corresponding institutional mechanisms can be identified empirically. The "New Jim Code" makes technological neutrality itself an object of critique: inequality need not appear as explicit discriminatory rules but may persist through apparently objective efficiency, optimization, and risk management. Benjamin accordingly situates algorithms within existing structures of housing, education, policing, and employment; technology's social consequences depend on the institutional histories it enters. Abolitionist tools also differ from mere fairness repairs. The question is not how existing systems can operate more evenly across groups, but whether some systems inherently serve unequal classification and punishment. Technological governance thus involves improving models and more fundamental institutional choices: which judgments should never be delegated to automation? Benjamin's abolitionist orientation is particularly useful in distinguishing technological reform from institutional reform. Even if a predictive system achieves statistical fairness across groups, fairness metrics cannot answer the deeper question of legitimacy when the system's purpose is to expand punishment, exclusion, or overpolicing. Technology ethics must therefore preserve the institutional negation that asks whether the system should exist at all.]
- Noble, Safiya Umoja. Algorithms of Oppression. NYU Press, 2018. (Read: Introduction, pp. 1–14).
[Note: Noble shows that algorithmic ranking does not operate in a vacuum: it amplifies existing social prejudices. Unlike Crawford's data/labor perspective, it emphasizes classification and the politics of representation in information environments. Surveillance has no single central agent. Noble's Algorithms of Oppression offers not a final name for the mechanisms through which commercial datafication and state governance coexist rather than replace one another, but a theoretical opening testable against competing explanations; this openness is itself part of its research value. The central contribution of Noble's Algorithms of Oppression is not merely demonstrating biased search results. It shows how commercial ranking, advertising logic, and social representation jointly shape "relevance." Search becomes a site where cultural classification and market incentives overlap; racialized representation is not simply a data error. Through search engines, Noble shows how commercial ranking can repackage racial and gender prejudice as the "most relevant" results. Search is no neutral mirror of public knowledge: advertising markets, content ecologies, and platform design shape its classifications. Representation is therefore more than a cultural issue; when search becomes an entry point for understanding groups and social facts, ranking power also becomes epistemic power. Noble's cases further reveal infrastructure's cumulative effects. Search results respond to existing content while influencing subsequent clicks, content production, and public understanding, reinforcing certain associations. Using past behavior as a relevance signal can technologize prejudice through feedback loops. Algorithmic ranking both processes information and produces future information environments. The politics of search also concern epistemic authority. Once a platform organizes vast information algorithmically, users can easily take what appears first as the world's own natural ordering. Noble reveals this naturalization through the interface: technical ranking gives commercial and social biases the appearance of objectivity, making them harder to recognize as institutional choices.]
- Lyon, David. Surveillance Studies: An Overview. Polity, 2007. (Read: Introduction; chapters on surveillance as social sorting).
[Note: Lyon gives surveillance studies a conceptual framework broader than "being seen," emphasizing surveillance as an institutional process of collecting information, classification, and social sorting. This is a useful foundation because platforms, states, and commercial databases often need not watch individuals continuously: they allocate opportunities through categories of identity, risk, consumption, and eligibility. Being seen and being datafied are distinct. Through Lyon's Surveillance Studies: An Overview, the analytical gains and nonnegotiable historical limits of colonial analogy enter organizational and institutional analysis. At the scale of Lyon's Surveillance Studies: An Overview, those gains and limits require attention to continuity and transformation together: old institutions acquire new operations through new technologies, while new technologies may also be reabsorbed into existing institutions. The value of Lyon's Surveillance Studies: An Overview is understanding surveillance as a relation of social organization rather than a single device. Identification, classification, visibility, and institutional purposes together determine surveillance's forms, furnishing a basis for comparing commercial platforms, state administration, and security systems. Lyon treats surveillance as social classification and organizational practice rather than mere "watching." Modern surveillance uses identity records, databases, and risk categories to allocate different treatment and rights of movement. Digital technology extends this capacity while continuing surveillance's long alliance with administrative organization. Situating surveillance studies within this sociological tradition helps avoid technocentric narratives confined to cameras, platforms, or privacy.]
- Andrejevic, Mark. Automated Media. Routledge, 2020. (Read: Introduction; selected chapters on automation, surveillance, and decision systems).
[Note: Andrejevic examines how automated media increasingly delegate perception, analysis, and response to data-intensive systems. The point is not to announce human replacement by machines, but to analyze how institutions expand monitoring beyond human processing capacity and manage information environments through automated correlations and predictions. Reading this alongside Zuboff distinguishes more carefully the relations among accumulation, media automation, and governing capacity. The question of bringing data justice, epistemic authority, and differentiated exposure into one framework, raised by Andrejevic's Automated Media, exceeds an internal conceptual debate. It therefore remains subject to empirical evidence and competing explanations beyond Andrejevic's Automated Media. Automated media's central problem is consequently not machine autonomy, but which judgments cease to receive public explanation when perception-analysis-response chains are compressed. Large-scale automation greatly expands monitoring and response, yet can distance the grounds of decisions from those affected. Andrejevic emphasizes precisely this politics of scale: automation changes how much institutions can see, how quickly they process it, and under what conditions they act. Automated media also alter political action's timescale. Systems flag, block, or recommend before humans understand; institutions may increasingly rely on machines reacting first and explanations arriving later. Speed becomes a resource of power: organizations that process enormous signal volumes in real time have a structural advantage over individuals who can only appeal afterward. The politics of automation must analyze this temporal asymmetry too. Once speed combines with institutional coercion, new procedural problems arise: flags occur instantly, while explanation and correction take a long time. Asymmetrical appeal costs thus become a significant measure of automated governance.]
- Turow, Joseph. The Daily You: How the New Advertising Industry Is Defining Your Identity and Your Worth. Yale University Press, 2011. (Read: Introduction; selected chapters on behavioral targeting and audience classification).
[Note: Beginning with organizational changes in advertising, Turow traces how audiences are segmented, valued, and continually targeted. This supplies an important industrial history for surveillance capitalism: extraction was not invented in a vacuum but connects with advertising markets, audience measurement, customer relationship management, and differential pricing. "Identity" here is primarily a commercial classification, not necessarily subjective self-identification. At the analytical scale of Turow's The Daily You: How the New Advertising Industry Is Defining Your Identity and Your Worth, the historical relations among extraction, surveillance, racialized classification, and coloniality first appear as concrete relationships. This industry's history also shows that data's value does not naturally reside in individual behavior: it is produced through market categories and tradable audience imaginaries. "High-value users," "customers likely to churn," and "high-risk populations" are categories organizations construct for specific purposes. Individualized marketing and social stratification may therefore reinforce one another on the same infrastructure rather than constitute independent processes. Turow's industrial history reveals personalization's other face: finer segmentation enables more differentiated treatment. Personalized recommendations not only improve matching efficiency but can allocate different prices, discounts, and information environments to different consumers. The boundary between commercial classification and stratified opportunity becomes blurred, giving advertising technology broader effects of social sorting. Advertising technology allocates not just product information but the attention and discounts businesses consider worth investing in, so commercial valuation directly participates in social differentiation.]
- van Dijck, José. "Datafication, Dataism and Dataveillance: Big Data between Scientific Paradigm and Ideology" Surveillance & Society 12, no. 2 (2014): 197–208. (Read: entire article).
[Note: Van Dijck distinguishes datafication, dataism, and dataveillance, helping us see "everything can become data" as both technical practice and cultural ideology with epistemological commitments. Data do not automatically possess the status of truth; platforms' and research institutions' trust in large-scale behavioral data itself requires explanation. This illuminates Rouvroy's governance through correlation, but here the emphasis is on the knowledge-cultural conditions legitimizing extraction. Van Dijck's "Datafication, Dataism and Dataveillance: Big Data between Scientific Paradigm and Ideology," Turow's The Daily You: How the New Advertising Industry Is Defining Your Identity and Your Worth, and D'Ignazio's Data Feminism place the coexistence rather than mutual displacement of commercial datafication and state governance within different intellectual traditions. Surveillance has no single central agent. Van Dijck's article releases these mechanisms from overly abstract diagnoses of an era. Dataism matters because it elevates datafication from a technical operation into an epistemological belief: if records are comprehensive enough and computation complex enough, social behavior seemingly becomes directly apprehensible. Critiquing this belief supplies algorithmic governance with a necessary epistemological dimension. Data volume cannot replace conceptual choices; correlation cannot itself answer questions of causality, meaning, or legitimacy.]
- D'Ignazio, Catherine, and Lauren F. Klein. Data Feminism. MIT Press, 2020. (Read: Introduction; Chapters 1–3).
[Note: D'Ignazio and Klein place data practices within feminist intersectional analysis, emphasizing power, context, embodiment, and omitted experience. Rather than a generic AI fairness manual, this asks researchers who defines datasets, who is recorded, who is absent, and who bears classification's consequences. Alongside Browne, Benjamin, and Noble, it moves critique from system bias to the politics of data production itself. Being seen and being datafied are distinct. Through D'Ignazio's Data Feminism, the analytical gains and nonnegotiable historical limits of colonial analogy can be separated into interrelated elements. From D'Ignazio's Data Feminism, those gains and limits cannot stand apart from institutional authorization and resource distribution: technical forms produce durable social consequences only when combined with organizational power. Data Feminism embeds power analysis directly in data practices. From posing questions through classification and visualization to interpretation, it tracks who is represented, who is omitted, and who bears consequences. More data on women or minorities are not a sufficient answer; existing power structures and binary classifications must be challenged. Data justice therefore concerns not just correcting bias but redesigning knowledge production's purposes and participatory relations. Data Feminism particularly emphasizes context and pluralism. Detached from collection contexts, datasets readily abstract particular social relations into natural facts, while plural epistemic positions expose this abstraction's costs. Democratizing data practice thus entails more than open downloads: it includes documenting provenance, retaining uncertainty, and involving affected groups in defining problems.]
- Birhane, Abeba. "Algorithmic Colonization of Africa" SCRIPTed 17, no. 2 (2020). (Read: entire article).
[Note: Birhane uses algorithmic colonization to describe asymmetrical power reproduced when digital technologies enter African societies through universalist, developmentalist, and efficiency discourses. Outside actors often control technology, capital, data, and standards, while affected communities become deployment sites or data sources. This speaks directly to Couldry and Mejias's data colonialism and insists that "colonial" must be demonstrated through specific ownership, epistemic authority, and institutional dependence, not merely used rhetorically. Birhane's "Algorithmic Colonization of Africa," D'Ignazio's Data Feminism, and Ricaurte's "Data Epistemologies, the Coloniality of Power, and Resistance" move the integration of data justice, epistemic authority, and differentiated exposure from macro-level judgment to mechanisms. Birhane's contribution is precisely to bring that broad judgment back to identifiable institutional mechanisms. Her article also resituates the question historically: similar digital appearances need not have identical mechanisms, and conceptual validity depends on preserving differences. Birhane returns "colonial" from metaphor to institutional relations: who designs models, cloud services, and standards, who defines knowledge needs, and where technological gains flow are concrete grounds for assessing coloniality. Algorithmic colonization then describes not all transnational technological circulation but persistent asymmetrical dependence and its erosion of local knowledge, autonomy, and public capacity.]
- Ricaurte, Paola. "Data Epistemologies, the Coloniality of Power, and Resistance" Television & New Media 20, no. 4 (2019). (Read: entire article).
[Note: Ricaurte connects datafication to the coloniality of power, emphasizing knowledge production as well as resource extraction: what can be known, how it is classified, and who may define data results as social reality. This moves data colonialism beyond an economic metaphor toward epistemic authority and importantly introduces Latin American decolonial theory into data studies. Ricaurte's "Data Epistemologies, the Coloniality of Power, and Resistance" gives clearer causal layers to the historical relations among extraction, surveillance, racialized classification, and coloniality. She further situates data within knowledge production's coloniality: not all worlds are datafied alike, and not all forms of knowledge enter databases and models without loss. Choosing categories, variables, and measurable objects already embeds epistemological politics within infrastructure. Resistance therefore concerns not only privacy or property, but who may determine what counts as knowledge. This critique also concerns data sovereignty. If local communities supply only raw data while outside institutions monopolize categories, model design, and interpretation, knowledge production reproduces a center-periphery structure. Decolonization must redistribute power over research agendas, infrastructure, and interpretation, not merely add representative data. Data decolonization accordingly includes infrastructure: epistemic autonomy remains limited when computing, storage, and modeling depend on external actors. Data infrastructure itself becomes contested terrain for knowledge politics and decolonial practice. Rights over knowledge and infrastructure cannot be cleanly separated here.]
- Magnet, Shoshana Amielle. When Biometrics Fail: Gender, Race, and the Technology of Identity. Duke University Press, 2011. (Read: Introduction; selected chapters on fingerprinting, facial recognition, and biometric failure).
[Note: Through biometrics, Magnet shows how supposedly objective identity depends on databases, sensors, thresholds, and existing social classifications. Failure is more than an engineering error: it is distributed unevenly across bodies and may recode race, gender, and nationality as problems of identification. This supplies a highly concrete technical object for broader surveillance/data coloniality analyses. Surveillance has no single central agent. Biometrics particularly clearly reveal a loop between categories and bodies: systems train and validate against existing demographic categories, then return those categories to borders, security, and identity regimes as technical identification. Failure rates are not merely engineering defects; when identification carries compulsory consequences, errors are distributed unevenly along existing social differences. Magnet emphasizes that failure is no accidental event outside the system. When biometrics are declared objective, automatic, and universal, institutions treat misidentification as individual abnormality rather than a classificatory problem. Technical failure is thus politically distributed: some bodies must bear greater costs of verification, suspicion, and proving themselves. Accuracy must be assessed alongside these unequal consequences. Political assessment must relate error rates to the consequences of error, since the same statistical error imposes entirely different social costs in different identity regimes. Technical precision and institutional justice cannot be unrelated evaluative dimensions. In identity cards, borders, or welfare systems, biometric failure is not merely a user-experience issue: it can affect movement, access to resources, and the ability to prove legal status. Technical performance must therefore be linked to institutional consequences: the same misidentification rate has entirely different political weight in entertainment applications and compulsory identity systems.]
- Mejias, Ulises A., and Nick Couldry. Data Grab: The New Colonialism of Big Tech and How to Fight Back. University of Chicago Press, 2024. (Read: Introduction; chapters on data territories, colonialism, and resistance).
[Note: This book continues and updates the authors' data colonialism argument, placing the ongoing datafication of everyday life and large technology firms' infrastructure expansion within colonial relations' long history. Compared with 2019's The Costs of Connection, it more explicitly addresses contemporary Big Tech and resistance strategies. It therefore works as subsequent debate within the same session rather than assigning two books the same function in separate sessions. Focus on colonial analogy's analytical gains and historical limits. Being seen and being datafied are distinct. Mejias's Data Grab: The New Colonialism of Big Tech and How to Fight Back turns those gains and nonnegotiable limits from a rhetorical proposition into institutional analysis. Data Grab continues the data colonialism framework but concentrates on Big Tech's conversion of connectivity, convenience, and infrastructure into continuing data appropriation. Its policy significance is shifting the problem from individual privacy choices to collective data relations: when leaving major platforms means losing work, social, or public-service opportunities, consent can hardly bear the entire burden of legitimacy. Resistance must therefore involve institutional restrictions and public alternatives. Mejias and Couldry further stress that the data grab does not depend on one-off coercive appropriation: it continues once connectivity becomes a condition of life. The harder platforms become to leave, the easier extraction becomes to package as a natural condition of service exchange. The institutional question then becomes whether data minimization, collective rights, and public infrastructure can make social participation cease to require unlimited surrender of data.]
- Taylor, Linnet. "What Is Data Justice? The Case for Connecting Digital Rights and Freedoms Globally" Big Data & Society 4, no. 2 (2017). (Read: entire article).
[Note: Taylor proposes data justice to connect privacy and data protection with wider social justice. It helps shift critique into normative questions: if datafication produces inequalities of visibility, representation, resource allocation, and agency, institutional evaluation cannot focus solely on individual consent. It can close this session by converting diagnoses of surveillance and coloniality into an agenda on rights, capabilities, and global inequality. Data justice extends digital rights beyond individual privacy to structural distribution. Who accesses data, who can refuse datafication, who bears automated errors, and who benefits from the data economy cannot be resolved by consent alone. Taylor connects governance with social justice while preserving institutional differences for cross-national comparison. Locating data justice within global digital rights avoids universalizing European and North American privacy-law experiences. Countries connect infrastructure differently to social protection, identity regimes, and development policy, giving digital rights different urgencies. Transnational justice must address both individual freedom and resource allocation rather than replicate one legal paradigm. Data justice's unit of analysis is therefore not merely an individual data subject, but data relations constituted by states, platforms, communities, and infrastructure. It also connects usually separate policy domains: privacy concerns control over data, antidiscrimination concerns classification's consequences, and development justice concerns the distribution of infrastructure and capabilities. Taylor brings these into one normative space and reminds us not to assume an individual consumer with ample choices as the default subject of digital rights.]
Session 4: Planetary Computation and Infrastructure Space
This session returns the "cloud" to land, energy, chips, data centers, submarine cables, satellites, spectrum, cities, and legal arrangements. Bratton's The Stack describes planetary computation as a layered architecture; Easterling's Extrastatecraft analyzes how infrastructure space produces quasi-governing effects through standards, logistics, and spatial formats. Star reminds us that infrastructure usually recedes into the background when working smoothly and reappears through breakdown, conflict, or exclusion. Platform power therefore resides not just in interfaces, algorithms, and business models, but in conditions of connection, material resources, and maintenance systems.
Infrastructuralization also changes the relation between authority and territory. Sassen's reassemblage of territory, authority, and rights, Graham and Marvin's differentiated urban networks, and Plantin and colleagues' encounter between platform and infrastructure studies all show the increasingly unstable boundary between platforms and infrastructures. Through APIs, cloud services, identity systems, and payment interfaces, some platforms become underlying conditions other organizations cannot easily bypass. Traditional infrastructures, meanwhile, increasingly adopt platform logics of datafication, service provision, and tiered access. The political question is not whether platforms simply replace states, but how different authorities are redistributed through networks spanning multiple scales.
Varoufakis and Durand's technofeudalism theses become testable only against this infrastructural background. Whether cloud capital, intangible assets, platform gateways, rents, and exit barriers constitute dependencies distinct from traditional capital-labor relations requires point-by-point comparison with platform capitalism, monopoly capitalism, and rent theory. Critics such as Morozov are important precisely because they force the concept to specify its additional explanatory value. If "feudalism" merely renames monopoly, rent, and market power, its theoretical necessity is limited. If platform control of infrastructural conditions genuinely changes market entry, ownership, and dependence, the institutional levels at which this change occurs must be specified.
Required Readings:
- Bratton, Benjamin H. The Stack: On Software and Sovereignty. 10th anniversary ed., with a new preface by the author. MIT Press, 2026. Originally published 2016. (Read: new preface; "The Nomos of the Cloud"; "Platform and Stack, Model and Machine"; "Cloud Layer").
[Note: Bratton's stack is not simply a theory of the cloud-computing state. "Nomos of the Cloud" examines planetary computation's challenge to political geography; "Platform and Stack" explains how computational architecture becomes a political model; "Cloud Layer" situates data centers, platforms, infrastructure, and state governance in one architecture. Rather than directly calling Google or Amazon states, closely examine how sovereignty is distributed across layers and organizations. Use MIT Press's 2026 tenth-anniversary edition: revisit the original architectural account of planetary-scale computation, then use the new preface to test its validity and blind spots a decade later. The six-layer model is illuminating not because reality neatly divides into Earth, Cloud, City, Address, Interface, and User, but because it traces relations across layers: how energy and minerals support the cloud, how addressing brings objects into a computable order, how interfaces present institutional rules as actionable choices, and how a user may be a person, organization, sensor, or machine agent. It is a model of models, not an empirical map. The 2026 edition invites the question of which changes in sovereignty, cloud platforms, AI agents, and geopolitics strengthen the stack and which require revising its hierarchical imagination. The Stack offers a cross-scalar architectural language: its six layers are not merely technical but analytical slices of computation's reorganization of territory, identity, and sovereignty. Its advantage is placing data centers, cities, and interfaces within a single planetary system; its risk is that architectural abstraction obscures regional differences. Use it as a comparative framework rather than a complete world map.]
- Easterling, Keller. Extrastatecraft: The Power of Infrastructure Space. Verso, 2014. (Read: Chapter 2, "Zone: The Infrastructure Commons," pp. 25–70).
[Note: Easterling is not simply saying special zones are lawless spaces. She examines how infrastructure space, through spatial products, protocols, standards, logistics, and institutional arrangements, generates governing effects beyond traditional state planning. Applied to digital platforms, the analogy needs limits: Amazon's cloud and a free-trade zone are different spaces, although both show rules operating through infrastructure and standards rather than conventional legal texts. Calling platforms states, empires, or territories requires care. Free-trade zones and clouds are comparable only at the level of infrastructural regulation; their legal status, spatial structures, and economic relations must be treated separately. Infrastructure space is political precisely because it organizes conduct through technical specifications, spatial templates, and operational standards rather than always through explicit legal commands. Zones, logistics networks, and data centers are not vacuums outside the state; state policy, private capital, and transnational standards often produce them together. Easterling supplies a spatial language for analyzing quasi-governing effects. More importantly, infrastructure space frequently works through replication. Packaged as a standardized spatial product, a zone, industrial park, or logistics model spreads between regions carrying predetermined governing logics. APIs, cloud architectures, and service terms share this replicability, but their spatial and legal mechanisms differ. Comparison should remain at the level of rules spreading with infrastructure. This spatial politics shows how infrastructural power acts through default environments: once rules are embedded in spaces and systems, they are harder to identify than explicit commands. The core is not being outside law but how rules gain lasting force in spatial, standardized, and operational forms.]
- Star, Susan Leigh. "The Ethnography of Infrastructure" American Behavioral Scientist 43, no. 3 (1999): 377–391.
[Note: Star provides this session's indispensable methodological foundation. Infrastructure is hard to see because it becomes work and life's background; breakdown, conflict, or exclusion suddenly makes it visible. Studying clouds therefore requires investigating electricity, networks, server rooms, maintenance labor, and standards, not just interfaces and software. If existing theories of capitalism adequately explain rents, network effects, and monopoly, "feudal" is rhetoric; if there are new dependencies, gateway controls, and nonmarket allocations, explain why they require a new concept of social formation. In Star's "The Ethnography of Infrastructure," the political-economic distinctions among rent, profit, gateway control, and exit costs thus exceed conceptual questions. Star also places maintenance, repair, and exceptional experiences at the center. For most users, stable infrastructure recedes into the background; for maintainers, excluded people, and users encountering failures, it remains visible. Social differences are often clearest in disconnection, service suspension, incompatibility, or denied permissions. Infrastructure's relationality also means one system does not become background identically for all. Stable broadband may be invisible to urban middle classes but remain an explicit constraint for remote communities, disabled users, or maintenance workers. Star's ethnography inherently attends to difference: whose infrastructure is it? The maintenance perspective also changes innovation narratives. Reliability often depends on long-term, repetitive, low-visibility repair work rather than original design alone. Failure logs and maintenance sites often reveal actual power relations better than promotional materials.]
- Plantin, Jean-Christophe, Carl Lagoze, Paul N. Edwards, and Christian Sandvig. "Infrastructure Studies Meet Platform Studies in the Age of Google and Facebook" New Media & Society 20, no. 1 (2018): 293–310.
[Note: This article almost serves as the session's methodological manifesto: platforms and infrastructures can no longer be fully separated, since platforms acquire infrastructural properties while platform logics reorganize traditional infrastructures. It connects Star with Bratton and makes the cloud a form of social organization rather than merely a technical layer. Planetary-scale concepts can make particular places disappear. Macro models explain only when grounded in nodes—minerals, chips, data centers, cables, cities, labor, or jurisdictions—and when they test whether regions bear symmetrical costs. In the session's context, Plantin's "Infrastructure Studies Meet Platform Studies in the Age of Google and Facebook" gives the tension between planetary models and particular places, labor, and maintenance a concrete structure. The infrastructuralization of platforms and platformization of infrastructures captures Google's and Facebook's movement from optional services to nodes of social dependence. Once platforms become infrastructural, stability, universal access, and public responsibility become pressing; traditional infrastructures adopting platform logics may become datafied, commercialized, and organized through interfaces. Both changes blur public utilities and private platforms. This two-way transformation also rewrites public responsibility. Infrastructure is expected to be stable, universal, and available over the long term, while platform firms quickly change services, prices, and rules. When private platforms become unavoidable nodes for education, work, or public communication, corporate flexibility comes into conflict with infrastructural responsibility to society.]
Recommended Readings:
- Sassen, Saskia. Territory, Authority, Rights: From Medieval to Global Assemblages. Princeton University Press, 2006. (Read: Chapter 7, "Digital Networks, Global Cities," pp. 323–377).
[Note: Sassen's importance is that globalization does not simply make state sovereignty disappear: territory, authority, and rights are reassembled in transnational networks. This gives a more precise framework than the "platform state": different authorities coexist and overlap at particular spatial nodes. Digital infrastructure is therefore not naturally supranational space, but a site recombining old and new governing scales. Global assemblages undermine narratives of state withdrawal. Global digital networks need legal recognition, urban nodes, financial institutions, and cross-border rules; they can weaken some state capacities while reorganizing and strengthening others. Authority is reassembled across scales rather than simply transferred from states to platforms. Sassen also shows deterritorialization and reterritorialization often occurring together. Data flows seem to escape place, yet server siting, taxation, jurisdiction, and talent concentration anchor them anew in particular cities and countries. Global networks do not abolish space but reorder spatial relations across scales. Digital globalization thus recombines territory, authority, and rights within infrastructural networks rather than abolishing territory. Transnational digital networks always entail renewed concentration at nodes and competition among jurisdictions. The global is produced through particular places. Reorganizing authority also means firms and states are not in a zero-sum relation. States shape clouds through taxation, localization, and security policies; platforms use cross-border architectures to alter regulatory costs. Sassen's assemblage perspective analyzes this coproduction rather than reducing reality to who replaces whom.]
- Graham, Stephen, and Simon Marvin. Splintering Urbanism. Routledge, 2001. (Read: Chapter 1, pp. 1–36).
[Note: Differentiated urban infrastructure provision shows how networks produce spatial divisions. Alongside Easterling, it demonstrates that infrastructure need not produce publicness: prices, priorities, connections, and spatial choices can instead create service tiers. Calling platforms states, empires, or territories requires care. Splintering urbanism reveals that infrastructure has no inherently universalist direction. Networked services tier connectivity through prices, standards, and spatial investment, delivering speed, redundancy, and stability to high-value districts while peripheral areas suffer interruption and shortage. For digital infrastructure, this reveals connectivity itself as a new social difference. Fragmentation concerns not only access versus nonaccess, but connection quality, latency, bandwidth, reliability, and affordability. As infrastructure becomes a condition of work, education, and public services, technical differences become differences of opportunity. Infrastructure politics must exceed a single penetration-rate metric. Tiered connection quality eventually settles into tiered educational, employment, and participatory opportunities, making infrastructural equity explicitly distributive. Research must therefore examine service quality and spatial differences rather than merely count connections. Infrastructure differences ultimately become social opportunity differences. The distribution of quality connections, redundancy, and maintenance resources often reveals class and regional divisions in digital cities better than nominal access rates. Network equity must enter design and investment decisions. Splintering Urbanism also shows commercialization replacing universal-service logics with profitability. High-value users and areas gain priority investment; low-return spaces remain undermaintained. Digital networks relying wholly on similar commercial choices may further entrench existing urban class and regional divisions.]
- Mattern, Shannon. Code and Clay, Data and Dirt: Five Thousand Years of Urban Media. University of Minnesota Press, 2017. (Read: Introduction; selected chapters on urban information infrastructures).
[Note: Mattern extends urban media history across writing, maps, architecture, logistics, databases, and sensing systems, refusing to treat the smart city as a suddenly arriving digital revolution. Her key method juxtaposes code and clay, data and dirt: information infrastructures always depend on material carriers, spatial institutions, and maintenance labor. This provides Bratton's macro stack with a more historical, urban-scale correction. Through Mattern's Code and Clay, Data and Dirt: Five Thousand Years of Urban Media, the political-economic distinctions among rent, profit, gateway control, and exit costs enter organizational and institutional analysis. Mattern treats cities as media environments long constituted by writing, sound, masonry, maps, sensors, and databases, resisting smart-city accounts of digitization as sudden revolution. Urban computation must rest on existing roads, utility networks, administrative archives, and social memory. Digital infrastructure is therefore not an invisible network covering old cities, but something continually layered over and rubbing against historical material strata. Her long urban media history also resists data-city narratives that dismiss older infrastructure as obsolete background. Sensors depend on roads, databases inherit archival categories, and interfaces are constrained by architectural and administrative boundaries. Digitization is never a blank-slate redesign: it continually layers with existing materials, institutions, and histories. This perspective substantially reduces the sense of technological rupture in smart-city discourse. Mattern also expands media from screens to the city itself. Street signs, archives, building materials, and sensors jointly organize information; the smart city is another layer of this long history of mediation, not the first time the city becomes computable.]
- Hu, Tung-Hui. A Prehistory of the Cloud. MIT Press, 2015. (Read: Introduction; selected chapters on cloud infrastructure, sovereignty, and labor).
[Note: Through the cloud's prehistory, Hu dismantles language that imagines networks as placeless, frictionless space. Connecting clouds with earlier networks, military systems, mainframes, data centers, and labor relations, he shows the cloud as historically formed infrastructure and political imagination, not a natural technical layer. Alongside Bratton, compare architectural modeling and media archaeology as approaches to the same object. Planetary concepts can erase particular places. The tension between planetary models and specific sites, labor, and maintenance revealed by Hu's A Prehistory of the Cloud exceeds internal conceptual debate; it remains testable through empirical evidence and competing explanations beyond Hu's A Prehistory of the Cloud. Hu's A Prehistory of the Cloud confines that tension to particular objects and causal chains; analogies beyond this scope require new evidence, not terminology alone. Tracing the cloud metaphor's prehistory reveals how placelessness, weightlessness, and unlimited expansion obscure military, network, and data-center foundations. The cloud's virtuality is an interface effect, not a material fact. Concentrated data centers, dispersed terminals, and hierarchical network routes show that clouds create spatial abstractions while reorganizing concrete geographical dependencies. The metaphor also has political-psychological effects: users more readily understand storage and computation as natural environments rather than corporate infrastructure built and sold as services. Services seem as ubiquitous as air while depending on a few hyperscale operators. Hu's historicization exposes this naturalization as a cultural product of the technology industry.]
- Parks, Lisa. Cultures in Orbit: Satellites and the Televisual. Duke University Press, 2005. (Read: Introduction; selected chapters on satellite infrastructures and global visibility).
[Note: Parks extends media infrastructure into orbital space, examining how satellites connect ground, signals, images, and geopolitics. For planetary computation, this reminds us that "planetary" is not a rhetorical "global," but a vertical space composed of orbits, spectrum, ground stations, state licenses, and investment. Clouds also need locating in these material layers. At the scale of Parks's Cultures in Orbit: Satellites and the Televisual, the connection of clouds, platforms, and infrastructure with land, energy, standards, and sovereignty appears first as concrete relations. Technofeudalism must be compared with monopoly, rent, and platform capitalism. If "feudal" only renames market power, its explanatory gain is limited; if infrastructural control changes dependence, entry, and returns, identify the institutional levels involved. Satellite media bring infrastructure's vertical dimension into digital politics. Orbital positions, ground stations, spectrum allocation, and remote sensing jointly constitute communication conditions, so global coverage always depends on specific geopolitical arrangements. Parks turns planetary scale from macro rhetoric into traceable technical objects and resource competition. From satellite television to satellite internet, orbital infrastructure shows communication sovereignty's vertical space. State borders cannot fully explain signal coverage, yet launch capabilities, spectrum coordination, and ground reception remain institutionally controlled. Planetary computation's globality is concretely produced through scarce positions, technical licenses, and capital investment. Satellite networks thus show that global coverage does not abolish boundaries; boundaries extend from the ground into institutional arrangements for orbits, spectrum, and ground stations.]
- Lehdonvirta, Vili. Cloud Empires: How Digital Platforms Are Overtaking the State and How We Can Regain Control. MIT Press, 2022. (Read: Introduction; selected chapters on platform governance and institutional power).
[Note: Lehdonvirta compares large platforms as institutions capable of rulemaking, enforcement, and dispute resolution, asking why they assume some traditional governing functions in cross-border digital markets. "Overtaking the state" in the title is a proposition to test, not a factual description. Its value is distinguishing state sovereignty, private governance, market infrastructure, and user dependence rather than declaring platforms already to be states. Lehdonvirta's Cloud Empires: How Digital Platforms Are Overtaking the State and How We Can Regain Control, Parks's Cultures in Orbit: Satellites and the Televisual, and Varoufakis's Technofeudalism: What Killed Capitalism place the two-way infrastructuralization of platforms and platformization of infrastructures in different intellectual traditions. Calling platforms states, empires, or territories requires care. Lehdonvirta's Cloud Empires: How Digital Platforms Are Overtaking the State and How We Can Regain Control frees this two-way process from overly abstract epochal judgments. Technofeudalism must be compared with monopoly, rent, and platform capitalism. If feudalism only renames market power, its explanatory gain is limited; if infrastructure control changes dependence, entry, and returns, identify the institutional levels involved. Describing platforms as empire-like governing structures concerns their powers of rulemaking, dispute resolution, identity management, and sanction, not a rhetorical analogy with states. Platform authority, unlike traditional states, derives from contracts and infrastructure dependence; it may reach across borders without territorial sovereignty. The comparison is most valuable in clarifying the sources and limits of different governing capacities.]
- Varoufakis, Yanis. Technofeudalism: What Killed Capitalism. Bodley Head, 2023. (Read: Chapter 3, "Cloud Capital," pp. 58–91; Chapter 4, "The Rise of the Cloudalists and the Demise of Profit," pp. 92–115; Chapter 7, "Escape from Technofeudalism," pp. 175–214).
[Note: Varoufakis's central claim is that cloud capital changes the traditional relation of competition and profit, producing new dependencies through platform gateways, data, and cloud infrastructure. The useful question is not whether to accept the word technofeudalism, but whether it explains rents, network effects, monopoly, and user dependence. If monopoly capital and platform capitalism still explain them, what does feudalism add? Conversely, does it reveal power relations difficult to explain with traditional capital theory? Distinguish cloud capital from ordinary digital capital. The argument is not that large tech firms are feudal because they are large: platforms accumulate infrastructure that shapes behavior, organizes transactions, and extracts cloud rent from ecosystem participants, increasingly arranging markets inside private platforms. Read Chapters 3 and 4 together: the former defines cloud capital, the latter introduces cloudalists and changes in profit/rent relations. Chapter 7 shows the account must confront exit and political reorganization, not merely diagnose an era. Test whether Amazon Marketplace, app stores, and cloud services share this structure. The cloud-capital thesis argues platforms are no longer just firms in capitalist markets: cloud infrastructure and user behavior let them shape transaction environments and extract cloud rent from dependency. The crucial dispute is not whether "feudal" is striking, but whether relations among profit, rent, markets, and platform power have structurally changed. If conventional monopoly theory already explains these returns, technofeudalism's additional explanatory power must be demonstrated explicitly.]
- Durand, Cédric. How Silicon Valley Unleashed Techno-Feudalism: The Making of the Digital Economy. Translated by David Broder. Verso, 2024. Originally published as Technoféodalisme: Critique de l'économie numérique in 2020. (Read: Chapter 2, "On Digital Domination"; Chapter 3, "The Rentier Class of the Intangible"; Chapter 4, "The "Techno-Feudal" Hypothesis").
[Note: Durand's structure particularly suits comparison with Varoufakis: Chapter 2 covers digital domination, Chapter 3 intangible assets and the rentier class, and only Chapter 4 systematically proposes the technofeudal hypothesis. Distinguish three levels: mechanisms of domination, forms of accumulation, and a hypothesis about social formation. Compare Marxist rent theory, monopoly capitalism, and platform capitalism rather than equating rent with feudal ground rent. Similar vocabulary conceals different paths: Durand emphasizes how digital domination, intangible assets, intellectual property, and rentier logics reorganize accumulation, placing technofeudalism at the boundary between capitalism's internal development and potential transformation of social form. Because Chapters 2–4 progress cumulatively, do not jump to the feudal hypothesis. Establish domination and intangible rent first, then ask whether these mechanisms warrant naming a social formation. Reading the books together is not a contest but a comparison of their criteria for feudalism. Durand's How Silicon Valley Unleashed Techno-Feudalism: The Making of the Digital Economy, Varoufakis's Technofeudalism: What Killed Capitalism, and Thorhauge's The Economic Lives of Platforms: Rethinking the Political Economy of Digital Markets bring the scalar tension between planetary models and particular places, labor, and maintenance down from macro judgment to mechanisms. Planetary concepts can erase particular places. Durand's argument is best separated into three questions rather than accepted as an epochal name: how domination forms, how intangibles generate rents, and whether these changes amount to a new social formation. Feudalism then becomes a testable political-economic hypothesis, not a predetermined conclusion. Platform dependency, intellectual property, and monopoly rent must be analyzed separately.]
- Thorhauge, Anne Mette, Andreas Lindegaard Gregersen, Eva Iris Otto, Jacob Ørmen, and Morten Axel Pedersen, eds. The Economic Lives of Platforms: Rethinking the Political Economy of Digital Markets. Bristol University Press, 2024. (Read: "Rethinking the Political Economy of Digital Markets," pp. 1–16; "Modes of Connectivity," pp. 17–36).
[Note: This collection provides finer-grained comparisons for the technofeudalism debate. Rather than begin with an overarching social formation, it examines how connectivity, market design, data, and mediation organize economic activity. After Varoufakis and Durand, it forces large terms such as cloud landlord and rent into researchable mechanisms: which gateways are controlled, how prices are set, how supply and demand are organized, and where exit costs actually occur. Platforms' economic lives exceed market intermediation. Connections, pricing, data accumulation, and rulemaking jointly determine value's creation and distribution, with significant differences across platform types. The collection moves political economy from a single business model to comparative analysis, preventing a few giant firms from standing for the whole platform economy. Connectivity modes also determine transaction entry, participants' statuses, and which relations become datafied. An economics confined to prices and network effects misses connections' institutional nature. Treating connection as designed relations reveals how market structures are written into technical interfaces. Platform political economy must study how interfaces institutionalize economic relations, not only seek power in financial statements or market shares. Connections themselves constitute institutional design, not just a precondition for transactions. This is an important distinction from ordinary market intermediaries. Defining eligibility, transaction units, and visible information already preorganizes market relations in interfaces. Platform power thus operates before market formation.]
- Mezzadra, Sandro, Niccolò Cuppini, Mattia Frapporti, and Maurilio Pirone, eds. Capitalism in the Platform Age: Emerging Assemblages of Labour and Welfare in Urban Spaces. Springer, 2024. (Read: "Introduction. The Platform Age," pp. 3–14; "Operations of Platforms," pp. 15–31; "The Politics of Platforms," pp. 81–95).
[Note: This open-access collection advances from classic digital-labor theory to recent platform-labor research. It treats platforms as infrastructures organizing labor, urban space, welfare, and social cooperation, bringing valorization, algorithmic management, worker subjectivity, and alternatives into one framework. Alongside Scholz and Woodcock, ask whether platform cooperativism changes only ownership or also labor organization and social reproduction more broadly. Calling platforms states, empires, or territories requires care. Data centers, cities, borders, supply chains, and maintenance sites show how macro architectures are executed and whether environmental and labor costs are symmetrical across regions. Capitalism in the Platform Age moves platform studies into the intersection of cities, welfare, and labor reproduction. Platforms are not isolated apps: they may reorganize transport, delivery, care, and public services, jointly platformizing labor markets and urban space. Analysis must therefore track the rewriting of local institutions and redistribution of social-reproductive costs, not only firm valuations or extraction. The urban dimension shows platforms and local states do not simply replace each other. Delivery, ride-hailing, short-term rentals, and care platforms must encounter transport rules, housing policy, labor law, and welfare institutions. Expansion exploits local institutional differences while prompting local governments to redefine regulatory objects. Welfare effects particularly matter. As work splits into temporary tasks and multiplatform income, insurance, taxation, and protections founded on stable employment become harder to apply. Platform capitalism not only innovates corporate organization; it forces welfare states to redefine labor statuses and social rights. Cities are the institutional scale where this conflict first becomes visible.]
- Morozov, Evgeny. "Critique of Techno-Feudal Reason" New Left Review, nos. 133/134 (2022). (Read: entire essay).
[Note: Morozov supplies a particularly useful opposition within this session. He questions the hasty treatment of digital monopoly, intellectual property, platform rents, and financialization as evidence that capitalism has yielded to feudalism, asking us to reconsider the continuing roles of competition, profit, and state policy. This prevents the two technofeudalism books from dictating the debate and makes their concept a falsifiable, comparable hypothesis. "Critique of Techno-Feudal Reason" transforms the political-economic differences among rent, profit, gateway control, and exit costs from rhetoric into institutional analysis. These differences do not lead to one conclusion; they require comparing configurations of power across organizational settings to determine explanatory scope and failure conditions. Morozov's critique offers a necessary counterargument: prematurely announcing contemporary capitalism as feudalism may obscure the continuing mechanisms of capital, competition, finance, and the state. Conceptual novelty cannot replace evidence for historical periodization. Productive debate compares theories' precision in explaining the same phenomena—platform rents, monopoly, infrastructural dependence. Morozov also reminds us that historical analogies shape strategy. Treating platforms as feudal lordships may favor restoring capitalist competition; treating the problem as capital concentration and privately owned infrastructure yields different priorities for antitrust, public ownership, and labor politics. Concept choice is therefore not a naming game: it organizes policy imagination.]
- Mosco, Vincent. To the Cloud: Big Data in a Turbulent World. Paradigm Publishers, 2014. (Read: Introduction; chapters on cloud computing, big data, and political economy).
[Note: From political economy and communication studies, Mosco demystifies the cloud's industry structure, labor, data, and ideology. Predating stack and technofeudalism accounts, it provides a historical marker: from the outset, cloud computing involved corporate concentration, data centers, energy, labor, and state policy, not merely technical outsourcing. Planetary concepts can erase particular places. Preserve falsifiability rather than maximize conceptual coverage; extend a framework only where evidence supports its central mechanisms. The tension between planetary models and specific places, labor, and maintenance thus remains testable through evidence and competing explanations beyond Mosco's To the Cloud: Big Data in a Turbulent World. Demystification means showing that clouds do not eliminate place: language hides servers, data centers, electricity, and labor from users' experience. Ideological effects and material architecture require simultaneous study. The more weightless and ubiquitous a service appears, the more its fixed investments in land, energy, and capital need tracing. Cloud concentration also has economies of scale. Capital, energy purchasing, and technical operations give large data centers cost advantages, so on-demand convenience and highly concentrated backends grow together. Demystification therefore concerns industrial structure as well as language: the more clouds resemble ubiquitous public environments, the easier their ownership becomes to obscure. Mosco also returns clouds to global communication politics. Concentrated data centers and services give a few firms cross-sector infrastructural positions, turning traditional media, governments, and ordinary businesses into clients. Clouds create infrastructural dependence; the political issues concern control over outages, pricing, and standards as well as privacy.]
Part III: Mediation, Organization and Labour (Weeks 5–6)
Session 5: Recursive Publics, Mediation and Digital Politics
This session shifts from control and extraction to the institutional conditions of publicness. Kelty's recursive public is not a crowd discussing things online, but a public form able to participate in maintaining, modifying, and reproducing its own technical and institutional foundations. Free-software communities matter theoretically not only because their code is open, but because licenses, collaborative norms, discussion procedures, version control, and infrastructure jointly constitute a recursively modifiable public life. Warner's analysis of publics and counterpublics emphasizes that publics are not fixed, preexisting groups: circulation, responses, and recirculation continually call them into being. Fraser further places unequal participation and counterpublics within public-sphere theory.
Kornbluh's Immediacy, or The Style of Too Late Capitalism gives this discussion a tension at the level of media form. Immediacy, transparency, immersion, and the absence of mediation are often treated as digital culture's natural values. Political organization, however, depends on procedures, representation, delay, explanation, memory, and chains of responsibility. Tufekci, Dean, Gerbaudo, and others further show that inexpensive mobilization and rapid circulation do not automatically produce sustained organizational capacity. Connections quickly enlarge participation, yet may also make decision-making, resource accumulation, and long-term coordination fragile.
Digital politics therefore turns not on an abstract choice between centralization and decentralization, but on which mediations participants can understand, hold accountable, and modify. A public's owning its servers is not the sole criterion; more important is its ability to influence rulemaking, resource allocation, structures of visibility, and dispute resolution. Only when both media conditions and organizational forms can become political questions does publicness avoid shrinking into a temporary attention space supplied by a platform. Recursive publics thus connect media technology, institutional design, and collective autonomy.
Required Readings:
- Kelty, Christopher M. Two Bits: The Cultural Significance of Free Software. Duke University Press, 2008. (Read: Chapter 1, "A Recursive Public," pp. 27–63).
[Note: This is the session's central text. Kelty's recursive public not only uses public media but can modify, maintain, and reproduce their infrastructural and institutional conditions. Focus on how software licenses, code, discussion, maintenance, and identity jointly form publicness. In comparison with the platform era, ask what remains of publicness when participants lose practical control over code, servers, accounts, and rules. Conversely, does enduring publicness require some autonomy over technical infrastructure? Participant counts and circulation speed cannot directly measure it. A highly visible group may still lack the ability to preserve archives, modify rules, own infrastructure, or determine exit conditions; distinguish these institutional dimensions from expression rights. Kelty's Two Bits: The Cultural Significance of Free Software grounds publics' ability to maintain and modify media conditions in institutional relations. Publicness is not synonymous with information traffic: it requires durable rules, memory, procedures, and media conditions. Platforms expand visibility but may place the interfaces, data, and communicative structures supporting publics under private governance. A recursive public is no ordinary online discussion group; it can reshape its own conditions of communication and collaboration. Free-software communities are recursive because members not only use code and licenses but maintain these institutional-technical foundations. This raises a sharp platform-era problem: how is organizational autonomy constrained when publics can form only on unmodifiable proprietary platforms? Recursion also means a public can in some sense repair itself. When software, licenses, or communication channels fail, members can change them through shared rules rather than wait for an outside platform's decision. Technical and democratic autonomy do not entirely coincide for Kelty, yet the right to maintain makes a crucial connection between them.]
- Kornbluh, Anna. Immediacy, or The Style of Too Late Capitalism. Verso, 2024. (Read: "Circulation," pp. 24–43; "Writing," pp. 65–112; "Antitheory," pp. 148–190).
[Note: Kornbluh's book is not directly about technology; it belongs here precisely to broaden the question of why platform-era publicness increasingly seeks directness, instantaneity, immersion, and no mediation. She calls this cultural form immediacy and argues that it lowers tolerance for representation, distance, and mediation. This creates a tension for recursive publics: rules, procedures, discussion, and maintenance sustain publicness, yet immediacy culture experiences these mediations as obstacles. What happens to organization's necessary delays, explanations, and institutional procedures when political communication demands immediate reaction? Read "Circulation," "Writing," and "Antitheory" together. Immediacy is not one media phenomenon, but a late-capitalist style linking circulation, first-person expression, immersion, transparency, and antitheoretical attitudes. Her polemical intensity also requires critique: does reducing mediation always mean political regression? When marginalized subjects seek to speak directly, how should we discuss representation's institutional value? Positioning Kornbluh among Warner, Fraser, and Kelty avoids one-sided praise of mediation; distinguish necessary, accountable, and monopolistic mediations. Mediation is not inherently antidemocratic: can it be visible, accountable, and reconfigured? Critiquing immediacy connects cultural form with political organization. When directness, immersion, the first person, and instantaneous circulation become values, mediation looks like alienation or delay. Yet public politics depends on irreducible mediations of representation, procedure, explanation, and organization. Digital speed thus has two sides: it extends responsiveness while potentially eroding the distance and time required for sustained judgment.]
- Warner, Michael. "Publics and Counterpublics" Public Culture 14, no. 1 (2002): 49–90.
[Note: Warner emphasizes that a public is not a fixed group existing before media: circulation, response, and recirculation continually summon it. This particularly matters for platforms, where recommendations, reposting, group chats, subscriptions, and algorithmic timelines alter public formation. Read with Kelty to examine how technical conditions produce publicness. Inexpensive mobilization and organizational capacity are distinct measures. Rapid assembly amplifies issues without necessarily accumulating trust, resources, divided roles, or conflict-resolution mechanisms; keep mobilization, organization, and institution distinct. "Publics and Counterpublics" offers a limited but effective scale for explaining differences among publics, counterpublics, networked publics, and recursive publics. It identifies mechanisms without replacing further analysis of organizations, law, and material conditions. Publics arise through texts' circulation and strangers' sustained responses rather than preexist them. Publicness is thus media-dependent: communicative rhythms, forms of address, and accessibility change who constitutes a public. Algorithmic timelines further intervene, so platform rankings and recommendations shape formation alongside participants' communicative acts. Counterpublics also organize themselves through expression and social rhythms different from the mainstream. Platforms lower their communicative costs while measuring all expression through standardized formats and metrics. This tension between standardization and counterpublic difference is essential to networked publicness.]
- Fraser, Nancy. "Rethinking the Public Sphere: A Contribution to the Critique of Actually Existing Democracy" Social Text, no. 25/26 (1990): 56–80.
[Note: Fraser revises classical public-sphere theory through subaltern counterpublics, participatory parity, and identity differences. Rather than make a unified sphere the only ideal, she asks how multiple publics, counterpublics, and marginal publics coexist. In this session, "Rethinking the Public Sphere: A Contribution to the Critique of Actually Existing Democracy" gives a concrete structure to the distinction among expression, infrastructural ownership, and democratic governance. Publicness is not information traffic: it requires durable rules, memory, procedures, and media conditions. Platforms expand visibility but may privately govern the interfaces, data, and circulation structures that sustain publics. Counterpublics also revise the normative image of one shared public sphere. Social inequality means resources, languages, and visibility for discussion have never been equal; marginalized groups often first need their own counterpublic spaces. Digital networks expand differentiation while platform ranking and commercial logics shape encounters among publics. Participatory parity further shows that formal openness does not guarantee equal participation. Resources, status, and expressive habits affect who is heard. Platforms lower speaking thresholds yet create new inequalities through attention competition. Multiple publics must therefore be studied alongside actual differences in participation conditions. Counterpublics' value is preserving difference within publicness rather than requiring everyone first to adopt one language and expressive norm.]
Recommended Readings:
- Gramsci, Antonio. Selections from the Prison Notebooks. Edited and translated by Quintin Hoare and Geoffrey Nowell Smith. International Publishers, 1971. Originally published in Italian selections from the prison notebooks, 1948–1951. (Read: "The Intellectuals," pp. 5–23).
[Note: This reading does not simply rename programmers organic intellectuals; it reconsiders the links among knowledge, production, and organization. Gramsci examines intellectuals' emergence from particular social forces and their roles in culture, organization, and hegemony. In digital environments, examine separately the knowledge positions of software engineers, protocol developers, platform managers, data workers, and researchers. Technical competence does not naturally represent a political class. Participant counts and circulation speed cannot directly measure publicness. Gramsci's Selections from the Prison Notebooks moves publics' capacity to maintain and modify media conditions from macro judgment to mechanism analysis; validity ultimately depends on identifying institutional relations empirically. Organic intellectuals connect intellectual labor with social groups and organized practice rather than give clever people special status. In digital society, the concept examines engineers, researchers, technical communities, and labor organizations forming shared worldviews and organizational capacities. Expertise itself does not determine political alignment: how knowledge is organized, which social forces it serves, and how it becomes collective practice are decisive. Distinguish technical authority from political representation too. Knowing a system does not automatically entitle engineers to define other groups' public interests. Expertise becomes collective political capacity through organization, negotiation, and shared practice. This particularly matters for AI expert governance. Gramsci also emphasizes reciprocity: intellectuals organize social experience, while organizations give knowledge work sustained social positions. Publicness dependent solely on short-cycle opinion leaders struggles to develop this long-term function. How engineering communities, labor organizations, and public research institutions accumulate knowledge and transmit organizational memory is the real core of the "new organic intellectuals" question.]
- Habermas, Jürgen. The Structural Transformation of the Public Sphere. Translated by Thomas Burger. MIT Press, 1991. Originally published 1962. (Read: Part V, pp. 141–180).
[Note: This supplies public-sphere theory's historical baseline. The point is not to equate platforms with Habermas's mass media, but to understand the institutional, medial, and social conditions a public sphere requires and how commercialization changes them. Mediation is not inherently antidemocratic; it must be visible, accountable, and reconfigurable. Habermas still offers a key benchmark: publicness depends on institutional spaces, media forms, and social qualifications, not information circulation alone. The platform-era issue exceeds everyone's ability to speak: how business models, ownership, attention allocation, and inequality organize discussion, and which institutions sustain critical communication, matter more. Commercialization takes new infrastructural forms digitally. Traditional mass media shape discussion through ownership and advertising; platforms additionally control identity, ranking, circulation, and feedback. Theory must extend from content production to distribution architectures and private technical systems' specification of communicative conditions. Who controls infrastructure becomes unavoidable: media ownership and technical architecture require joint discussion. Publicness's material conditions now extend from newspapers, coffeehouses, and broadcasting to servers, interfaces, and recommenders. Media architecture becomes part of public-sphere institutions. Revisiting the sphere means not restoring old media orders, but analyzing how new infrastructures affect equal participation, issue formation, and public judgment. Habermas's normative framework cannot directly encompass fragmented network publics, but still offers measures: are participatory conditions equal, can agendas escape private control, and are reasons open to public contestation? Joined to platform infrastructures, these retain critical force without reviving the historical social model.]
- Benkler, Yochai. The Wealth of Networks. Yale University Press, 2006. (Read: Chapter 3, "Peer Production and Sharing," pp. 59–90).
[Note: Benkler offers an institutional and economic analysis of peer production. With Kelty, ask whether commons-based peer production supplies organizational forms distinct from firms and markets, and how platforms reabsorb their public outputs. Inexpensive mobilization is not organizational capacity. Through Benkler's The Wealth of Networks, differences among publics, counterpublics, networked publics, and recursive publics enter institutional and organizational analysis. Peer production's political potential comes from changing the relation between productive cooperation and exclusive property, not networks themselves. Open licenses, modular collaboration, and shared resources lower participation barriers but still require governing rules, maintenance, and sustainable material foundations. This sharply contrasts with platformization: open collaboration is not free platform labor. Distinguish nonmarket cooperation from unpaid corporate contribution. The former usually has shared rules, common resources, and institutions through which participants influence production; the latter may leave platforms in unilateral control of infrastructure and revenue. Both use networks, but their property and governance structures differ completely. This is crucial for open-source models. Even freely downloadable weights leave a significant institutional distance from commons-based peer production if one company still decides training, governance, and deployment.]
- Coleman, Gabriella. Coding Freedom. Princeton University Press, 2012. (Read: Chapter 3, pp. 61–84).
[Note: Coleman's ethnography brings free software and hacker culture from abstract technological freedom into particular communities. Focus on relations among technical ability, ethics, political identities, and organizational culture. The distinction between expression, infrastructural ownership, and democratic governance revealed by Coleman's Coding Freedom is not merely an internal conceptual debate. It remains subject to empirical evidence and competing explanations beyond Coleman's Coding Freedom. Coleman's Coding Freedom confines this distinction to particular objects and causal chains; analogies beyond that scope need new evidence rather than terminology alone. Coleman's Coding Freedom grounds free-software ethics in community practices: technical competence, humor, legal consciousness, hacker ethics, and organizational boundaries intertwine. Publicness arises not just from open licenses, but from participants' sustained commitments to code, rules, and shared identities. This ethnography particularly matters because free-software politics do not follow automatically from license texts. Humor, craft, reputation, legal consciousness, and everyday collaboration jointly shape community ethics. Technical autonomy is thus a practiced and organized cultural form, not a political property naturally appearing when code is opened. Coleman also shows how technological communities' politics often appear in legal details and everyday norms rather than grand declarations. Specific understandings of copyright, anonymity, free speech, and collaborative obligations shape conduct. Free-software culture is an institutionalized ethics stabilized by long practice, not by members sharing a single ideology.]
- Tufekci, Zeynep. Twitter and Tear Gas. Yale University Press, 2017. (Read: Introduction and Chapters 1–2).
[Note: Tufekci examines how networked media lower mobilization costs without necessarily strengthening long-term organizational capacity. With Kornbluh, distinguish immediacy's rapid circulation from organization's needs for memory, trust, accumulated resources, and procedures. Participant counts and circulation speed cannot directly measure publicness. At the scale of Tufekci's Twitter and Tear Gas, publics' ability to maintain and modify media conditions appears first as concrete relations. Tufekci's Twitter and Tear Gas explains that ability only insofar as corresponding institutional mechanisms are empirically identifiable. Through Tufekci's Twitter and Tear Gas, the question becomes more rigorous: which mechanisms actually change, which involve only a change of medium, and who bears the consequences? Tufekci's Twitter and Tear Gas further insists on differences among objects and institutional boundaries; rhetorical force cannot replace empirical explanation. Tufekci's Twitter and Tear Gas particularly shows that networked movements can gain enormous communicative scale almost instantly without simultaneously developing negotiation, resource management, or internal conflict-resolution capacities. A capacity deficit is not a dismissal of network mobilization, but an account of asymmetry between organizational foundations and communication speed. Networked protests show that digital media dramatically reduce mobilization costs, quickly producing scale and visibility without necessarily developing bargaining, decision-making, and conflict resolution. The experience traditional organizations slowly accumulate may be precisely what rapid mobilization lacks. Network movements' fragility may therefore arise from unequal growth of organizational and communicative capacity rather than failure of communication.]
- Dean, Jodi. Crowds and Party. Verso, 2016. (Read: Chapter 3, "The Distributive Party," pp. 119–156).
[Note: Dean questions the assumption that more connections mean stronger political organization. Networked circulation produces many individualized gatherings without necessarily producing enduring organizations. With Kelty, distinguish recursive publics from network crowds. Dean's Crowds and Party, Tufekci's Twitter and Tear Gas, and Gerbaudo's Tweets and the Streets: Social Media and Contemporary Activism place the conflict between immediacy and speed, on one hand, and organization's necessary mediations and delays, on the other, within different intellectual traditions. Mediation is not inherently antidemocratic; the question is whether it can be visible, accountable, and reconstructed. Dean's Crowds and Party frees this conflict from abstract epochal diagnoses. The contribution of Dean's Crowds and Party is to preserve theory's falsifiability. Dean's Crowds and Party does not simply reject crowd mobilization: it stresses the structural difference between momentary assembly and organizational forms capable of decisions, discipline, and sustained action. Digital media magnify the former without automatically producing the latter. This organizational question strips network connection of a naturally democratic aura. Rapid circulation produces shared feelings and short-term gatherings without necessarily creating organizations responsible for decisions, resources, and long-term commitments. The tension between crowd and party becomes crucial to digital activism: political capacity requires durable organizational forms as well as expression.]
- Gerbaudo, Paolo. Tweets and the Streets: Social Media and Contemporary Activism. Pluto Press, 2012. (Read: Chapter 1, pp. 1–20).
[Note: Gerbaudo's discussion of leaderless networks analyzes social media's redistribution of political visibility and leadership. It does not prove network movements undemocratic, but suggests interfaces may generate new centers. Low mobilization costs and organizational capacity are different measures. Gerbaudo's Tweets and the Streets: Social Media and Contemporary Activism separates distinctions among publics, counterpublics, networked publics, and recursive publics into connected elements. His correction to leaderless-network accounts is particularly important. Social media do not eliminate leadership: account visibility, emotional choreography, and calls to action produce new soft leadership. Networks can be formally distributed yet highly concentrated in attention and agenda-setting. Study implicit curation, coordination, and representation rather than judge decentralization from organizational self-descriptions alone. The choreography of assembly reveals organizational labor connecting online circulation and offline gatherings. Who selects places, times, and slogans, and who maintains information flows, shapes supposedly spontaneous movements. Digital media do not abolish organization but make some of its labor less conspicuous as curation, reposting, and emotional mobilization. Soft leadership also illuminates new representational problems. Highly visible accounts can coordinate effectively while acquiring the authority to interpret a movement without formal authorization. Network politics needs new accountability so communicative advantages do not automatically become unquestionable representative authority. Leadership and horizontal participation are questions of organizational design, not mutually exclusive choices.]
- Milan, Stefania. Social Movements and Their Technologies. Palgrave Macmillan, 2013. (Read: Chapter 2, pp. 31–54).
[Note: This examines movements building their own communication and media infrastructures. Extending recursive publics to political actors reveals a potentially vast difference between owning and using media. Milan's Social Movements and Their Technologies, Gerbaudo's Tweets and the Streets: Social Media and Contemporary Activism, and Juris's Networking Futures bring the distinction among expression, infrastructure ownership, and democratic governance from macro judgments to mechanisms. Historical analogies are valuable only at that level. Declaring an epochal shift merely because algorithms, platforms, or models have appeared destroys theory's ability to distinguish continuity from rupture. Milan's Social Movements and Their Technologies contributes by returning this broad judgment to identifiable institutions. The distinction consequently remains open to empirical evidence and competing explanations beyond Milan's Social Movements and Their Technologies. Movement technologies concern not just which media to choose but whether relative autonomy over communicative conditions can be built. Independent servers, encrypted communication, independent media, and technical volunteers may constitute movements' own infrastructural capacities. Milan thus turns media use into media governance: an organization's degree of control over technological conditions directly affects action's continuity.]
- Juris, Jeffrey S. Networking Futures. Duke University Press, 2008. (Read: Chapter 1, pp. 1–26).
[Note: Taking movements against globalization as cases, Juris observes networked organization emerging in actual action, avoiding its reduction to a purely technical structure. Participant counts and communication speed cannot directly measure publicness. Juris's Networking Futures clarifies causal layers in publics' ability to maintain and modify media conditions. Juris's Networking Futures explains this ability only where corresponding institutional mechanisms can be identified empirically. Around Juris's Networking Futures, trace connections among mechanisms rather than seek one term covering every phenomenon. The ethnography in Juris's Networking Futures further shows networked organization is no direct projection of technical topology onto political form. Meetings, trust, resources, traditions of action, and interorganizational negotiation sustain networks together; horizontalism always involves coordination and informal leadership. The ethnography of antiglobalization movements shows networking as organizational norms and political culture as well as technical structure. Horizontal deliberation, autonomy, and diversity are practiced through meetings, mailing lists, and action networks, with coordination costs and informal power. Network organizations' democracy is an institutional achievement requiring maintenance, not an automatic property of connected nodes. Juris also shows horizontalism's practical difficulties: informal networks do not eliminate power, since technical skill, resources, language, and ability to travel still generate centers. Organizational analysis must examine both formal structures and informal hierarchies. Horizontalism is better treated as an ongoing practical norm than an accomplished organizational fact.]
- Papacharissi, Zizi. Affective Publics: Sentiment, Technology, and Politics. Oxford University Press, 2015. (Read: Introduction; selected chapters on affective publics and networked storytelling).
[Note: Papacharissi centers affect, emotional expression, and narrative structures in networked publicness. Publics form not only through rational arguments and information exchange: affective rhythms, hashtags, visibility, and circulating stories temporarily organize them too. After Warner and Fraser, this supplements overly institutional or normative tendencies while remaining distinct from Berlant: here the focus is public formation, whereas the next session concerns attachment in precarious lives. Mediation is not inherently antidemocratic; visibility, accountability, and reconstruction are the questions. Affective publics free affect from simple opposition to public reason. Circulating hashtags, images, and narratives assemble dispersed individuals into temporary publics of shared feeling; emotional intensity itself becomes political connection. Can such connection become sustained organization, and how does platforms' optimization of visibility and interaction amplify selected emotions? Affect does not necessarily produce superficial politics. It sustains memory, identity, and common concerns, or may be depleted by short attention cycles. Distinguish formation from institutionalization: the former explains how connections happen, while the latter still requires organizational theory. Endurance ultimately depends on organization, memory, and institutional supports; emotional intensity cannot replace long-term political capacity. Affect's circulation also clearly depends on platforms. Certain emotions travel readily through short texts, visual impact, and engagement metrics, and ranking may further amplify highly engaging content. Affective publics are not spontaneous psychological aggregates: media forms and algorithmic visibility shape them. This infrastructural condition must inform any account of digital political emotion.]
- boyd, danah. "Social Network Sites as Networked Publics: Affordances, Dynamics, and Implications" In A Networked Self: Identity, Community, and Culture on Social Network Sites, edited by Zizi Papacharissi, 39–58. Routledge, 2011. (Read: entire chapter).
[Note: Boyd's networked publics explain how persistence, replicability, scalability, and searchability alter public interaction. The point is not that Facebook or Twitter automatically generate democratic public spheres, but to supply middle-range concepts explaining how affordances change speech's temporality, visibility, audience boundaries, and recirculation. Reading with Warner's circulation is especially productive. Low mobilization costs and organizational capacity are different measures. "Social Network Sites as Networked Publics: Affordances, Dynamics, and Implications" turns differences among publics, counterpublics, networked publics, and recursive publics from rhetorical propositions into institutional analysis. The framework emphasizes affordances' shaping of communication: persistence, replicability, scalability, and searchability change expression's time, reach, and retrievability. Social networks are thus not simply online traditional public spheres. Preserved visibility, copyable content, and blurred audiences bring identity management, context collapse, and participation together. Networked publics also alter imagined audiences. Speakers often address potential, invisible, algorithmically expandable audiences rather than defined groups, changing self-presentation and risk assessment. Context collapse is not individual failure to manage privacy: multiple social settings are technically compressed into one interface. The social tensions of network publics thus enter interface design.]
- Gillespie, Tarleton. Custodians of the Internet: Platforms, Content Moderation, and the Hidden Decisions That Shape Social Media. Yale University Press, 2018. (Read: Chapter 2, "The Myth of the Neutral Platform," pp. 24–44; Chapter 5, "The Human Labor of Moderation," pp. 111–140).
[Note: The book analyzes platforms' content governance through policies and technical rules while emphasizing moderation's dependence on extensive hidden human labor. It corrects the image of automatic algorithmic governance: algorithms, rules, human review, and commercial risk management commonly govern together. Content governance thus resembles a quasi-public institution yet lacks public bodies' usual democratic authorization and procedural constraints. Rulemaking, automated detection, and human review jointly determine visibility's boundaries, while firms control both interpretation and infrastructure. Gillespie makes platform responsibility concern governance's legitimacy, not merely deleting or retaining content. Moderation also exposes a private-platform paradox: firms claim neutral intermediation while constantly making public-order-like rules. Scale forces reliance on automation and global human labor, so interpretation becomes a continuous organizational process. Disputes cannot be reduced to a single opposition between free speech and censorship. Moderation is simultaneously content policy, a labor institution, and infrastructural governance; together these determine publicness's actual boundaries. Platforms are private firms and de facto communicative infrastructures, a tension underlying governing disputes. Moderation's scale also creates continual friction between rules and cases. Uniform policies encounter linguistic, cultural, and political differences, while automated detection cannot fully understand context. Platforms constantly adjust between global consistency and local judgment. Research should focus not only on contested individual decisions but on how this vast institution produces executable categories and exceptions. The politics of moderation also appear in handling exceptions. The more complex contexts are compressed into uniform policies, the more borderline judgments are required; these are precisely where platforms exercise de facto public power.]
Session 6: Platform Labour, Precarity and New Class Relations
This session brings the political economy of platforms into the labour process and the time of everyday life. Platforms reorganize employment through task allocation, ratings, dynamic pricing, task decomposition, identity verification and automatic deactivation. The power of the "boss" can thus appear as interfaces, metrics and models, without always being exercised directly by a visible manager. Scholz, Woodcock, Rosenblat and Aloisi/De Stefano approach this from cooperativism, workers' struggles, drivers' experiences and labour law. They show that algorithmic management has not abolished labour relations; information asymmetry and dispersed responsibility may instead make labour control harder to identify.
Terranova's free labor, Casilli's research on microtasks, and Gray and Suri's account of ghost work extend digital labour to user participation, content moderation, data annotation and training support. "Automation" often depends on extensive human labour that is geographically dispersed and erased from its public narrative. AI systems therefore cannot be understood solely through models and computing power. Platformization further transfers work-related risks to individuals: waiting, accepting tasks, maintaining ratings, paying for equipment and remaining continuously available become parts of the labour process. Class relations are consequently expressed not only in wages, but also in information rights, conditions of exit, data ownership and control over time.
Berlant's Cruel Optimism adds an affective and temporal dimension to this structure of experience. Precarious labour persists not simply because of external coercion, but also because visions of "flexibility," "freedom," "entrepreneurship" and "self-realization" can become objects of attachment that sustain people's life narratives. Platform cooperatives, unions, collective bargaining and public infrastructure therefore confront more than ownership: they must also address the distribution of risk, the organization of everyday time, and whether workers can constitute a sustainable collective political subject. Institutional alternatives must reach these levels to move beyond the limited idea of simply changing a platform's owner.
Required Readings:
- Berlant, Lauren. Cruel Optimism. Duke University Press, 2011. (Read: Chapter 1, "Cruel Optimism," pp. 23–50; Chapter 6, "After the Good Life, an Impasse," pp. 191–222; Chapter 7, "On the Desire for the Political," pp. 223–264).
[Note: Berlant does not offer a direct theory of platform capitalism, but provides a powerful concept for understanding why platform labour remains attached to "hope." Cruel optimism names a subject's attachment to an object that obstructs their flourishing. Chapter 6 situates this relation within the precarious present and the time of work and life; Chapter 7 develops the difficulties of political desire and action. Read after platform labour, it raises a question technical solutions often overlook: why do workers remain dependent on "flexible, autonomous, entrepreneurial" forms of work even when these continually erode security and the ability to plan a future? Institutional alternatives must change platform ownership and address an attachment sustained jointly by hope, everyday rhythms and precarity. Chapter 1's definition must be read alongside the impasse of Chapter 6 and political desire of Chapter 7. Berlant is concerned not merely with why people cannot relinquish a harmful object, but with how they improvise a life in a persistently unstable present after the infrastructures of the "good life" fail, and why political desire does not automatically crystallize into a collective programme. This matters especially for platform labour, since flexibility may be both a platform slogan and a real need under conditions of care, migration, education or unemployment. Only by acknowledging both the material rationality of attachment and its structural costs can cruel optimism avoid becoming the simplistic claim that "workers are brainwashed." Berlant offers an affective politics of attachment rather than an economics of platform labour. People may remain attached to promises of "flexible work," "self-employment" or a "good life" even as these undermine the conditions of a stable existence. The precarious present is therefore constituted not only by wages and contracts, but also by temporal experience, hope and everyday adaptation. Labour reform that ignores these attachments will struggle to explain the enduring appeal of precarious forms.]
- Terranova, Tiziana. "Free Labor: Producing Culture for the Digital Economy" Social Text 18, no. 2 (2000): 33–58.
[Note: Terranova's free labor examines the relations between voluntary participation, social interaction and capitalist valorization in digital culture. It warns against reducing digital labour to "paid online work": users' creation, socializing and data production can themselves become conditions of capital extraction. Algorithmic management does not mean managers disappear. Rule designers, customer service, operations and risk teams, and capital owners remain; some commands are simply executed through interfaces, ratings and automatic deactivation. Tracing responsibility must therefore penetrate the organizational claim that "the system decided." "Free" participation, paid employment and microtask labour must remain analytically distinct, or digital labour loses its capacity to distinguish degrees of coercion. Algorithmic management has not abolished management relations; it converts some commands into task allocation, ratings, dynamic pricing and automatic deactivation. An impersonal interface can disperse responsibility without reducing workers' exposure to control. The strongest aspect of free labor is its recognition that voluntary participation and capitalist valorization can occur simultaneously. Enjoying creation, socializing or games does not stop users from producing value; conversely, calling every online activity labour erases differences in wage relations and coercion. Terranova thus offers an analytical continuum, not a simple formula that turns all of life into labour. She also reveals that capitalist valorization and cultural autonomy are not a zero-sum relation. Digital culture can genuinely create communities, aesthetics and knowledge while platforms convert these activities into traffic and data. Critique should not deny participants' autonomous meanings, but explain how value is captured within infrastructures they do not fully control. The force of free labor lies precisely in this duality: participation can be a real autonomous practice while supplying unpaid conditions for platform accumulation.]
- Scholz, Trebor. Uberworked and Underpaid: How Workers Are Disrupting the Digital Economy. Polity Press, 2016. (Read: Chapter 4, "Crowd Fleecing"; Conclusion, especially the sections on "Tactical Refusal, Defection, and Withdrawal from Data Labor" and platform cooperativism).
[Note: The book's central concern is not "developing a workers' Uber," but reopening the questions of who bears platform-economy risks, who controls data and who owns platforms. Chapter 4 discusses crowd fleecing; the conclusion connects refusal, defection and withdrawal with platform cooperativism. This reading avoids reducing cooperativism to a business venture, restoring its place within worker ownership, collective bargaining and infrastructure governance. Discussions of automation must especially avoid substitution fetish: a machine replacing a task does not mean the disappearance of a profession or all labour. The institutional core of platform cooperativism is not "copy an existing platform and change its owner," but the reorganization of control, surplus distribution and data governance. Cooperatives that still depend on external cloud services, capital finance or closed algorithms may find their autonomy constrained by infrastructure. Scholz's questions thus ultimately concern governance capacity and technological sovereignty beyond ownership. Platform cooperativism must also confront scale. Cooperative ownership can alter returns and decision-making rights, yet network effects, finance, software maintenance and user acquisition can still force cooperatives to depend on incumbent giants. The viability of alternatives depends on a wider institutional ecosystem of shared technology, public procurement and alliances among cooperatives. The difficulties of cooperativism extend from business ownership to the technology stack, financing and market access: platform autonomy needs a fuller institutional ecology. Only when worker ownership combines with infrastructural capacity can it produce an organizational form meaningfully different from conventional platforms. The technical conditions of cooperation cannot be treated as neutral external resources.]
- Woodcock, Jamie. The Fight against Platform Capitalism: An Inquiry into the Global Struggles of the Gig Economy. University of Westminster Press, 2021. (Read: Chapter 3, pp. 45–67).
[Note: Woodcock returns collective action in platform labour to workers' practices: stoppages, disconnecting, organizing, unions and cross-platform cooperation. This creates a tension with Scholz's route of institutional alternatives: even where cooperative platforms exist, workers still need real organizational capacity and bargaining power. Cooperativism is no technological panacea. Read within this session, Woodcock's The Fight against Platform Capitalism: An Inquiry into the Global Struggles of the Gig Economy gives a more concrete structure to the question of which layers of power cooperatives, unions, labour law and commons governance each change. Algorithmic management has not abolished management relations; it turns some commands into task allocation, ratings, dynamic pricing and automatic deactivation. The interface's impersonality can disperse responsibility without weakening control over workers. Woodcock restores platform conflicts to workers' organizing, beyond alternative business models. Strikes, collective log-offs, unions and cross-platform networks show that collective action has not disappeared even when labour relations are individualized and broken into tasks. Algorithmic management may itself create shared objects of discontent. Platform labour politics thus remains centred on organizing: how can scattered experiences become sustainable collective power? Organization must also confront the technical mediation of the labour process. Personalized task allocation and dynamic pricing can weaken shared experience, yet the same interface can help workers recognize similar injustices and develop new organizing tools. Technology is both a management medium and a potential object and resource of collective action. These practices show that platform workers are not passive objects of algorithms: they develop knowledge, strategies and collective countermeasures around algorithmic rules.]
Recommended Readings:
- Srnicek, Nick. Platform Capitalism. Polity, 2017. (Read: Chapter 2, pp. 35–63).
[Note: Platforms are understood as economic organizations for data extraction and market intermediation. This gives the session's problem of compute concentration a prehistory in capitalist economies of scale and network effects, but platform capitalism must not simply be equated with AI capitalism. Theoretical resemblance does not automatically imply institutional identity. Srnicek treats platforms as specific business models formed through data extraction, network effects and market intermediation, supplying an economic basis for understanding their concentration. Platforms are not homogeneous: advertising, cloud, industrial and lean platforms depend on different revenue structures and asset configurations. Distinguishing these types prevents conceptual overgeneralization when examining platform labour and concentrated computing power. This typology also explains why "platforms" cannot be treated as one uniform class relation: their reliance on fixed assets, labour contracts and data differs, as do their paths to concentration. Another strength of platform capitalism is its placement of data within business competition rather than an abstract "information society." Firms accumulate data because it improves services, locks in users, trains models and erects entry barriers. Concentrated computing power and data must therefore be connected to concrete profit models, not explained only by technological scale. Platform concentration arises from network effects and positive feedback among data, capital and infrastructure; every alternative policy must specify the level at which it acts. Platform typology is thus a prerequisite for policy analysis: different mechanisms of concentration require different regulations and public alternatives. Economic distinctions precede any uniform scheme of "platform governance." Srnicek's typology also reveals differing capital intensities. Asset-light platforms depend on outsourcing and risk transfer; cloud platforms need enormous fixed capital. Both are called platforms, yet their labour organization and entry barriers differ entirely. Connecting platform capitalism with AI infrastructure especially requires distinguishing these capital structures, or "platform" will conceal the compute industry's heavy dependence on assets.]
- Huws, Ursula. Labor in the Global Digital Economy: The Cybertariat Comes of Age. Monthly Review Press, 2014. (Read: Chapter 1, pp. 11–38).
[Note: Huws approaches digital labour through global divisions of labour and commodification. She reminds us that digital labour encompasses not just software engineers, but a new global labour structure formed by services, outsourcing, platforms and knowledge work. Algorithmic management does not mean that managers disappear. Labor in the Global Digital Economy: The Cybertariat Comes of Age offers not a definitive label for the relation between affective attachment to precarious life and labour institutions, but a theoretical opening that competing explanations can test; this openness is itself part of its research value. Huws situates digital labour within global commodification and divisions of labour, rejecting "knowledge workers" as the digital economy's sole subjects. Customer service, outsourcing, logistics, manufacturing and office work are all restandardized and organized at a distance through informatization. The cybertariat thus connects high and low skills, centres and peripheries, requiring digital class structures to be understood through global labour chains. A global view also corrects the equation of digital labour with remote white-collar work. High-end software and low-paid back-office services often occupy the same value chain; the former's appearance of automation may rest on the latter's standardized labour. Class analysis must address these spatially separated but functionally interdependent jobs together. It must attend to both software interfaces and supply chains, restoring visible knowledge work and hidden support work to one structure. Huws also brings households and social reproduction into the picture. Remote and flexible work does not eliminate care burdens and may allow working time to penetrate further into the home. Platforms' flexible schedules must therefore be examined with gendered reproductive responsibilities: the same "flexibility" may mean autonomy for some workers and the continual erosion of work-life boundaries for others.]
- Casilli, Antonio A. Waiting for Robots: The Hired Hands of Automation. Foreword by Sarah T. Roberts. Translated by Saskia Brown. University of Chicago Press, 2025. Originally published in French as En attendant les robots: Enquête sur le travail du clic in 2019. (Read: Introduction; selected chapters on micro-work, clickwork, data labor, and automation).
[Note: The central point is that labour remains behind automation. Microwork, data annotation and content moderation show that AI training and operation often rely on highly dispersed, barely visible human labour. This connects "platform labour" with "AI labour." The 2025 English translation is used to bring the 2019 French study directly into dialogue with automation discourse after the rise of generative AI; its core remains hidden labour, and the book should not be misrepresented as a retrospective commentary on large models. Casilli's Waiting for Robots: The Hired Hands of Automation brings the concealment of microtasks, annotation and maintenance by automation narratives into organizational and institutional analysis. His principal contribution is to decompose "automation" into the organization of extensive human microlabour. Clicking, annotating, verifying and moderating are not incidental patches outside machine intelligence; they are often necessary conditions of model training and service operation. Automation is thus not the simple disappearance of labour, but its rearrangement across space, contracts and visibility. Casilli further shows how human labour is embedded in automated feedback loops. Models continually need annotation, verification, correction and content judgments; platforms distribute these stages through crowdsourcing to global labour markets. Human-in-the-loop may be a lasting organizational form of large-scale AI services, not merely a transitional stage. Automation grows together with new labour markets. The global distribution of microlabour also complicates labour law's identification of employers. Workers may face multiple contractors, crowdsourcing platforms and the ultimate AI company, with a fragmented chain of responsibility. Automation ethics must therefore reconnect supply-chain responsibility with labour standards.]
- Gray, Mary L., and Siddharth Suri. Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass. Houghton Mifflin Harcourt, 2019. (Read: Chapters 1–3).
[Note: This book shows the microtasks, content moderation and platform-mediated human labour behind AI. Read with Casilli and Crawford, it reframes automation as labour redistribution and the production of labour's invisibility. Cooperativism is not a technological cure-all either. The question of which layer of power cooperatives, unions, labour law and commons governance each change, illuminated by Gray's Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass, exceeds conceptual debate within the text. Ghost work exposes the infrastructural paradox of automation narratives: the more a system is described as requiring no human input, the more easily the human judgment sustaining its "intelligence" disappears. Global microtask markets split labour into discrete, on-demand units that are hard to organize, grounding technical efficiency in new forms of labour precarity. AI labour policy must therefore extend beyond technical occupations: microtask workers' wages, occupational safety, complaint channels and cross-border rights also belong to AI governance. Gray and Suri reveal an organizational paradox as well: workers are isolated into individual accounts yet rely on informal communities to exchange experiences, identify scams and help one another. Where platforms provide no stable work organization, workers create their own cooperative networks. The collectivity of digital labour has not vanished; it has moved from formal workplaces into more fragile spaces with less institutional recognition. This organization also shows that "on demand" does not mean demand arises naturally. Platforms actively shape labour markets through task decomposition, pricing and quality control. Worker replaceability is jointly manufactured by technical and contractual design, creating a direct institutional connection between labour's invisibility and technological myth.]
- Rosenblat, Alex. Uberland: How Algorithms Are Rewriting the Rules of Work. University of California Press, 2018. (Read: Chapters 2–4).
[Note: Rosenblat examines algorithmic dispatch, prices, ratings and information asymmetry through drivers' experiences. Her concrete ethnography of algorithmic management prevents platform labour from remaining an abstract capital-labour relation. At the scale of Uberland: How Algorithms Are Rewriting the Rules of Work, platform labour's task decomposition, algorithmic management, risk transfer and collective action first appear as concrete relations. Automation is often achieved by hiding labour: annotation, moderation, microtasks and maintenance are distributed across global labour markets, so the claim that "machines perform the tasks" conceals new organization and risk transfer. Uber's algorithmic management does not eliminate management, but distributes commands across price signals, task allocation, ratings and information asymmetry. Drivers formally retain freedom of choice while continually responding to a platform-designed incentive environment. Rosenblat's ethnography therefore specifies how autonomy and control coexist within one labour regime. It also reveals information asymmetry as a managerial instrument. Platforms see demand, prices and driver performance together; drivers see only partial information on which to base decisions. Algorithmic management draws power not only from automated commands, but also from monopolizing a market-wide view and controlling when information is disclosed. This is a key difference from traditional on-site supervision: workers face a platform that can observe them but that they cannot easily observe. Such an information structure makes "autonomous labour" distinctly asymmetric: choice exists, yet rests on the platform's superior informational position.]
- Aloisi, Antonio, and Valerio De Stefano. Your Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labour. Bloomsbury, 2022. (Read: chapters on algorithmic management and worker protection).
[Note: This book approaches algorithmic management through labour law and institutional design. The "boss" need not be one person: it can take the form of scheduling models, performance ratings, automatic deactivation and rule systems, linking political economy with labour institutions. Aloisi's Your Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labour, Rosenblat's Uberland: How Algorithms Are Rewriting the Rules of Work, and Irani's Chasing Innovation: Making Entrepreneurial Citizens in Modern India place the relation between affective attachment to precarious life and labour institutions within different intellectual traditions. Algorithmic management does not mean managers disappear. Aloisi's Your Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labour releases this relation from overly abstract judgments about the age. Automation is often accomplished by hiding labour. The contribution of Your Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labour to this relation lies precisely in keeping theory open to refutation. Aloisi and De Stefano approach algorithmic management through labour law, restoring institutional content to the popular phrase "your boss is an algorithm." Task allocation, performance management, pricing and deactivation may be automated, but businesses still design control within labour relations and bear its legal consequences. The central question is how traditional employment classifications, rights to information and collective negotiation can adapt to impersonal management, not how to hand responsibility over to software.]
- Irani, Lilly. Chasing Innovation: Making Entrepreneurial Citizens in Modern India. Princeton University Press, 2019. (Read: chapters on entrepreneurial citizenship, design and development).
[Note: Irani studies how "innovation" becomes a moral language in governance and economic policy, and how entrepreneurial subjects are formed. Read with Berlant, she invites a question: when workers must become "innovators," "entrepreneurs" and self-managers, does power enter labour relations through the language of hope and self-realization? Through Chasing Innovation: Making Entrepreneurial Citizens in Modern India, the concealment of microtasks, annotation and maintenance by automation narratives can be broken into interconnected processes. Automation often works by hiding labour. From the perspective of Chasing Innovation: Making Entrepreneurial Citizens in Modern India, this concealment cannot be separated from institutional authorization and resource allocation: technical forms produce enduring social consequences only in conjunction with organizational power. Irani examines how innovation turns from an economic slogan into a mechanism of subject formation. Design thinking, entrepreneurship education and development projects encourage individuals to redescribe social problems as entrepreneurial opportunities, while turning structural risks into personal responsibilities. Alongside platform labour, entrepreneurial citizenship explains why "precarity" can be repackaged as autonomy, flexibility and innovation. The "entrepreneurial citizen" also changes how failure is attributed. Once innovation becomes an individual duty, project failure is more readily explained as personal inadequacy than as an outcome of infrastructure, capital and policy. Self-branding, ratings and freelance culture in the platform economy continue this subjectification, recasting labour risks as personal choices.]
- Schor, Juliet B. After the Gig: How the Sharing Economy Got Hijacked and How to Win It Back. University of California Press, 2020. (Read: Introduction; selected chapters on platform business models, inequality, and alternatives).
[Note: Through empirical research on sharing-economy platforms, Schor distinguishes business models, forms of labour participation and distributive outcomes, avoiding the reduction of every gig-economy experience to one model of exploitation. She is especially useful after Scholz and Woodcock: cooperativism and workers' struggles matter, but workers differ in their dependency, income structures and motivations, so alternatives cannot rely on one organizational template. Schor's After the Gig: How the Sharing Economy Got Hijacked and How to Win It Back, Irani's Chasing Innovation: Making Entrepreneurial Citizens in Modern India, and van Doorn's "Platform Labor: On the Gendered and Racialized Exploitation of Low-Income Service Work in the 'On-Demand' Economy" shift the question of which layers of power cooperatives, unions, labour law and commons governance change from broad judgments to mechanisms. Cooperativism is no technological panacea. The contribution of After the Gig: How the Sharing Economy Got Hijacked and How to Win It Back is precisely to return this question to identifiable institutional mechanisms. Automation often works by hiding labour. Schor distinguishes the sharing economy's early social visions from platforms' later commercialization, showing how venture capital and expansion reorganize narratives of efficiency, community and supplementary income. Her research indicates that the platform economy has no single path: labour conditions, ownership and community rules can produce different outcomes. The question becomes which institutional arrangements can turn sharing from marketing language into actual resource governance.]
- van Doorn, Niels. "Platform Labor: On the Gendered and Racialized Exploitation of Low-Income Service Work in the 'On-Demand' Economy." Information, Communication & Society 20, no. 6 (2017): 898–914. (Read: entire article).
[Note: Van Doorn emphasizes that on-demand service work is not abstract "flexible labour," but highly gendered, racialized labour embedded in urban service relations. Connecting platform capitalism with existing low-paid services, migrant labour and social reproduction corrects analyses that begin only with interfaces or algorithmic dispatch. "Platform Labor: On the Gendered and Racialized Exploitation of Low-Income Service Work in the 'On-Demand' Economy" clarifies the causal levels of task decomposition, algorithmic management, risk transfer and collective action. By putting gendered, racialized, low-income services at the centre, van Doorn challenges the tendency to let drivers or highly skilled freelancers represent the entire gig economy. Cleaning, care and delivery already occupy unequal structures of social reproduction and migrant labour; platformization further digitizes ratings, matching and payment. Algorithms did not create these differences, but can recode and amplify them. Gender and race in the on-demand economy also show that technical mediation does not erase labour markets' historical stratification. Customer ratings, geographical matching and platform classifications can translate existing prejudices into apparently individualized transaction outcomes. Labour fairness therefore concerns algorithm design as well as institutions beyond platforms, such as migration status, care responsibilities and social protection. Platformization can shift prejudice from employers' direct judgments to dispersed customer reviews and system rankings, making responsibility for discrimination harder to locate. Algorithmization does not strip inequality of identity; it can instead sustain inequality through apparently individualized data feedback.]
- Graham, Mark, Isis Hjorth, and Vili Lehdonvirta. "Digital Labour and Development: Impacts of Global Digital Labour Platforms and the Gig Economy on Worker Livelihoods" Transfer 23, no. 2 (2017): 135–162. (Read: entire article).
[Note: Graham, Hjorth and Lehdonvirta situate digital labour platforms within global development and cross-border labour markets. They examine how online outsourcing expands opportunities while generating unequal bargaining, excess supply and geographical wage differences. This extends the session from Uber-style local gigs to global remote labour, reminding us of platform labour's different spatial scales. Algorithmic management does not mean managers disappear. Their global perspective examines online outsourcing's effects on development and geographical inequality. The internet lowers cross-border transaction costs without eliminating wage disparities, language hierarchies or dependence on place. Platforms can globalize demand while leaving workers in separate social-protection systems: digital labour both crosses and exploits borders. This market also produces new replaceability. Employers can quickly move tasks between countries, while workers remain constrained by local living costs, currencies and regulations. Remote platforms create asymmetric spatial freedom: capital and tasks move more readily, while workers' social reproduction remains intensely local. The spatial politics of cross-border digital labour is thus not the "disappearance of geography," but the conversion of geographical differences into cost advantages and bargaining inequalities. Its developmental effects are double-edged: it can provide new income in peripheral regions while exploiting low wages and weak social protection to sustain global cost disparities. Developmental value must be assessed not only by job numbers, but also by bargaining power, skills accumulation, income stability and opportunities to move up the value chain. Moreover, platform ratings and global competition individualize development: workers must improve their position through skills and reputation, while exchange rates, regional wages and platform bargaining power recede into the background. Digital development policy must restore these structural factors to the picture.]
- Karatani, Kojin. The Structure of World History: From Modes of Production to Modes of Exchange. Translated by Michael K. Bourdaghs. Duke University Press, 2014. Originally published 2010. (Read: Chapter 12, "Toward a World Republic," especially the discussion of Mode D, pp. 285–308).
[Note: Karatani's Mode D is not a digital-currency proposal, but a rethinking of exchange within historical dynamics. Moving beyond capitalism understood solely as a mode of production, he places capital, the state and reciprocity within a longer history of exchange. Here, this confronts Web3, DAOs and platform cooperativism with a higher-order institutional question: if the mode of exchange remains unchanged, can changing the technical interface alone transform capital-labour relations? Through The Structure of World History: From Modes of Production to Modes of Exchange, the concealment of microtasks, annotation and maintenance in automation narratives can move from rhetoric to institutional analysis. Mode D's normative force lies in refusing to reduce alternatives to capitalism to a single change of ownership. Exchange, state and reciprocity are interwoven; every digital cooperative proposal must address public finance, social reproduction and coordination across regions. Karatani thus confronts "decentralized organization" with a world-historical scale beyond technical governance. His framework also makes the digital left rethink the state. Capital, state and nation are not links that can be dismantled in isolation: market expansion often works together with state redistribution and protection. Proposals to replace existing organizations with DAOs or cooperative platforms cannot easily assume the overall functions of social reproduction without addressing taxation, public services and coordination between communities. Mode D therefore cannot be reduced to a technological utopia. The real difficulty is establishing institutions that transcend the complementary structure of capital and state while providing public goods, redistribution and cross-regional coordination. Technology can only be one component.]
- Ostrom, Elinor. Governing the Commons. Cambridge University Press, 1990. (Read: Chapter 1, pp. 1–28).
[Note: Ostrom's empirical research challenges the reduction of common-resource governance to the "tragedy of the commons." Read alongside DAOs, the questions are how rules arise, are enforced and revised, and which participant structures can sustain shared resources. Code can implement rules, but cannot automatically create trust, participation or responsibility. Cooperativism is no technical panacea. Rather than maximizing conceptual coverage, theory should remain refutable: this framework should extend only where evidence supports its core mechanisms. Governing the Commons ultimately makes the question of which layers of power cooperatives, unions, labour law and commons governance change open to further empirical analysis, instead of a completed diagnosis of the age. Ostrom is especially useful for correcting the idea that commons mean rule-free sharing. Durable governance depends on boundaries, participatory rule-making, monitoring, graduated sanctions and conflict resolution, adapted to each community. Although digital commons have different copying costs from natural resources, open access alone cannot solve their governance problems. Ostrom's design principles cannot be mechanically transferred to digital resources. Software and data can be copied non-rivalrously, and membership may span the world; servers, maintenance and governance attention nevertheless remain scarce. What can travel is her empiricism about institutional diversity and self-organization, not a licence to call every open project a commons. The issue thus returns to institutional design: reproducibility does not remove the need for member governance, maintenance obligations and conflict resolution. Digital reproducibility reduces resource-depletion problems but heightens the scarcity of maintenance and governing attention. Code review, server costs, community management and conflict handling all require continuing investment. Ostrom reminds us that commons are never only about openness, but about durable governance suited to the character of the resource.]
Part IV: General Intellect and Technological Future (Weeks 7–8)
Session 7: General Intellect and Resistance to Compute Concentration
This session reconnects Marx's general intellect with contemporary AI infrastructure, without claiming that "Marx predicted large models." The "Fragment on Machines" in the Grundrisse addresses contradictions among social knowledge, fixed capital, labour time and the value-form: as scientific knowledge and social cooperation enter machinery more deeply, the relation between direct labour and productive power changes. Virno re-emphasizes general intellect as a living, public capacity for language and cognition, distinct from the stock of knowledge already congealed in machines. Their tension offers more productive questions about concentrated computing power than whether "AI resembles humans."
Generative AI converts socially produced text, code, images, feedback and judgment into training resources on a vast scale. At the same time, chips, data centres, energy, capital investment and specialized labour make it a form of fixed capital with high entry barriers. Pasquinelli's social history of AI, Crawford's tracing of its material and labour conditions, and studies of knowledge capitalism, cognitive capitalism and digital labour jointly show that "intelligence" is not a pure capacity detached from social relations, but an outcome of knowledge, labour, infrastructure and property arrangements together. Compute concentration is therefore first an industrial and institutional problem open to empirical analysis, not an a priori political conclusion.
Resistance cannot be reduced to "open source" or "a GPU for everyone." Open weights may improve access without altering training resources, cloud infrastructure, energy costs or governing power. Public computing resources can lower entry barriers but still face questions of data, maintenance, responsibility and distribution. Virno's exit, commons and non-state public sphere offer no ready-made solution, but a political direction for reorganizing general intellect: can shared cognitive capacities acquire institutions and infrastructure of their own, so that models, data and computation no longer exist solely as private fixed capital?
Required Readings:
- Marx, Karl. Grundrisse: Foundations of the Critique of Political Economy. Translated by Martin Nicolaus. Penguin Books, 1973. Originally written 1857–58; first published 1939–1941. (Read: "Fragment on Machines," pp. 690–712).
[Note: We must resist the retrospective claim that "Marx predicted AI." Marx addresses historical contradictions among social knowledge, general intellect, fixed capital, labour time and value: as scientific knowledge and social cooperation become more embedded in machinery, labour time's place in production changes. AI can be treated as a new historical setting for this problem, but LLMs cannot simply be equated with Marx's "system of machinery." The question is whether contradictions among value, labour and ownership reappear in new forms as social intelligence increasingly congeals into privately owned computing infrastructure. The "Fragment on Machines" must be restored to Marx's analysis of fixed capital and productive forces. Machinery absorbs scientific knowledge and social cooperation, increasingly making workers supervisors and regulators of production; this creates a tension between labour time as the measure of wealth and the socialization of productive power. General intellect's place in Marx's text is not identical to its later autonomist interpretation, and this difference must be preserved. For AI, the most useful comparison concerns how knowledge enters fixed capital: do models, software, chips and data centres form a new combination of fixed capital, and which parts of training data and living cooperation resist fixation? This is first a theoretical question about social knowledge, common capacities and relations of production. Contemporary applications must explain how these capacities become fixed capital, rather than anthropomorphizing model parameters as humanity's shared mind. Marx's fragment addresses the contradictions of fixed capital, social knowledge and labour time, rather than predicting AI. General intellect's entry into machinery means that production increasingly relies on socially accumulated science and cooperation, even as the value-form retains labour time as its measure. Foundation models can provide a new setting for this contradiction, but models, data centres and GPUs are particular historical forms; they cannot retrospectively define Marx's concepts.]
- Virno, Paolo. A Grammar of the Multitude: For an Analysis of Contemporary Forms of Life. Translated by Isabella Bertoletti, James Cascaito, and Andrea Casson. Semiotext(e), 2004. Originally published 2001. (Read: Day One, "Common Places and 'General Intellect'"; Day Two, "Intellect as Score" and "Reason of State and Exit"; Day Three, "The Emotional Tonalities of the Multitude").
[Note: Virno's crucial development of Marx's general intellect is his refusal to understand it merely as the "total stock of knowledge" congealed in machines, fixed capital and technical systems. In "Common Places and 'General Intellect'" and "Intellect as Score," it first appears as socially shared capacities for language, knowledge, judgment and cooperation: a general condition of living labour. The central issue for AI is then not whether machines have acquired human intelligence, but how existing social cognitive capacities are reorganized through platforms, models and infrastructure. "Reason of State and Exit" advances the political question: if capital absorbs general intellect, why does emancipation not follow automatically? Virno's exit asks us to find a politics whose sole aim is not taking over existing state-corporate organizations, but reorganizing public conditions of cooperation. Today, resisting compute concentration therefore cannot mean simply replicating big tech in a more democratic form. The real question is whether society can regain data, model, computing and knowledge infrastructures, making computation serve general intellect rather than turning general intellect into raw material for compute capital. "Compute multitude" must remain a problematic concept rather than being attributed directly to Virno: it tests whether new political connections can form among general intellect, infrastructural ownership and autonomous organization. The metaphor "Intellect as Score" is especially important: a score is not an accomplished performance, but a set of public capacities realized only in concrete action. General intellect is thus not merely an object stored in fixed capital, but the common condition of speaking, cooperating, learning and judging. "Reason of State and Exit" then turns to the relation between the state and non-state publicness. Exit is neither private escape nor a technical community's flight from regulation, but a reorganization of cooperation that withdraws support from old relations. Applied to open models, public computing or federated infrastructures, each proposal must be tested: does it really change power over the organization of common capacities, or merely make tools cheaper? Virno brings general intellect out of fixed capital and back to living language, judgment, habits and cooperation. It can be organized by capital as a productive force, yet also holds possibilities for non-state publicness. Exit is not simply leaving institutions, but reorganizing the conditions of cooperation. For contemporary computing, the political focus is not whether machines possess intelligence, but whether common cognitive capacities can establish their own data, model and computing infrastructures.]
- Pasquinelli, Matteo. The Eye of the Master: A Social History of Artificial Intelligence. Verso, 2023. (Read: Chapter 4, "The Origins of Marx's General Intellect"; Chapter 7, "The Automation of Pattern Recognition"; Conclusion, "The Automation of General Intelligence").
[Note: Pasquinelli places AI back within histories of the division of labour and pattern recognition, rather than treating it as a computational miracle detached from society. Chapter 4 retraces Marx's general intellect; Chapter 7 examines pattern recognition and automation; the conclusion carries these histories into the automation of general intelligence. For this session, "model training" becomes a question of organizing labour and knowledge rather than a purely technical operation: does AI automate one person's skill, an occupation, or collective cognitive capacities already formed by society? The book's social history asks us to understand intelligence alongside the division of labour: machine intelligence often extracts rules, patterns and statistical order from human work, then reorganizes that work in automated form. Chapter 4 returns to general intellect not to announce that Marx already explained AI, but to ask how knowledge becomes a productive apparatus. Chapter 7's historicization of pattern recognition dismantles the myth of deep learning "learning by itself from data." General intelligence in the conclusion is therefore first a social and institutional question: which general capacities are formalized, who determines the patterns, and who must adapt to patterns recognized by machines? Reconstructing AI through labour division and pattern recognition means machine learning is no longer intelligence suddenly emerging from pure mathematics. Automation often first extracts and formalizes rules of judgment from social practice, then delegates their execution to machines. General intelligence's history connects to histories of organization, measurement and management. What AI "automates" is not an abstract mind, but already socialized cognitive operations.]
- Hardt, Michael, and Antonio Negri. Multitude: War and Democracy in the Age of Empire. Penguin Press, 2004. (Read: Part 2, selected sections on immaterial labor, biopolitical production, and the common).
[Note: Hardt and Negri connect multitude, immaterial labour, biopolitical production and the common. As production increasingly depends on language, knowledge, affects and cooperation, they argue that political subjectivity need not take the homogeneous industrial worker as its sole premise. After Marx and Virno, they offer a more explicit vision of political organization, but one that must face empirical correction from platform-labour research: widespread social cooperation does not automatically produce autonomous subjects, and capital can still capture cooperation through property, platforms and infrastructure. In the context of this session, Multitude: War and Democracy in the Age of Empire makes the difficulties of moving from theoretical multitude/common to sustainable institutions more concrete. General intellect cannot simply be equated with model parameters, datasets or the internet. It first names socialized capacities for language, knowledge, judgment and cooperation; fixed capital can absorb these capacities without making them fully identical with machines. Hardt and Negri emphasize the conjunction of common productive capacities and differentiated subjects, seeking a politics distinct from a homogeneous people or the traditional working class. This openness connects well to digital cooperation, but is easily romanticized through network metaphors. The real question is how dispersed cooperation acquires institutional durability, resources and decision-making structures; otherwise, "multitude" may remain political rhetoric that describes connections. Its affinity with digital networks readily invites technological optimism. Networked common production does not automatically generate common politics, nor do different subjects naturally form a stable will. The difficult task is building institutions for decision-making and resources without erasing difference. This brings multitude theory directly to governance.]
Recommended Readings:
- Vercellone, Carlo. "From Formal Subsumption to General Intellect" Historical Materialism 15, no. 1 (2007): 13–36.
[Note: Vercellone examines the entry of knowledge, cooperation and social intelligence into capital accumulation. He bridges Sessions 6 and 7: questions of labour organization ultimately lead to who owns general intellect and knowledge commons. Theoretical resemblance does not automatically imply institutional identity. Compute concentration concerns industrial structure and infrastructure. The scale requirements of chips, energy, capital and cloud services shape who can train, deploy and maintain models, so open weights cannot by themselves remove dependence on private infrastructure. The account of general intellect, fixed capital, cognitive labour and concentrated computing power in "From Formal Subsumption to General Intellect" is therefore valid only where the relevant institutional mechanisms can be identified empirically. Vercellone connects general intellect with capital's absorption of social knowledge, emphasizing production's increasing dependence on cooperation and knowledge beyond factory walls. Cognitive capitalism places rent, intellectual property and socialized production within a shared contradiction. For AI, it offers not the conclusion that knowledge has replaced labour, but an analysis of how capital re-establishes scarcity and control over common knowledge that is difficult to enclose completely. It also returns rent to analysis of the knowledge economy. Where common knowledge resists full privatization, capital may increasingly earn returns through patents, platform gateways and infrastructure ownership. AI models' enclosure of knowledge and API access can be examined this way, but whether they constitute rent requires concrete analysis of competition, property and pricing. By connecting formal/real subsumption with general intellect, Vercellone also tries to explain capital's movement from directly organizing labour toward controlling conditions of socialized knowledge. Even if this periodization is contested, it asks an important question: when production depends on broad social cooperation, which institutions convert dispersed cooperation back into private returns? Platforms and AI infrastructure supply new empirical objects for this question.]
- Moulier Boutang, Yann. Cognitive Capitalism. Translated by Ed Emery. Polity, 2011. Originally published in French as Le capitalisme cognitif in 2007. (Read: Introduction; selected chapters on knowledge, externalities, and cognitive labor).
[Note: Moulier Boutang's cognitive capitalism places knowledge, innovation, cooperative networks and positive externalities at the centre of value production, explaining why production increasingly exceeds traditional factory boundaries. It is close to general intellect but not synonymous with it: Marx and Virno address social intelligence's relation to capital, whereas cognitive capitalism is an empirical-theoretical name for a particular capitalist phase. Reading them together tests what it means for knowledge to become the main productive force. Compute concentration is a matter of industry and infrastructure. Cognitive Capitalism offers not a final designation for the reorganization of common capacities through property, platforms and infrastructure, but a theoretical opening testable by competing explanations; that openness is itself valuable. Moulier Boutang describes knowledge, innovation and positive externalities as prominent in accumulation, incorporating cooperation beyond the factory into value analysis. "Cognitive" must not obscure material production, but it helps explain why networks, patents, education and knowledge flows become crucial to capitalist competition. Applied to AI, this requires attending both to models' cognitive resources and to the energy, hardware and labour supporting them. Cognitive capitalism also confronts labour measurement. Knowledge production relies heavily on cooperation, education and social interaction, making individual contributions harder to isolate; capital nevertheless distributes income through wages, intellectual property and firm boundaries. Model training amplifies this contradiction: its materials come from extensive social production, while returns concentrate among a few infrastructure owners.]
- Gorz, André. The Immaterial: Knowledge, Value and Capital. Translated by Chris Turner. Seagull Books, 2010. Originally published in French as L'Immatériel in 2003. (Read: selected chapters on knowledge, value, and the crisis of measurability).
[Note: Gorz examines a contradiction: knowledge, capacities and living social experience resist complete measurement by conventional labour time, yet capital continually converts them into appropriable resources. He belongs in the general-intellect lineage because he restores knowledge from abstract information to subjective capacities and social relations while raising a crisis of value measurement. For AI, we must ask whether training commodifies knowledge itself, or turns its expressions, data traces and access rights into assets. The Immaterial: Knowledge, Value and Capital brings the distinction between openness, interoperability, public ownership and autonomous governance to organizational and institutional analysis. Compute concentration concerns industrial structure and infrastructure. Gorz locates cognitive capitalism's central contradiction between reproducible knowledge and the manufacture of scarcity. The easier digital knowledge is to copy, the more capital must recreate exclusion through intellectual property, platform enclosure and service provision. Foundation models and the API economy give this a new form: absorbed public knowledge re-enters the market through controlled access. By stressing knowledge's immeasurability, Gorz makes value impossible to explain fully through individual labour time. Knowledge often comes from long education, public research and social interaction, while copying costs approach zero; patents, closed platforms and brands must reconstruct scarcity. Generative AI's mass absorption of public text makes the contradiction between common knowledge and private access more visible. Gorz also helps explain model-access rights: capital need not own knowledge itself if control of computation, interfaces and service conditions makes access to common knowledge scarce again. Appropriation shifts from ownership of content to ownership of infrastructure.]
- Roggero, Gigi. The Production of Living Knowledge: The Crisis of the University and the Transformation of Labor in Europe and North America. Translated by Enda Brophy. Temple University Press, 2011. (Read: Introduction; selected chapters on living knowledge, the university, and cognitive labor).
[Note: Roggero examines the production of living knowledge through universities and knowledge work. Cognitive labour is not an abstract "creative class," but forms within educational institutions, labour markets, cooperative networks and conflict. This returns general intellect from grand philosophical claims to universities, research, training and intellectual-property regimes, making it useful for examining tensions between common knowledge production and institutional appropriation as generative AI enters education and research. The difficulty of moving from theoretical multitude/common to sustainable institution-building in The Production of Living Knowledge: The Crisis of the University and the Transformation of Labor in Europe and North America exceeds conceptual debate within the text. Compute concentration is an industrial and infrastructural problem. Roggero places living knowledge within university crisis and labour transformation, stressing that knowledge is continually generated in social cooperation, learning and networks, not only formal research institutions. Universities produce public knowledge while commodifying and quantifying it. Foundation models' mass absorption of academic and public texts intensifies the contradiction, creating new appropriative relations between open social knowledge and private computing infrastructure. For Roggero, the university is not a pure public space outside knowledge capitalism, but a laboratory of labour transformation: research, teaching and learning increasingly undergo performance measurement, project-based organization and competitive funding. Connected with AI, this perspective shows academic knowledge becoming open text for model ingestion while publishing, platforms and computing resources restratify it.]
- Beller, Jonathan. The Message Is Murder: Substrates of Computational Capital. Pluto Press, 2018. (Read: Introduction; selected chapters on computational capital, attention, and racialized extraction).
[Note: Beller places computation within histories of capital, visual culture, financialization and racialized violence. Computation is not confined to operations inside machines, but belongs to broader social extraction of value. This creates a necessary tension with general intellect: treating social knowledge only as an emancipatory common capacity risks overlooking its production within differentiated violence, attention capture and global divisions of labour. At the scale of The Message Is Murder: Substrates of Computational Capital, relations among general intellect, fixed capital, cognitive labour and compute concentration first appear as concrete connections. Beller understands computational capitalism as a system in which perception, attention and infrastructure jointly participate in valorization, extending digital labour's boundaries. His strong claims require empirical limits, yet identify a crucial issue: computation is not a neutral background tool, but may become a general substrate through which capital organizes vision, time and social relations. The concept of substrate connects visual culture, computation and material foundations. Screen-based attention becomes capitalizable only within systems of chips, networks, energy and labour. Computational capital is therefore neither a pure information economy nor a simple extension of industrial capital, but a form of value capture organized through multiple infrastructural layers. Computation's apparent immateriality is rematerialized: the valorization of attention, images and data always depends on equipment, networks and labour below. Its analysis needs a double view of the politics of perception and material infrastructure. Losing either side abstracts computational capital again.]
- Tronti, Mario. Workers and Capital. Translated by David Broder. Verso, 2019. Originally published 1966. (Read: "Lenin in England," pp. 78–88).
[Note: Tronti's significance is methodological: begin not with what capital wants, but with how workers' struggles force capital to change. For today's platform economy, this is a research method, not proof that "workers' struggles produced AI." Workers and Capital, Beller's The Message Is Murder: Substrates of Computational Capital, and Doctorow's The Internet Con: How to Seize the Means of Computation situate the reorganization of shared social capacities through property, platforms and infrastructure within different intellectual traditions. Workers and Capital releases this question from overly abstract judgments about the age. The contribution of Workers and Capital lies precisely in keeping the theory refutable. Tronti reminds us that technological change cannot be explained only through capital's innovation strategies. Workers' refusal, mobility, organization and conflicts in production can also force capital to find new management technologies and labour arrangements. Applied to automation, this means tracing historical links between concrete struggles and investment decisions, not romanticizing automation itself as a direct product of workers' struggles. His workerism also insists that political knowledge comes from inquiry within struggles, rather than defining workers' interests in advance. For platform and AI labour, research should begin with specific conflicts involving drivers, annotators, moderators and engineers, examining how workplaces reinterpret technology. Theory must not position workers merely as passive recipients of technical change.]
- Doctorow, Cory. The Internet Con: How to Seize the Means of Computation. Verso, 2023. (Read: Introduction; chapters on interoperability, switching costs, and antitrust).
[Note: Doctorow translates "seizing the means of computation" into concrete institutional tools—interoperability, adversarial compatibility, antitrust and exit rights—rather than an ownership slogan. This provides a valuable intermediate policy level: if platform infrastructure encloses general intellect, must resistance first reduce lock-in and restore interoperability and migration? We must also ask whether these reforms reach ownership of data centres, computation and capital, or mainly improve competition. The Internet Con: How to Seize the Means of Computation allows the distinction between openness, interoperability, public ownership and autonomous governance to be broken into interconnected elements. "Seizing the means of computation" restores competition policy, interoperability and user rights to the centre of infrastructure politics. Technical openness that leaves exit costs, data migration and market control unchanged often offers only superficial choice. This reformist path allows productive comparison with more radical public ownership or cooperative governance. Interoperability has a specific meaning for power: if users can take their data, social connections and content elsewhere and connect different services, lock-in decreases. Yet interoperability also brings complex costs involving security, privacy and standards governance. "Opening interfaces" is therefore no unconditional answer, but a mechanism of competition and autonomy requiring public rules. Its political significance is to shift competition from "more platforms" to whether users genuinely possess the institutional capacity to leave and migrate.]
- Tarnoff, Ben. Internet for the People: The Fight for Our Digital Future. Verso, 2022. (Read: Introduction; chapters on privatization, public ownership, and democratic control).
[Note: Tarnoff connects the history of internet privatization with public control, rejecting explanations that attribute platform power only to a few bad firms or algorithms. His approach is infrastructural politics: ownership, public investment, democratic governance and decommodification. With Doctorow, this opens a debate over reform's scale: are interoperability and competition policy enough, or must some internet layers be reconceived as public infrastructure? Internet for the People: The Fight for Our Digital Future, Doctorow's The Internet Con: How to Seize the Means of Computation, and Schneider's Governable Spaces: Democratic Design for Online Life shift the difficulty of moving from theoretical multitude/common to durable institutions from broad judgment to concrete mechanisms. The contribution of Internet for the People: The Fight for Our Digital Future is to return these broad claims to identifiable institutional mechanisms. By treating privatization as historical, Tarnoff breaks with the assumption that networks naturally belong to markets. Public funding, commercialization and advertising together shaped today's platforms; decommercialization must therefore involve financing and public institutions, not merely appeals to corporate ethics. The internet's public character becomes a question of ownership and institutional design.]
- Schneider, Nathan. Governable Spaces: Democratic Design for Online Life. University of California Press, 2024. (Read: Introduction; selected chapters on governance, ownership, and democratic design).
[Note: Schneider begins with online communities' governance practices. Digital spaces are neither naturally anarchic nor governable only by corporations; rules, membership, ownership and decision procedures can all be redesigned. There are distant resonances with Virno's exit and Kelty's recursive public, but the focus here is institution-building: without sustainable governance, "exit" from platform power can amount merely to moving to another private platform. Governable Spaces: Democratic Design for Online Life clarifies the causal levels connecting general intellect, fixed capital, cognitive labour and compute concentration. Schneider treats online spaces as capable of constitutional governance, asking how rules form, members gain representation and power is constrained, rather than praising democracy abstractly. Democratization requires enforceable procedures, not simply a voting feature added to an interface. Governable spaces require more than community rules: members must have an institutional position from which to change them. Platform democracy cannot be equated with surveys or voting plug-ins. If servers, code and final enforcement remain with outside owners, participation may lack binding force. The substantive test is not whether a participation button exists, but whether members can durably constrain rules, infrastructure and enforcement. Only when governing rights can bind infrastructure owners does digital participation become institutional power rather than consultation. Democratic design must ultimately reach enforcement.]
- Söderberg, Johan. Hacking Capitalism: The Free and Open Source Software Movement. Routledge, 2008. (Read: Introduction; selected chapters on free software, labor, and political economy).
[Note: Söderberg explains free/open-source software through labour and political economy, recognizing cooperation's potential beyond conventional firms while examining capital's recapture of unpaid contributions. This tests a recurring question: does open code mean common ownership? Where cloud computing, deployment channels and data remain concentrated, where is the political force of software commons weakened? Free software shows that control of technical means of production can be institutionalized through licences, cooperative norms and shared code. Yet it can both challenge property and be absorbed by capital as cheap innovation. Openness is not a pure space outside capital, but a continually contested production relation. The movement also demonstrates non-state institutional innovation: copyleft repurposes law to protect sharing and resist private re-enclosure. "Exiting capital" need not mean exiting law; autonomous practices often use existing law to build lasting boundaries. This directly informs data and model commons. Commons do not require abolishing institutional boundaries: appropriate legal design can support public production. Licences and governance for public models and open data may similarly prevent re-enclosure. Sustained openness is itself an institutional achievement. Free software also develops material and institutional dependencies. Projects need hosting, maintenance, funding and contributor governance; if these concentrate again in a few firms, open code can coexist with infrastructural dependence. Open models likewise require distinct examination of property relations governing weights, training data, computation and deployment platforms.]
- Lazzarato, Maurizio. "Immaterial Labor" In Radical Thought in Italy: A Potential Politics, edited by Paolo Virno and Michael Hardt, 133–147. University of Minnesota Press, 1996. Originally published in Italian in 1994. (Read: entire chapter).
[Note: Lazzarato's immaterial labour names the importance of information, cultural content, communication and subjectivity production under post-Fordism. Influential for general intellect, it requires a double reading today: it captures production's spread into social communication, yet "immaterial" can obscure chips, minerals, logistics, annotation and bodily labour. Reading it with Session 6 exposes this limit. "Immaterial Labor" can turn the distinction among openness, interoperability, public ownership and autonomous governance from rhetoric into institutional analysis. The concept captures language, affect, knowledge and cooperation entering value production, but its boundaries require care. Cognitive and communicative activities do not become detached from material labour; they depend on equipment manufacturing, logistics, energy and reproductive work. Reconnecting the two prevents "immaterial" from hiding the digital economy's material foundations. Historically, the concept links post-Fordism, communication and subject production, emphasizing labour's increasingly direct production of cultural content and social relations. Its difficulty is the potential concealment of equipment, logistics and bodies. It is better read as a change in labour's content and forms of control, not a declaration that material production has departed. Only then can it connect to digital labour's layered structure. Its greatest value lies not in the label, but in identifying value production's growing dependence on communication, signs and subjective relations. Digital capitalism concerns how these activities and material foundations are recombined, not which replaces the other.]
- Dyer-Witheford, Nick, Atle Mikkola Kjøsen, and James Steinhoff. Inhuman Power: Artificial Intelligence and the Future of Capitalism. Pluto Press, 2019. (Read: Introduction; selected chapters on AI, labor, and capital).
[Note: This book examines AI automation and capital through Marxist political economy, rejecting an exclusive focus on whether future machines will be conscious. It studies how machine learning reorganizes capital's powers of control as it enters production, management and labour markets. This reconnects general intellect with AI's actual political economy while setting empirical limits for the next session's philosophy and future imaginaries. Inhuman Power locates AI within capitalist automation's historical logic, rather than treating it as an independent technical subject. Machine learning expands capital's capacities for prediction, coordination and supervision while depending on existing divisions of labour and property structures. "Inhuman power" is therefore first a technical capacity organized through social relations. Inhuman Power also juxtaposes AI's autonomy with capital's: machine systems appear to gain agency, yet their goals, deployment and expansion remain shaped by accumulation. Technical autonomy and capital's autonomy are different concepts. Distinguishing them prevents corporate investment and management decisions from being anthropomorphized as "AI demanding acceleration." Machine autonomy becomes an organizational question again: expanding capabilities do not remove capital's decisions over deployment and returns. AI's political economy must distinguish technical capability growth from the social direction given by capital's organization. The capitalist framework of Inhuman Power also corrects the philosophical image of AI as an autonomous actor. Algorithms' goals do not naturally arise inside machines, but are supplied by investment, strategy, competition and management needs. Capabilities may develop complexities beyond individual designers' expectations, yet deployment still has identifiable institutional actors. System complexity cannot justify anthropomorphically transferring political responsibility.]
Session 8: Artificial Intelligence, Automation and Posthuman Futures
The final session places AI at the intersection of philosophy, engineering, political economy and future imaginaries. Turing's classic questions about machine intelligence, Russell's account of human-machine goals, Bostrom's superintelligence risk framework, and Brassier's, Land's and Negarestani's different approaches to reason, technology and acceleration are not the same kind of argument. They involve operational tests, engineering control, thought experiments, normative philosophy and metaphysical inference, and cannot substitute for one another under a single evidential standard. Nor should AGI be treated as an empirical object with an already stable definition and timetable. It must be differentiated into questions of capability, autonomy, generality, deployment conditions and institutional power.
Haraway, Hayles, Braidotti and Hui reconsider the boundaries of "the human" through posthumanism, embodiment, relationality and technological plurality. Posthumanism does not mean enhancing humans or uploading consciousness, nor does it necessarily imply technological optimism. It first challenges the traditional placement of reason, subjectivity and political standing within the individual human. In another register, Crawford, Bender/Hanna, Narayanan/Kapoor and Hao repeatedly return grand intelligence narratives to data, labour, energy, corporate organization, industrial competition and marketing, bringing the "future" back into present institutions.
The session therefore asks how narratives of the future themselves become political forces. Industry roadmaps, risk discourse, accelerationism and emancipatory visions of technology predefine what counts as feasible, necessary or inevitable. AI critique must neither reduce every capability to hype nor accept the industry's unified category of "intelligence" because some systems do possess significant capabilities. Actual performance, speculative risk, political-economic structures and normative futures require distinct arguments. Only clear connections among these levels can take discussions of automation and the posthuman beyond oscillation between technological worship and fear.
Required Readings:
- Crawford, Kate. Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press, 2021. (Read: Introduction; Chapter 2, "Labor"; Chapter 3, "Data").
[Note: Crawford returns general intellect to planetary materiality. AI depends not only on algorithms and models, but also on minerals, energy, data centres, annotation, moderation and global supply chains. Chapter 2 makes AI's labour visible again; Chapter 3 shows how data becomes a resource. With Marx and Virno, this raises a concrete question: if AI privatizes social intelligence while hiding material labour and ecological costs behind models, why does "intelligence" consistently appear disembodied? Crawford corrects our understanding of AI's actual conditions rather than directly extending general intellect. Atlas of AI uses scales such as Earth, Labor and Data to examine an industrial system of extraction: minerals, energy and space make computation possible; annotation and maintenance are concealed by automation narratives; datasets carry histories of classification into models. Alongside Hao's organizational investigation and Bender/Hanna's critique of hype, "what is AI?" becomes a composite of technical capability, industrial foundations and institutional power rather than an abstract question of intelligence. This final session must distinguish present capability judgments from philosophical possibility. A system's inability today cannot logically prove machines will never do something; conversely, the conceptual possibility of superintelligence cannot establish the political priority of present deployments. Crawford's atlas disassembles AI into a planetary assemblage of minerals, energy, labour, data, classification and state power. AI is not floating software capability, but a highly material and institutional system. Restoring workers, mines and data centres changes risk's timescale: political consequences do not begin only with future superintelligence, but already exist in extraction, work organization and classificatory institutions.]
- Bender, Emily M., and Alex Hanna. The AI Con: How to Fight Big Tech's Hype and Create the Future We Want. Harper, 2025. (Read: Introduction; selected chapters on AI hype, synthetic media, labor, and institutional power).
[Note: Bender and Hanna first disassemble "AI" into industrial discourse, institutional choices and concrete technical systems, opposing the anthropomorphism of machine-learning products as intelligent subjects with a unified will. They examine how hype diverts attention: society may debate distant machine domination while overlooking data, labour, resources, business models and present automation harms. Published in 2025, this text returns discussion to the new narratives following generative AI, complementing Crawford's material politics. AI is a highly heterogeneous umbrella term. Predictive and generative models, robots, optimization systems and hypothetical AGI should not share one capability assessment. Judgments must identify their specific technical object, rather than allowing "AI" to replace classification. Critique of industry discourse must stay connected to particular systems, organizations and observable effects, avoiding another grand anthropomorphic story in place of technical myth. Their focus is not only model limitations, but how exaggerated abilities serve investment, concentration and policy agendas. Compared with technical-risk accounts, this turns attention to existing institutional consequences: who can define AI, who benefits from its myths, and who bears deployment costs? Hype has political effects because it changes resource allocation. Grand narratives of general intelligence and automated substitution influence investment, talent, regulatory priorities and public attention, marginalizing some current problems. Critiquing hype does not deny technical change; it demands a stricter evidential connection between capability claims and social decisions. Demystification therefore does not diminish technology's importance, but returns judgments to testable capabilities, institutions and resource relations.]
- Narayanan, Arvind, and Sayash Kapoor. AI Snake Oil: What Artificial Intelligence Can Do, What It Can't, and How to Tell the Difference. Princeton University Press, 2024. (Read: Chapter 1, "Introduction"; Chapter 2, "How Predictive AI Goes Wrong"; Chapter 4, "The Long Road to Generative AI").
[Note: Narayanan and Kapoor offer a technological-epistemological "disenchantment." They distinguish predictive AI, generative AI and broader AI discourse, stressing capability boundaries and deployment conditions. This is crucial here: critical theory must distinguish actual technical capabilities from the "AI" of industry promotion. AI Snake Oil: What Artificial Intelligence Can Do, What It Can't, and How to Tell the Difference thus offers a limited but effective scale for explaining how resources, labour, ecology and organizational history constrain AGI futurism. It identifies mechanisms without replacing further analysis of institutions, law and material conditions. Different technical objects require different evidence; predictive failure, generative capabilities and AGI imaginaries are not interchangeable. Critique that treats all systems as one intelligence reproduces marketing's vagueness. Demystification begins with classification. A key method of AI Snake Oil is matching every capability claim to its use case. Benchmark performance cannot automatically demonstrate suitability for hiring, education or medical decisions; fluent text cannot demonstrate factual reliability. Evaluation must move from abstract abilities to deployment conditions, error costs and alternatives. This distinction is especially useful at the reading group's close, preventing philosophical futurism and real system capabilities from substituting for each other within one concept of "AI."]
- Hao, Karen. Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI. Penguin Press, 2025. (Read: Introduction; selected chapters on OpenAI's scaling strategy, data labor, compute infrastructure, and global environmental impacts).
[Note: Through long-term investigative reporting, Hao reconstructs OpenAI and the AI race, connecting model expansion with computing power, capital, data labour, global supply chains and environmental conflict. Rather than abstract AI philosophy, organizational history and global cases show how "scaling" became industrial strategy. Its place among the final session's required readings demands that philosophical projections of AGI, superintelligence and posthuman futures face actual organizations and resource politics: who builds these systems, who bears the costs, and who can define the future? Posthuman critique must not escape material politics. Challenging anthropocentrism matters, but concrete institutions still allocate minerals, labour, energy, ecological costs and intellectual property; relational theories of subjectivity must be read with these structures. In this session, Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI makes conflicts over technological futures among risk governance, emancipatory imagination and democratic institutions concrete. AI is first a heterogeneous collection of objects: predictive and generative models, robots, optimization systems and hypothetical AGI have different capability and risk structures. One undifferentiated concept of intelligence quickly blurs empirical judgment. Through OpenAI's history, Hao places AGI ideals, financing, organizational governance and the compute race in one narrative. Its value is not only corporate inside information, but showing how the normative aim of benefiting all humanity encounters institutional pressures in commercial expansion, competition for talent and infrastructure investment. AI futurism is not a history of ideas detached from organizations, but bound to governance and capital structures. Corporate history also shows why governance cannot rely solely on mission statements. Organization, boards, investors and computing resources alter how far a mission can constrain commercial decisions. Public interest existing only as corporate self-regulation is easily reinterpreted under competition to scale.]
Recommended Readings:
- Turing, Alan M. "Computing Machinery and Intelligence" Mind 59, no. 236 (1950): 433–460. (Read: entire article).
[Note: Turing's most important move is not asking "Can machines think?" but reformulating the question. Through the imitation game, a catalogue of objections and learning machines, he asks which behavioural criteria can make machine intelligence an operational object of research. Read after AI discourse of the 2020s, intelligence appears not as a transparent natural property, but as something defined by tests, tasks and institutional expectations. This final session must distinguish actual capability from philosophical possibility. Theoretical resemblance does not automatically imply institutional identity. Turing's paper is better understood as a methodological shift in how to discuss machine intelligence than as an AGI roadmap. The imitation game temporarily suspends metaphysical debates about thought's essence, replacing them with operational behavioural criteria. This opens computational intelligence research while leaving a persistent question: tests can measure performance, but cannot automatically settle understanding, subjectivity or social responsibility. Turing also anticipates objections to machine intelligence and shows their logical weaknesses. The point is not that he proves machines think, but that he moves debate from essential definitions to testable questions. We should preserve this restraint today: passing a test warrants judgments about that tested capability. Behaviour can be evidence without immediately yielding conclusions about consciousness, understanding or subjectivity. Turing offers conceptual discipline for discussing intelligence, not a prophecy of technology's future. His operationalism also leaves an acute contemporary problem: increasingly diverse tests reveal systems excelling at some tasks while lacking robustness at others. Should intelligence remain a single umbrella term, or be differentiated into abilities? This directly concerns the scope and evidential standards of generative-AI capability claims.]
- Russell, Stuart. Human Compatible: Artificial Intelligence and the Problem of Control. Viking, 2019. (Read: Introduction; selected chapters on objectives, uncertainty, and the control problem).
[Note: Russell raises control and alignment from within AI research. Rather than predicting a definite superintelligence catastrophe, he critiques the design assumption that fixes machine goals in advance as a single utility function. He proposes that machines remain uncertain about human preferences. With critical theory, we should ask how those preferences form, who may represent them, and whether social conflict can be reduced to preference learning. AI is a highly heterogeneous umbrella term. Russell formulates advanced-AI risk through goal specification and uncertainty management: if human-given goals are treated as a perfectly certain utility function, increasing capability may magnify specification errors. This deserves dialogue with social science, because preferences are neither unified, transparent nor static. Alignment thus concerns not only control engineering but value conflict, democratic authorization and institutional representation. "Human values" cannot simply be entered into a system as one unified object. Groups genuinely conflict, and values change with context. Treating alignment only as preference learning risks compressing democratic disagreement, rights and non-negotiable principles into optimizable variables. Technical control necessarily leads to political philosophy. In actual institutions, Russell's problem must also handle multiple actors. When a system serves a hospital, company, state or individual, whose preferences prevail? Values cannot always be compressed into one objective function. Alignment therefore needs procedural justice, limits imposed by rights and democratic decision-making; otherwise, "human compatible" may presuppose a unified human will that does not exist.]
- Haraway, Donna. "A Cyborg Manifesto" In Simians, Cyborgs, and Women, 149–181. Routledge, 1991.
[Note: Haraway's cyborg is not merely a prediction of human-machine fusion, but a theory of political subjects that dismantles boundaries. It asks whether politics must rest on a pure biological body and unified identity, offering the multitude an image of coalition without homogeneity. "A Cyborg Manifesto" brings the question of how resources, labour, ecology and organizational history constrain AGI futurism into institutional and organizational analysis. Cyborg politics rejects pure, unified, naturalized identities as foundations for coalition. Technical practices continually redraw human/machine and nature/culture boundaries, while heterogeneous relations constitute political subjects. Its relevance to AI lies not in a fusion prophecy, but in reorganizing responsibility and solidarity without an essentialist account of human nature. Cyborg politics is also anti-essentialist coalition-building: instead of pure womanhood, humanity or natural identity, Haraway stresses partial connections formed where technology, bodies and power intertwine. For the AI age, the point is not that everyone becomes half-machine, but that political subjects are already constituted through technical and social relations. Responsibility and solidarity cannot therefore rest on recovering a pure pre-technological humanity. The cyborg also challenges technology as an external tool. Bodies, medicine, communication and labour have long been remade by technical devices; humans have never been independent subjects outside technology. "Protecting humanity" in AI governance must specify the capacities, relations and conditions of life to be protected, rather than invoking abstract pure human nature. Haraway turns this hybridity into a political resource instead of a technology-induced identity crisis.]
- Hayles, N. Katherine. Bacteria to AI: Human Futures with Our Nonhuman Symbionts. University of Chicago Press, 2025. (Read: Chapters 1–2, 6–7, and 9–10).
[Note: In this 2025 book, Hayles proposes an integrated cognitive framework, extending cognition and meaning-making from humans to living organisms and some artificial systems while retaining an emphasis on materiality, embodiment and relation. It is more suited than How We Became Posthuman to contemporary AI debates because it directly addresses GPT-era claims about machine cognition without equating language models with human minds. Its value is a third theoretical route between strong anthropocentrism and the declaration that machines are already subjects. Posthuman critique must not become an escape from material politics. The contest over technological futures among risk governance, emancipatory imagination and democratic institutions in Bacteria to AI: Human Futures with Our Nonhuman Symbionts extends beyond internal conceptual debate. Hayles distributes cognition across relations among humans, nonhuman life and technical systems, so intelligence is no longer simply an isolated subject's internal capacity. This 2025 account connects ecology with AI: technical systems are embedded in biological symbiosis while altering how humans understand cognitive boundaries. Posthuman futures are thus first reorganizations of relations, not upgrades to one species. Hayles further treats cognition as a process across scales: bacteria and technical systems may sense and respond to environments differently. Cognitive pluralism rejects human reason as the sole measure without equating all information processing with human intelligence. Its key contribution is a vocabulary for distinguishing forms of cognition, freeing posthuman debate from oscillation between human exceptionalism and machine anthropomorphism.]
- Bostrom, Nick. Superintelligence: Paths, Dangers, Strategies. Oxford University Press, 2014. (Read: Chapter 8, "Is the Default Outcome Doom?", pp. 115–142).
[Note: Bostrom offers an existential-risk framework quite different from accelerationist philosophy. One need not accept a doom scenario in advance; the task is to examine the argument: if we posit a highly autonomous, goal-directed agent, why does alignment become central? Comparison with political economy asks whether AI risk can be abstracted into a purely technical issue detached from property, labour and institutions. The final session must distinguish actual capability from philosophical possibility. At the scale of Superintelligence: Paths, Dangers, Strategies, distinguishing AI capabilities, industrial hype, philosophical intelligence and political-economic reality begins with concrete relations. Technological futures are not purely objects of prediction. Industry roadmaps, risk narratives and accelerationism affect investment, regulation and public imagination, so what will happen and who can define the future must be discussed together. Through Superintelligence: Paths, Dangers, Strategies, the distinction becomes a stricter question: which mechanisms actually change, which changes are only shifts of medium, and who bears their consequences? Bostrom assumes extreme capability growth and goal preservation to derive existential risk and the control problem. The value is a clear structure of reasoning about future risks; the limitation is heavy dependence on assumptions about agents, goals and technical paths. Juxtaposing it with present political economy distinguishes long-range scenario analysis from already observable infrastructural power.]
- Brassier, Ray. "Prometheanism and Its Critics" In #Accelerate: The Accelerationist Reader, edited by Robin Mackay and Armen Avanessian, 467–488. Urbanomic, 2014.
[Note: Brassier's value is not an AI development roadmap, but his problematization of technology, reason and political emancipation. The focus is why he refuses to treat naturalness, authenticity and organic life automatically as measures of progress. This forces a distinction between anti-technology and anti-capitalism: if technology can itself be reorganized, can rejecting it really bring emancipation? We must also test Prometheanism's dangers: might a politics of universal reason again dominate difference and ecological conditions? "Prometheanism and Its Critics," Bostrom's Superintelligence: Paths, Dangers, Strategies, and Land's "Teleoplexy: Notes on Acceleration" locate the different problems of automation, control, alignment and posthuman subjectivity in distinct intellectual traditions. AI is a heterogeneous umbrella term. Brassier's essay releases these problems from excessively abstract judgments of the age. Technological futures are not merely predictive objects. The essay contributes by keeping theory refutable. Brassier connects technical transformation with reason's self-correction, rejecting the assumption that every practice exceeding existing nature or tradition expresses domination. This challenges technological pessimism, but must also answer who can transform the world and which institutions determine technical direction. Detached from political economy, Prometheanism can easily abstract unequal real technical capacities into the universal capacity of "humanity."]
- Land, Nick. "Teleoplexy: Notes on Acceleration" In #Accelerate: The Accelerationist Reader, edited by Robin Mackay and Armen Avanessian, 511–520. Urbanomic, 2014.
[Note: Land is this session's strongest counter-pole. He describes capital, technology and self-acceleration as dynamics beyond human will. The value of reading him lies not in accepting his extremes, but in asking what place remains for political subjects once capital is imagined as a self-accelerating system. With Brassier, differences between left and right accelerationism over reason, capital and the future become visible. "Teleoplexy: Notes on Acceleration" allows the constraints that resources, labour, ecology and organizational history place on AGI futurism to be separated into connected processes. Technological futures are not purely predictive objects. From the perspective of Land's essay, these constraints cannot be considered apart from institutional authorization and resource distribution: technical forms produce durable social effects only with organizational power. Land describes capital and technical acceleration as self-organization beyond human political control, taking accelerationism to an anti-humanist extreme. As a counter-text, it exposes the logic that naturalizes capital's dynamics: once acceleration becomes unstoppable, politics is left with adaptation or exit. Critique must re-establish how institutions, labour and collective action can still alter technical trajectories. Land also provides an extreme test of inevitability rhetoric. When capital's acceleration transcends political choice, institutions and collective action are declared ineffective in advance. Critiquing this requires more than advocating deceleration: we must distinguish technical speed, capitalist competition and social goals, showing that their directions do not necessarily coincide.]
- Negarestani, Reza. Intelligence and Spirit. Urbanomic, 2018. (Read: Part I, especially Chapter 1, pp. 1–52).
[Note: Negarestani shifts intelligence from whether machines can reproduce the human brain to concepts, inference, self-correction and normativity. This is speculative philosophy, not a technical roadmap. It confronts critical theory with a difficult question: if intelligence should not be defined as a natural human property, how should the political difference between "artificial" and "natural" intelligence be understood? Intelligence and Spirit, Land's "Teleoplexy: Notes on Acceleration," and Hui's Recursivity and Contingency move contests over technological futures among risk governance, emancipatory imagination and democracy from broad judgments to mechanisms. Posthuman critique must not escape material politics. Historical analogies are useful only at the level of mechanisms. Declaring a new era merely because algorithms, platforms or models have appeared deprives theory of its ability to distinguish continuity and rupture. The contribution of Intelligence and Spirit is to return broad judgments to identifiable institutional mechanisms. Technological futures are not merely predictive objects. How they are contested must therefore remain answerable to evidence and competing explanations beyond Intelligence and Spirit. Negarestani understands intelligence as a practice of giving reasons, revising commitments and participating in a normative space, not merely solving problems. Intelligence is therefore not monopolized by a biological substrate, but neither can it be equated casually with model performance. Whether machines enter a space of normative reasoning concerns responsibility, conceptual use and self-correction—a more demanding question than whether they resemble humans.]
- Hui, Yuk. Recursivity and Contingency. Rowman & Littlefield, 2019. (Read: Introduction, pp. 1–28).
[Note: Hui challenges the assumption of a single modern technological path. Recursivity and contingency make technology a historical practice shaped by different cosmologies, cultures and institutions, rather than linear progress alone. AGI futures need not default to ever greater concentration, power and automation; technical institutions can follow different historical paths. The final session must distinguish actual capability from philosophical possibility. Recursivity and Contingency clarifies causal levels among AI capabilities, industrial hype, philosophical intelligence and political-economic reality. The account in Recursivity and Contingency of these distinctions holds only where the relevant institutional mechanisms can be identified empirically. Recursivity is not a simple celebration of technical autonomy: together with contingency, it reopens technological history's multiple paths. Modern technology is not the only possible organization of the world; different cosmologies and institutional traditions can produce different technical relations. AI's future need not be presumed to follow one path of scaling and concentration. Cosmotechnics further shows that technological diversity is not different cultural labels on identical equipment, but different combinations of technical purposes, conceptions of nature and social orders. Facing global AI infrastructure, this asks a deeper question than product localization: can different societies determine the scale, pace and normative goals of technical development? Technological diversity is first a question of political capacity—whether societies can genuinely determine their own technical institutions, rather than choose cultural styles atop a predetermined global infrastructure.]
- Braidotti, Rosi. The Posthuman. Polity, 2013. (Read: Chapter 1, "Post-Humanism: Life beyond the Self," pp. 13–54).
[Note: Braidotti distinguishes the posthuman subject from the traditional humanist subject, offering a posthumanism different from transhumanist enhancement. Must technological futures aim for "stronger humans," or can they rethink life, ecology and relational subjectivity? AI is a heterogeneous umbrella term. The Posthuman returns analysis of automation, control, alignment and posthuman subjectivity to evidential limits: without explicable connections among technology, institutions and political consequences, theoretical analogies remain suggestive. The Posthuman must remain distinct from transhumanist enhancement. Rather than stronger humans, it centres the subject's relations with nonhuman life, technology and environment, giving AI futurism an important counterpoint. Braidotti's posthumanism differs fundamentally from technological enhancement. It critiques humanism's claim to universality around an autonomous, rational European male subject, insisting that subjects are always embedded in life, technology and ecology. Posthuman politics thus concerns not more powerful individuals, but redistribution of interdependence and responsibility among humans, nonhuman life, technical systems and environments. Affirmative politics also reminds us that critique cannot stop at exposing humanist exclusions. Relational subjectivity can ground a new ethics: recognize dependence among humans, animals, ecology and technical systems, and redistribute responsibility accordingly. AI futures concern not only capability upgrades, but which lives and environments enter the valuation of technical development.]
- Srnicek, Nick, and Alex Williams. Inventing the Future: Postcapitalism and a World Without Work. Verso, 2015. (Read: Introduction; chapters on full automation, post-work politics, and institutional strategy).
[Note: Srnicek and Williams place automation within left political strategy rather than technical destiny. Full automation, shorter working hours and universal basic income seek to redirect technical capacity away from capital's profit logic. In the final session, they form a triangle with Land and Bostrom: is automation capital's self-acceleration, a technical risk requiring control, or a condition that political organization can turn into reduced necessary labour? Inventing the Future: Postcapitalism and a World Without Work moves the constraints that resources, labour, ecology and organizational history place on AGI futurism from rhetoric to institutional analysis. Inventing the Future situates automation in a post-work institutional imagination: reducing necessary labour becomes emancipatory only alongside changes to income, public services, working time and ownership. Otherwise, identical capabilities can create unemployment risks and stronger management. No automatic passage connects technical potential to political results. Post-work politics also distinguishes reduced working time from eliminating meaningful activity. Some tasks can be automated, but the value of care, education, culture and political participation cannot simply be subsumed under efficiency. The normative aim should be redistributing necessary labour and free time, not treating "no work" as an end in itself. The radical element is considering automation together with social rights. Shorter hours, basic income, public services and workers' bargaining power are all necessary; otherwise capital may absorb productivity gains as profits and unemployment pressures. Technical progress has no inherently emancipatory direction. Only institutions can turn productivity into greater autonomy over time.]
- Benjamin, Ruha. Imagination: A Manifesto. W. W. Norton, 2024. (Read: Introduction and chapters on imagination as a site of struggle).
[Note: Benjamin treats imagination as a political resource, not a soft cultural supplement. Read with Fisher, Kornbluh and AGI futurism, she asks: if firms can shape public visions of the future through product roadmaps, must resistance reclaim the capacity to determine which futures deserve imagination? Posthuman critique must not escape material politics. How technological futures are contested among risk governance, emancipatory imagination and democratic institutions must therefore face evidence and competing explanations beyond Imagination: A Manifesto. Benjamin raises imagination from private mental activity to public political capacity. The future is not a product awaiting delivery by technology companies, but a space contested through narratives, design, institutions and collective action. At the end of the group, this returns posthuman and AGI debates to democracy: who can propose futures, whose experiences enter their design, and which institutions make alternatives practicable? The politics of imagination also includes the capacity to negate existing institutions. Technical governance often limits the future to incremental upgrades of current products and business models; Benjamin insists it can be reconstructed from suppressed experiences, histories and community practices. This openness is the group's ultimate concern: technological futures are objects not only of prediction, but also of institutional creation. Imagination thus connects with institutional practice. Alternative futures must become organizations, public resources and durable rules, or retreat again into branded visions. This politics also connects directly with design. Datasets, roadmaps and policy scenarios imagine "future users" and "normal society"; design is already distributing possibilities. Expanding imagination means including affected groups in defining problems and setting goals, rather than seeking their feedback only after systems are complete.]
「デジタルインテリジェンス時代の批判理論」夏季読書会への序論
フォーディズムと産業資本主義の全盛期、すなわち私たちになじみ深い初期の、あるいは「古典的」とさえ呼べる近代の段階では、機械制大生産は厳密な物理的空間の囲い込み(enclosure)と時間の線形的分割の上に成り立っていた。資本は労働力を標準化された工場の生産ラインへ強制的に集中させ、プロレタリアートの身体と時間という二つの次元を物理的に規律化した。空間的な高密度の集積に基づくこの搾取形態は、古典的な階級的連帯、労働組合運動、集合的な政治意識の覚醒に不可欠な組織上の前提を、客観的に提供した。同じ屋根の下で労働者が物理的に共在すること(co-presence)によって、搾取の可視性は抵抗の結束力へと転化した。しかし、サイバネティクス(cybernetics)が主導するポスト・フォーディズムの段階に入ると、搾取のトポロジーは根本的な存在論的変容を遂げた。資本蓄積の論理は、物理的空間における絶対的剰余価値の直接的な搾取だけに依存するのではなく、非物質的労働(immaterial labour)、認知的剰余、そしてデジタル・プラットフォームが絶えず捕捉する「行動余剰(behavioural surplus)」の継続的な抽出へと向かうようになった。
今日のデジタルインテリジェンス・プラットフォームとアルゴリズム資本主義(algorithmic capitalism)の文脈では、従来の意味での物理的工場は徹底的に解体され、再編成されている。労働力はAPI、クラウドソーシング・プラットフォーム、外注されたデータ・ラベリングのネットワーク、ギグ労働を配分するアルゴリズムによって分散され、高度に分散的、ブラックボックス的、非中央集権的な特徴を帯びる。統制の媒体は、目に見える煉瓦や石の壁、監督者のまなざしから、不可視のTCP/IPプロトコル、ディープ・パケット・インスペクション(DPI)、密かに作動する予測的機械学習モデルへと沈み込み、拡散している。このような背景の下では、「ウォール街を占拠せよ」に代表される従来の物理的空間をめぐる闘争は、道具的・認識論的な水準で深刻な構造的ずれを抱え、「コードは法である(code is law)」という基底的なネットワーク構造を前に、その効力を失わざるをえない。搾取がもはや直接的な暴力的強制として現れず、日常生活に継ぎ目なく埋め込まれたインターフェース操作や個人向け推薦へと姿を変えるなら、抵抗の物理的な足掛かりもまた解消されるのだろうか。
この夏季読書会は、従来の人文主義的な道徳批判や単純すぎる技術拒否論をいったん括弧に入れ、理論的探究の視野を、プロトコル、暗号技術の基盤、惑星規模の計算スタック(The Stack)、そして異質な主体をめぐるミクロな権力作用のネットワークへと直接向けることを目指す。「デジタルインテリジェンス時代」は技術進歩を総括する便利な標語ではなく、歴史化され、理論化されるべき問題領域である。デジタル・プラットフォーム、機械学習、クラウド・コンピューティング、データセンター、自動化された管理、生成AIは、資本蓄積、国家統治、メディアによる媒介、労働組織、知識生産、政治的主体の関係を組み替えつつある。しかし、これらの変化は産業資本主義の単純な代替としても、旧来の制度に技術的なインターフェースをかぶせただけのものとしても説明できない。より問うべきなのは、どの制度的関係が計算によって規模、速度、可視性を変えたのか、どの権力形態が新しい媒体を通じてより精密な作動条件を得ただけなのか、そして「断絶」と名づけられたどの現象が、実際には管理科学、広告業、植民地統治、インフラ、分業の長い歴史を継続しているのか、ということである。デジタル化と知能化は新たな研究対象を形成すると同時に、批判理論に自らの概念の適用限界を再検討するよう迫る。
第一の問いの系列は、統制の形態と計算可能な主体の生成をめぐる。Wienerのフィードバックと通信から、Foucaultによる規律装置の分析、さらにDeleuzeの変調(modulation)、Gallowayのプロトコル、Rouvroyのアルゴリズム的統治性へと至る諸理論は、同一の権力機構が連続的に高度化する過程を記述しているのではない。規律は空間の分割、時間配分、試験、規格化に依存し、プロトコルは技術標準を通じて相互運用性を組織する。一方、アルゴリズム・システムは分類、スコアリング、予測、順位づけ、推薦によって行為の環境を形づくりうる。これらは重なり合うこともあれば、国家の法的権限、企業統治、技術的アーキテクチャの間に対立を生むこともある。したがって問題は、「アルゴリズムが新しい主権者になったか」ではなく、計算結果がどのように制度的手続きに入り込み、どの組織が確率的出力を資格、リスク、処罰へと変換するのか、誰が解釈、変更、不服申立ての権限を持つのか、そしてモデル内の不確実性が組織の過程でいかに現実の帰結へと固定されるのか、という点にある。
第二の問いの系列は、監視、抽出、植民地性、インフラ化へと進む。データは自然に存在して収集を待つ資源ではなく、センシング、ログ、ラベル、識別システム、プラットフォームのインターフェース、ビジネスモデル、統治上の要求を通じて絶えず生産される。Zuboffの監視資本主義、CouldryとMejiasのデータ植民地主義、Browneによる人種化された監視の歴史分析、Mbembeによる植民地性、敵意、生の曝露をめぐる政治的考察は、相互に完全には両立しない説明枠組みをなす。「植民地」「封建」「帝国」といった概念は、対応する機構を説明できる場合にのみ分析的な意味を持つ。資源の占有、法的不均衡、知識の権威、強制的依存、空間の分割、差異化されたリスクは、相互に置き換えられる修辞ではない。同様に、「クラウド」もインターフェース上の比喩にとどまってはならない。データセンター、海底ケーブル、半導体、エネルギー、鉱物、土地、電波周波数、標準、物流、保守労働が、計算の物質的条件をともに構成する。それゆえ、プラットフォームの権力はインフラ、領土、レント、退出費用と併せて論じられなければならない。
第三の問いの系列は、プラットフォームの政治経済、労働、新たな依存関係に焦点を当てる。VaroufakisとDurandのテクノ封建主義という命題は、ここではプラットフォーム資本主義、独占、レント理論、インフラ研究の間に位置づけて検討される。本当に見極めるべきなのは、「資本主義はすでに死んだのか」という修辞的な問いではなく、クラウド上の入口、ネットワーク効果、無形資産、プラットフォーム統治が、従来の資本理論では十分に説明できない依存構造を形成しているのか、そしてプラットフォームのレントと封建的地代がどの水準で比較可能であり、どの水準で比較できないのか、という点である。プラットフォーム労働研究は、これらの巨視的な問題を、注文、評価、配車・配達の割当て、マイクロタスク、データ・ラベリング、コンテンツ・モデレーション、企業家的主体、グローバルな労働連鎖へと引き戻す。BerlantのCruel Optimismが提示する情動と時間性の理論は、とりわけ重要である。不安定な「自由」「柔軟性」「自己起業」が、なぜ主体にとって手放しがたい愛着の対象であり続けるのか。また、制度変革は所有権、リスク配分、生活時間、未来への期待にどう同時に働きかけられるのか。
第四の問いの系列は、メディアによる媒介、公共性、組織へと向かう。デジタル政治はしばしば、接続費用の低下、情報流通の加速、ゲートキーピングの弱体化として語られる。しかし公共性はトラフィックと同じではなく、組織は即時的な集合と同じでもない。Keltyの再帰的公共は、公共が自らの存在を支える技術的・制度的条件を維持し変更できるかを問う。WarnerとFraserは流通の循環、対抗的公共、参加の不平等を論じる。KornbluhのImmediacy, or The Style of Too Late Capitalismは、透明性、即時性、没入、脱媒介化への魅惑を文化形式の面から批判する。政治的行為はしばしば、手続き、代表、記憶、遅延、解釈、責任の連鎖、持続可能な組織に依存するが、プラットフォーム・メディアは「すぐに見えること、すぐに応答すること、すぐに流通すること」をそれ自体の価値に仕立てる場合がある。したがって、分散化は本来民主的であるわけではなく、媒介は本来抑圧的であるわけでもない。より重要なのは、媒介を責任あるものにし、変更し、参加者が共同で形づくることができるか、という問いである。
第五の問いの系列は、一般的知性、計算資源の集中、技術の未来に焦点を当てる。MarxはGrundrisseの「機械にかんする断章」で、社会的知識、固定資本、労働時間、価値形態の間の緊張を論じる。Virnoはさらに、一般的知性が機械に凝固しているだけではなく、言語、判断、記憶、協働、共通の能力のうちにも存在すると強調する。生成AIはこの問題を新たな形で浮上させる。モデルの訓練は公共の文章、コード、画像、フィードバック、ラベリング労働、膨大な物理的計算資源に依存する一方、これらの社会的条件は私有モデル、インターフェース、クラウド・サービスへと再び包み直されうる。したがって分析すべきなのは、「LLMこそ一般的知性なのか」ではなく、社会が共同で生み出した認知の痕跡がどのような所有権とインフラの機構を通じて集中されるのか、アクセス、訓練、配備、収益の権利がどう配分されるのか、そしてオープン・ウェイト、公共的計算資源、相互運用性、協同的統治がそれぞれどの層の関係を変えられるのか、ということである。
最後に行うAI、自動化、ポストヒューマンな未来をめぐる議論は、AGIが実現する時期の予測を目的としない。Turing、Russell、Bostrom、Brassier、Land、Negarestani、Haraway、Hayles、Braidottiは、「知性」「人間」「機械」「統制」「解放」「未来」を大きく異なる仕方で理解している。一方、Crawford、BenderとHanna、NarayananとKapoor、Haoは、これらの哲学的問題を、現実の技術、組織、資源の関係へと繰り返し引き戻す。哲学的思弁、工学的判断、産業予測、政治経済学、経験的社会研究は異なる証拠基準を持ち、相互に代替できない。記述的判断、因果的説明、規範的判断も区別されるべきである。ある権力機構を明らかにしても、そこからただ一つの政治的方策が自動的に導かれるわけではなく、ある制度設計が技術的に実行可能であっても、その正当性が保証されるわけではない。この読書会では、概念史、経験的機構、規範的問題の間に検証可能なつながりを築くことを試みる。それによって、古典理論を万能の鍵として崇めることも、時代の違いを理由に現実を解釈し直す能力を奪うことも避けたい。
| 回次 | 中心となる問い |
| 第1回:サイバネティクス、統制、デジタル統治の生成 | サイバネティクス、規律、管理社会、プロトコルは、いかにデジタル統治の歴史的前提を構成するのか。 |
| 第2回:アルゴリズム的統治性と生政治 | 分類、予測、スコアリング、順位づけは、いかに計算可能な主体を生み出し、制度的な効力を獲得するのか。 |
| 第3回:監視、抽出、データの植民地性 | データの抽出、監視、人種化された分類、植民地性の間に、単なる修辞ではない厳密な関係をどう確立できるか。 |
| 第4回:惑星規模の計算とインフラ空間 | クラウド、プラットフォーム、データセンター、インフラは、空間、権威、レント、退出費用をどのように再編成するのか。 |
| 第5回:再帰的公共、媒介、デジタル政治 | 公共は自らのメディア条件をどう所有し変更できるか。媒介と遅延はなぜ政治的組織化の条件となりうるのか。 |
| 第6回:プラットフォーム労働、不安定性と新たな階級関係 | プラットフォーム労働、アルゴリズム管理、不安定な生活は、労働関係と集団的行為をどのように再編成するのか。 |
| 第7回:一般的知性と計算資源の集中への抵抗 | 社会が共同で生み出す知性はどのように固定資本化、インフラ化され、再び公共化されうるのか。 |
| 第8回:人工知能、自動化とポストヒューマンの未来 | AIの現実の能力、産業的な誇大宣伝、リスクをめぐる議論、ポストヒューマンな未来の想像を、どう区別できるか。 |
第1部:統制のトポロジー(第1–2週)
第1回:サイバネティクス、統制、デジタル統治の生成
本回は、デジタル政治理論がしばしば見落とす前史から出発する。「アルゴリズム」が日常語となる以前に、20世紀はすでに通信、フィードバック、情報、統制によって機械、組織、社会を記述し直す知の言語を形成していた。Wienerのサイバネティクス、冷戦期の指揮統制、産業管理における情報処理の需要、分散型通信ネットワーク、ソフトウェア・プロトコルは、デジタル統治の歴史的条件をともに構成する。ただし、これらの条件は今日のアルゴリズム統治へと必然的に至る線形の連鎖をなすわけではない。Foucaultの規律権力は空間の分割、時間配分、試験、規格化に依存し、Deleuzeの変調は連続的かつ可変的な統制の形態を指す。Gallowayのプロトコルは、分散型ネットワークが接続を認めると同時に操作可能性を規定する仕組みを扱う。三者の違いは、まず対象、尺度、機構の違いである。
この系譜を考える際、統制を「中心が存在するか」という問いへと還元することはできない。標準、プロトコル、インターフェース、識別、権限、ハードウェア、法はそれぞれ異なる層で制約を形成する。実行は高度に分散されていても、標準策定、資源の所有権、変更権は集中したままかもしれない。Baranの分散型通信はまず工学的設計であり、Mayの暗号アナーキズムは政治的宣言であり、Berners-LeeのWeb提案は情報管理上の問題に起源を持つ。DeNardisは、技術標準がつねに国際的機関、企業、国家権力の交差する場所にあることを示す。同じネットワーク・トポロジーが、軍事的な強靱性、開かれた協働、商業プラットフォーム、監視システムのいずれにも奉仕しうる。ゆえに、「分散型」「オープン」「暗号化」といった構造上の名称から、その政治的性格を直接導くことはできない。
こうして本回は、多層的な統制概念を構築する。技術システムは規則のコード化によって操作可能性を限定することもあれば、市場の集中、組織的手続き、国家の法的権限に依存して実際の効力を得ることもある。いわゆる「プロトコルの政治」は法の政治に取って代わるものではなく、標準、ソフトウェア、制度、主権が入れ子状に組み合わさった結果である。この区別が、次回のアルゴリズム統治の検討の基礎となる。規則がアーキテクチャ、プロトコル、組織、法のどの層で作用するのかを明確にして初めて、スコアリング、予測、推薦のシステムが行動を計算可能な対象へと変える仕方を分析できる。
必読文献:
- Wiener, Norbert. The Human Use of Human Beings: Cybernetics and Society. Houghton Mifflin, 1950. (Read: Chapters 1–2, pp. 15–47).
【注:Wienerは通信、フィードバック、統制によって、人間と機械、そして社会組織の関係を記述し直す。ここでの目的は、サイバネティクスをそのままアルゴリズム統治の理論とみなすことではなく、「情報とフィードバックによってシステムを管理する」という知の言語がいかに形成されたかを考察することにある。これは後に扱うプロトコルとアルゴリズムに歴史的背景を与えるとともに、今日「データ駆動型統治」と呼ばれるものの基本的な想像力の一部が、インターネットに始まるのではなく、より長いサイバネティクスの系譜を持つことを思い出させる。より重要なのは、まず原典の対象と技術的な層を明確にし、その後で構造的な同型性を探すことである。法的環境、組織的主体、物質的条件が異なるなら、類比は証拠ではなく示唆として位置づけなければならない。この枠組みを現代のデジタルな対象に適用する際は、類似性がどの層で成立するのかを見極める必要がある。法的環境、組織的主体、物質的条件が変われば、概念の説明範囲も変わる。WienerのThe Human Use of Human Beings: Cybernetics and Societyはさらに、「サイバネティクス、規律、管理社会、プロトコル統治の歴史的差異」を制度的関係へと具体化する。デジタル統治の鍵は中心が消えるかどうかではなく、命令、フィードバック、標準、保守の権限が異なる組織の間でどう組み替えられるかにある。実行は高度に分散されていても、標準策定と資源所有は集中しうるため、ネットワーク・トポロジーと政治的分権は分けて判断しなければならない。Wienerの意義は、フィードバックを機械内部の技術的過程としてだけ扱わず、通信、エントロピー、組織を結びつけ、工学と社会理論を横断できる言語を形成した点にもある。その高い移植可能性ゆえに、この言語は後の経営学、軍事システム、計算文化において、当初の工学的文脈をはるかに超える影響を得た。今日サイバネティクスを読み直すなら、フィードバックがいつ管理原理となり、いつ社会の自己調整に関する普遍的モデルとして神話化されるのかを考察するのが適切である。】
- Foucault, Michel. Discipline and Punish: The Birth of the Prison. Translated by Alan Sheridan. Vintage Books, 1977. Originally published 1975. (Read: Part III, Chapter 3, "Panopticism," pp. 195–228).
【注:本章は比較のための理論的出発点をなす。Foucaultが論じるのは「監視」そのものではなく、規律が空間の分割、時間配分、試験、記録、可視性の制度化を通じて、いかに統治可能な身体を生産するかである。パノプティシズムの鍵は、誰かを絶えず見つめる監視者の存在ではなく、観察される者を原理的につねに「見られうる」状態に置き、外的強制を主体の自己調整へと転化する点にある。この枠組みをデジタル・プラットフォームへ移す際には距離を保たなければならない。データ・システムは、建築空間における身体の配置を単純に複製するのではない。むしろログ、権限、評価、推薦、識別システムを通じて、行為を計算し再利用できるデータ対象へ変える。そのため、監視が「誰が私を見ているのか」から「どのシステムが私の行為を記録し処理できるのか」へ移るとき、パノプティシズムの理論はどう書き換えられるべきか、と問うことができる。プラットフォームによるデータ化と規律装置の関係は具体的機構の比較によってのみ考察でき、「可視性」という共通の語から直接導くことはできない。デジタル・プラットフォームとパノプティシズムの間で本当に比較可能なのは、「見る」という表象ではなく、可視性がどう制度化され、主体形成に組み込まれるかという点である。プラットフォームの記録は、商業的予測、リスク管理、組織管理に同時に利用されることが多く、そのデータの可視性もすべての行為者に対称的に開かれてはいない。したがってデジタル監視は、Bentham的建築モデルの単純な電子化よりも、規律技術とデータ・インフラの再結合として理解する方が適切である。さらに、デジタル環境は規律権力の時間構造を変える。従来の試験は明確な時点で行われることが多かったのに対し、プラットフォームの記録は連続的な蓄積と事後的な再計算を可能にする。過去には無関係に見えた行動が、新しいモデルや方針の下で新たな意味を得る場合もある。この遡及可能で再計算可能な可視性によって、デジタル統治は規律の記録論理を継承しながらも、監獄建築とは異なる時間の政治を形成する。】
- Deleuze, Gilles. "Postscript on the Societies of Control" October 59 (1992): 3–7. Originally published 1990.
【注:Deleuzeの「管理社会」はインターネットを経験的に予測したものではなく、権力形態の変化を扱う高度に圧縮されたモデルである。その核心は、閉じられた規律的な鋳型(mold)から連続的な変調(modulation)への移行にある。「分人(dividual)」は、個人が複数のデータ断片、信用、パスワード、アイデンティティの形で異なるシステムに入ることを意味する。このテクストは問いを生み出す装置として扱うべきであり、分人をそのまま「データ・プロファイル」と訳して済ませるべきではない。今日のデジタル統治に適用する際、最も補う必要があるのは技術用語ではなく実装の連鎖である。規則はどこでコード化され、誰が配備し、誰が保守し、どの法的・市場的関係がその規則に効力を与えるのか。ゆえにDeleuzeの"Postscript on the Societies of Control"において、「分散型トポロジーと政治的分権は同一ではない」という指摘は単なる概念上の問題ではない。デジタル統治の鍵は中心が消えるかどうかではなく、命令、フィードバック、標準、保守の権限が異なる組織の間でどう組み替えられるかにある。実行は高度に分散されていても、標準策定と資源所有は集中しうるため、ネットワーク・トポロジーと政治的分権は分けて判断しなければならない。鋳型と変調の区別は、権力形態が段階的な認証から継続的な更新へと移る変化を記述するのにとりわけ適している。アカウント状態、信用枠、価格、権限、推薦は、データの流れに応じてリアルタイムに調整されうる。ただし「管理社会」はあらゆるデジタル現象の総称ではない。変調に分析力があるのは、変わり続ける参入条件とアイデンティティの分割を指し示すからであり、国家法、組織規則、物理的強制はそれと併存しうる。「管理社会」から最も保持すべきなのは歴史的段階のラベルではなく、権力の時間性の記述である。管理はパスワード、限度枠、資格、更新され続ける状態を通じて、行為の空間を絶えず調整する。それは規律と相互排他的ではない。学校、監獄、企業などの閉鎖的制度は存続しながら、継続的評価や場面を横断するデータの流れとますます結びついている。】
- Galloway, Alexander R.. Protocol: How Control Exists after Decentralization. MIT Press, 2004. (Read: Introduction; Chapter 1, "Physical Media"; and Chapter 2, "Form," pp. 1–92).
【注:Gallowayの中心的な貢献は、統制の問題を単純な「中心―周辺」モデルから移し替えることにある。彼が問うのは、プロトコルが分散型ネットワークでどのように規則的関係を成立させるかである。TCP/IPやDNSなどのシステムは、多数のノード間の通信を認める一方、標準化されたアドレス、形式、権限、変換方式によって、どのような行為が生じうるかを定める。ここでは「プロトコルは統制である」と「プロトコルは法である」の違いをとくに区別する必要がある。Gallowayが論じるのは技術システムがいかに政治的形態を持つかであり、国家法がネットワーク・プロトコルに置き換えられたということではない。理論的にはFoucaultの規律の問題を引き継ぎながら、後のアルゴリズム統治、ブロックチェーン、クラウド・プラットフォームへの概念的な橋渡しを行う。さらに問うべきなのは、プラットフォームが登録、識別、インターフェース、順位づけ、アクセス権限をすべて技術構造に書き込むとき、統制は明示的な命令への依存を弱めるのか、ということである。このテクストをプラットフォーム、クラウド、AIへと拡張するには、歴史的連続性と機構上の差異の双方を保持する必要がある。前者は問題がなぜ再出現するかを説明し、後者は概念が越えてはならない境界を定める。その結果、プロトコルの政治は一見逆説的な構造を示す。通信が共通標準に依存するほど、ネットワークは実行の層では分散しながら、標準策定、命名体系、重要資源の面では新たな集中を形成しうる。Gallowayの価値は、「分散化」を規範的な賛辞から分析可能な技術的・政治的関係へと戻し、トポロジー、統治権限、所有権の混同を解く点にある。プロトコルはまた、政治的対立をバージョン、互換性、標準の変化のうちへ移す。誰が変更を提案する資格を持つのか、どの変更が分岐をもたらすのか、旧来のノードはなお接続できるのか。これらはいずれも技術共同体における統治の問題である。したがってプロトコルは静的な規則ではなく、継続的に保守される制度的過程である。統制は実行時だけでなく、標準が策定され更新される時点でも働いている。】
推薦文献:
- Lessig, Lawrence. Code and Other Laws of Cyberspace. Basic Books, 1999. (Read: Chapter 7, "What Things Regulate," pp. 85–99).
【注:Lessigの「コード」論で最も注目すべきなのは、技術システムと法制度を厳密に切り離していない点である。彼は、行動が法、社会規範、市場、アーキテクチャという四つの規制力によってともに影響されると指摘する。ネットワーク上のコードは、行為の条件を変えることで規制を実施できるが、その効力はなお法的、経済的、社会的関係に埋め込まれている。読む際は「Code is Law」を標語にせず、プラットフォームがパスワード、インターフェース、本人認証、初期設定、支払い規則、コンテンツの順位づけを通じて、ある行為を「違法」なものから「そもそも操作不可能」なものへどう変えるのかを分析すべきである。この問いが「アーキテクチャ的権力」を理解する基礎となる。技術構造の政治的効力は、つねに組織と制度を経由する。プロトコルは相互運用性を規定し、ハードウェアは実現可能な範囲を定め、市場集中は退出費用を形づくり、国家の法的権限は特定の規則がいつ強制力を持つかを決める。これらの層がデジタル統制の実際の形態をともに構成する。Lessigの四つの規制力、すなわち法、規範、市場、アーキテクチャによって、デジタル統治を「コードがすべてを決める」へと縮減せずに済む。アーキテクチャは特定の行為を技術的に不可能にしたり、その費用を高めたりできるが、コード自体の設計と実行も企業のビジネスモデル、法的責任、社会規範の影響を受ける。プラットフォーム統治で最も研究すべきなのは、この四者がどう重なり合うかである。初期設定、価格機構、コミュニティ規則、国家規制は、しばしば同じインターフェースの中でともに作用する。この枠組みはまた、プラットフォーム権力の最も安定した形態が、単一の規制手段ではなく、複数の力の相互補強であることを示唆する。たとえば商業的誘因が特定のインターフェース設計を促し、そのインターフェースが利用者の規範を変え、続いて法がプラットフォームの既成の技術条件を統治対象とする。デジタル・アーキテクチャの政治性は、コード自体が社会を超える独立した主権を持つことからではなく、こうした循環から生じる。】
- Beniger, James R.. The Control Revolution: Technological and Economic Origins of the Information Society. Harvard University Press, 1986. (Read: Introduction and Chapter 1).
【注:Benigerは近代の情報社会を、産業化がもたらした「統制の危機」の中へと位置づけ直す。生産、輸送、分配、消費が加速するにつれ、旧来の組織形態では複雑性への対応が困難になり、情報処理と管理技術が不断に発展した。この見方は、「インターネットが突然統制を発明した」という物語を避け、プラットフォーム、物流、アルゴリズムによる配分を、より長い管理の歴史に置くことを可能にする。議論の焦点は、統制がどのような経済的・組織的要求に応えるのか、という点にある。技術構造の政治的効力はつねに組織と制度を経由する。この長期的視点は、プラットフォーム物流、リアルタイム価格設定、アルゴリズムによる配分の説明にとりわけ適している。それらは無から現れた「知能」ではなく、速度、規模、複雑性の継続的増大への管理上の応答だからである。Benigerはこうして、情報技術史を資本主義の組織史へと接続し直す。いわゆるデジタル転換も、生産と流通の加速が新たな調整圧力をどう生み、誰がその調整を担う能力を持つのかを併せて考察しなければならない。組織史から見れば、「統制の危機」は情報処理と規模拡大の共生関係も明らかにする。企業は在庫、輸送、消費者、労働過程を追跡できて初めて、より広い範囲の調整を維持できる。新たな情報技術は複雑性に応えると同時に、さらなる拡大を可能にする。デジタル・プラットフォームの成長もこの循環の中に置けるのであり、純粋な技術的突破と描く必要はない。したがって情報処理能力そのものも、組織拡大の生産条件と捉えられる。それは既存の複雑性を解決しつつ、より大規模な集中調整への空間を開く。この組織と情報の双方向の強化が、Benigerを産業管理史とプラットフォーム統治史の重要な媒介者にしている。この視点は管理技術と資本主義の規模の問題を同一の歴史的枠組みに置く。情報革命は、組織能力と経済構造の共進化の結果として捉え直されるのである。また、この視点は「リアルタイム性」がなぜ近代的管理の追求し続ける能力となったかも説明できる。】
- Baran, Paul. "On Distributed Communications Networks" IEEE Transactions on Communications Systems 12, no. 1 (1964): 1–9.
【注:Baranの論文は工学史のテクストとして読むべきである。中心的な問いは、高リスク環境で通信システムの冗長性と生存能力をどう高めるかであり、政治的な意味での「分散化」の制度を構築することではない。だからこそ、後のデジタル政治における類比を検証するのに適している。同じ分散型トポロジーが、なぜ軍事的強靱性、ネットワークの自由、プラットフォームの弾力性、さらには監視インフラとして解釈されうるのか。技術構造そのものは単一の政治的方向性を帯びない。「分散型トポロジーと政治的分権は同一ではない」という命題は、Baranの"On Distributed Communications Networks"を通じて組織と制度の層へ入っていく。技術構造の政治的効力は、つねに組織と制度を経由する。この論文の尺度では、この命題は連続性と変形の双方を扱う必要がある。旧来の制度が新技術を通じて新たな作動方式を得る場合もあれば、新技術が既存の制度へ再吸収される場合もある。Baranの工学的議論はまた、分散型ネットワークがまず信頼性のための設計であり、政治制度ではないことを示す。ノードの冗長性は破壊下での通信システムの生存能力を高められるが、誰がネットワークを所有し、誰がプロトコルを定め、誰が接続を切断できるかには答えない。後に"distributed"をそのまま自由、自治、民主主義へと翻訳することは、工学的トポロジーを未証明の規範判断へ変換することにほかならない。工学的冗長性は「強靱性」と「自治」の違いも明らかにする。多経路設計によってネットワークの破壊が困難になっても、設備、周波数、接続許可を少数の機関が握ることはありうる。Baranの価値は、技術史の内部にこの違いを保持し、後の分散型ネットワークをめぐる政治的想像に、トポロジーだけから権力関係を推定するのではなく、所有と統治の問題に答えるよう求める点にある。】
- DeNardis, Laura. The Global War for Internet Governance. Yale University Press, 2014. (Read: Chapter 1, pp. 1–24).
【注:DeNardisはインターネットの技術的統治を直接グローバル政治の中に置く。IPアドレス、ドメイン名、標準、ルーティング、セキュリティは純粋な「技術問題」ではないが、国家主権へ単純に還元することもできない。Gallowayと併読すれば、プロトコル、企業、国際組織、国家機関がネットワーク秩序を共同で生産する仕方が見えてくる。DeNardisのThe Global War for Internet Governanceが明らかにする「技術設計、保守労働、統治の正統性のつながり」は、テクスト内の概念論争にとどまらない。技術構造の政治的効力はつねに組織と制度を経由する。DeNardisのThe Global War for Internet Governanceは、このつながりを特定の対象と因果連鎖に限定できるようにする。その範囲を超える類比には新たな証拠が必要であり、用語そのものに頼って成立させることはできない。DeNardisはインターネット統治をより精密な層へと進める。ドメイン名、アドレス配分、ルーティング、セキュリティ証明書、技術標準は工学的属性を持つと同時に、グローバルな権力関係を構成する。ネットワーク秩序は国家とインターネットの二者択一ではなく、政府、標準化団体、通信事業者、プラットフォーム企業、技術共同体が共同で維持する。ゆえにインフラ統治は、公衆には気づきにくいが接続可能性と表現条件を深く変える技術層で行われることが多い。その複雑さは、技術層が政治以前の段階ではないことも示す。標準化団体がいかに合意に達し、事業者がいかにトラフィックを交換し、認証局がいかに信頼を築くかという、一見舞台裏の取り決め自体が制度なのである。DeNardisはこうしてネットワーク主権を複数の識別可能な統治点へ分解し、「国家がインターネットを統制する」と「インターネットは国家を超える」という二つの包括的命題が、いずれも粗雑すぎることを明らかにする。】
- Turner, Fred. From Counterculture to Cyberculture: Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianism. University of Chicago Press, 2006. (Read: Chapter 1, pp. 11–39).
【注:Turnerは、アメリカのカウンターカルチャー、Whole Earthのネットワーク、パーソナル・コンピューティング文化の間の歴史的な翻訳過程をたどる。本書は、シリコンバレーの起源についてのよくある物語を検証するのに適している。1960年代の共同体、個人の自治、技術的媒体、ネットワークの想像力は、自然にプラットフォーム資本主義へ向かったのではなく、異なる組織的、商業的、文化的な力の中で絶えず組み替えられてきた。TurnerのFrom Counterculture to Cyberculture: Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianismの分析尺度では、「サイバネティクス、規律、管理社会、プロトコル統治の歴史的差異」は、まず具体的な関係群として現れる。より重要なのは、フィードバック、情報、適応が組織的思考の一部となり、その後の規律、プロトコル、プラットフォーム統治が、それぞれ異なる仕方でその関係を再編成する点である。Turnerの歴史叙述は、カウンターカルチャー共同体と後のデジタル資本主義の関係が、単純な裏切りではないことを示す。個人の自治、脱官僚化、ネットワーク協働、起業家文化は、特定の組織環境の中で組み直され、共同体の想像力と市場の論理を併せ持つ技術文化を徐々に形成した。この変容を理解すると、オープン、参加、革新という語が、反体制政治と企業管理の双方に役立ちうる理由がわかる。この歴史は、反官僚的言説が高度に集権化された企業組織としばしば併存するという、シリコンバレー文化の持続的な逆説も説明する。個人の自由は利用者インターフェースや起業神話で強調される一方、舞台裏のインフラ、資本統制、労働管理は一層集中することがある。したがってカウンターカルチャーの遺産は、プラットフォーム資本主義の単純な対立物ではなく、その文化的正統性の源泉の一つでもある。】
- Kittler, Friedrich. Optical Media. Translated by Anthony Enns. Polity, 2010. Originally published 2002. (Read: "Theoretical Presuppositions," pp. 29–46).
【注:Kittlerは、メディアの物理的条件と技術的な保存・伝送能力が文化形式を形づくることを強調する。プロトコルの政治に先立って読むことで、コードを純粋な抽象記号と考えることを避けられる。プロトコルは、ハードウェア、電波周波数、記憶装置、計算能力の上で作動しなければならない。彼のメディア決定論は批判的に読む必要があるが、まさにその極端な立場が、デジタル・メディア研究にデジタル権力の物質的基盤を真剣に扱わせる。KittlerのOptical Media、TurnerのFrom Counterculture to Cyberculture: Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianism、Barbrookの"The Californian Ideology"は、「標準、プロトコル、ソフトウェア、ハードウェア、制度的権力の層状の関係」を異なる知の伝統に置く。KittlerのOptical Mediaは、この層状の関係を過度に抽象的な時代診断から解放する。KittlerのOptical Mediaがこの問題にもたらす貢献は、理論を反証可能なものに保つ点にある。メディアの物質性の強調は、ソフトウェア政治の分析の方向も変える。「クラウド」「リアルタイム」「シームレスな体験」は、いずれも具体的な保存、伝送、表示、計算の装置に依存する。これらの装置が、どの情報を記録でき、どの速度で流通させ、どこで歪みを生むかを決める。Kittlerの強硬なメディア論は制度分析に取って代わることはできないが、批判がインターフェースと言説の層にとどまるのを防ぐことができる。】
- Barbrook, Richard, and Andy Cameron. "The Californian Ideology" Science as Culture 6, no. 1 (1996): 44–72.
【注:本論文は、デジタル技術文化が自由主義的個人主義、反官僚主義、技術進歩を結びつけるイデオロギー形態を批判する。Turnerと併読することで、歴史とイデオロギーの対照が可能となる。本論文が文化的想像力の再編成を論じるのに対し、Turnerは歴史的系譜を与える。"Californian Ideology"をシリコンバレーの万能説明とせず、そこにどのような現実の緊張が示されるかを考察すべきである。Barbrookの"The Californian Ideology"を軸にすると、「分散型トポロジーと政治的分権は同一ではない」という命題は、相互に関連する複数の環へと分解できる。この視点からは、その命題は制度的な権限付与と資源配分から切り離して成立しない。技術形態は組織的権力と結合して初めて、安定した社会的帰結を生む。本論文で最も保持すべきなのは、固定的な「シリコンバレー・イデオロギー」というラベルではなく、そこに示される政治的混合物である。反官僚主義、個人の解放、市場の自由は、技術的ユートピアの中で絡み合いうる。ゆえに、開かれたネットワークやパーソナル・コンピュータは、特定の左派的・右派的政治に生来対応するわけではない。その政治的方向は、所有権構造、労働関係、公共投資の語られ方と配分のされ方に左右される。"Californian Ideology"の批判は、公共投資が技術神話の中で消去される仕方にも関わる。インターネット、コンピュータ、研究体制は長く国家と大学の資源に依存してきたが、商業的成功は個人の起業と市場の自発性の勝利として語り直されがちである。この記憶の政治が、規制、課税、公共的責任に対するシリコンバレー特有の態度を形づくる一因となる。このイデオロギーが長く有効なのは、衝突する価値を一貫した物語へ包み直せるからでもある。個人の自治とベンチャー資本、反官僚主義と企業拡張、オープン文化と知的財産は、「イノベーション」の名の下に共存しうる。批判の課題は純粋な起源を探すことではなく、これらの矛盾が技術進歩の時間的物語によって暫定的に縫い合わされる仕方を分析することにある。】
- May, Timothy C.. "The Crypto Anarchist Manifesto" In High Noon on the Electronic Frontier, edited by Peter Ludlow, 41–44. MIT Press, 1996. Originally written in 1988 and circulated in the early cypherpunk milieu; read at the September 1992 founding meeting. (Read: entire manifesto, pp. 41–44).
【注:このテクストはまず歴史的文脈を示す資料であり、単独で規範判断を支えられる理論的権威ではない。Mayは強力な暗号、匿名通信、デジタル通貨、電子契約を、国家の規制、徴税、情報統制の能力を弱める技術的条件として想像する。同時に、サイファーパンク文化に顕著なリバタリアニズムと技術決定論の傾向も露呈させる。ここでの意義は、後にさまざまな陣営が繰り返し用いる「分散化」「プライバシー」「匿名性」「コードによる自治」といった語が、当初どのような政治的想像の中で結びつけられたかを知れる点にある。とくにGalloway、Lessigと比較する必要がある。技術的アーキテクチャが実際に規制能力を持つなら、暗号はなぜ国家監視に抵抗する道具としても、新たな私的権力のインフラとしても理解されうるのか。Mayの"The Crypto Anarchist Manifesto"、Barbrookの"The Californian Ideology"、Berners-Leeの"Information Management: A Proposal"は、ともに「技術設計、保守労働、統治の正統性のつながり」を巨視的判断から機構の層へ下ろす。Mayがこのつながりにもたらす貢献は、巨視的判断を識別可能な制度的機構へと押し戻すことにある。また、暗号技術の政治的意味が最初から一つではなかったことも示す。プライバシー技術は国家や企業に可能な可視性を制限できる一方、徴税、労働保護、公的規制の実施条件を弱めることもある。したがって暗号をめぐる議論は「国家―個人」の二項対立にとどまれず、プラットフォーム資本、金融仲介者、インフラ運営者、公共的責任を同時に分析へ組み込む必要がある。】
- Berners-Lee, Tim. "Information Management: A Proposal" CERN, 1989. (Read: entire proposal).
【注:この提案は完成されたWebユートピアの宣言ではなく、CERNの情報管理上の問題を解決するための工学的提案である。原初の設計文書として重要なのは、ハイパーテキスト、リンク、ネットワーク共有、開かれた情報の当初の意図が見えるからである。後のウォールド・ガーデン、プラットフォーム経済、クラウド・サービスと比較すれば、設計理念が商業化、標準化、資本蓄積の過程でどう転換するかを論じられる。Berners-Leeの"Information Management: A Proposal"から出発すると、「サイバネティクス、規律、管理社会、プロトコル統治の歴史的差異」は、より明確な因果の層を得る。工学的設計自体に固定的な政治的方向性はない。オープン、暗号化、分散型、モジュール化は、異なる制度では反対の結果をもたらしうる。その政治的意味を判断するには、標準策定権、変更権、資源所有権、障害対応権限へ戻る必要がある。この初期提案からプラットフォーム化したWebを振り返ると、「リンク」と「入口の統制」の歴史的分岐も見えてくる。開かれたハイパーテキストが異質な情報間の拡張可能な関連を重視するのに対し、プラットフォーム化はアカウント体系、推薦の論理、専有インターフェースにますます依存して可視性を組織する。この違いは、ネットワーク技術の政治的効果が、当初の工学的設計によってあらかじめ決まるのではなく、後に形成される商業的・制度的構造からしばしば生じることを示す。Berners-Leeの提案には、後にますます重要となる制度的問題も残されている。オープンなプロトコルは相互接続を保証できても、情報生態系が少数の入口へ再集中しないことまでは保証できない。Webの基礎標準は開かれたままであっても、検索、ソーシャルメディア、識別、クラウド・サービスはその上に新たなボトルネックを形成しうる。開かれた下位層と集中的な上位層の併存は、現代のインターネット政治を理解するうえで無視できない構造である。そのため、オープン標準とプラットフォーム集中の関係はWeb史を理解する重要な手掛かりになる。基底での相互接続は、上位の入口が再び独占を形成することを自動的には防がない。】
- Edwards, Paul N.. The Closed World: Computers and the Politics of Discourse in Cold War America. MIT Press, 1996. (Read: Introduction; selected chapters on command-control systems and Cold War computing).
【注:Edwardsは、冷戦期の計算、軍事的指揮統制システム、そしてより広い「閉じた世界」という政治的想像を、同じ歴史的枠組みで考察する。その価値は、すべてのデジタル技術を軍に帰因させることではなく、計算アーキテクチャ、シミュレーション、指揮系統、政治的物語が互いを形づくった仕方を示す点にある。本回にとっては、Wienerのサイバネティクスと後のネットワーク統治の間に不可欠な制度史的媒介を与え、「統制」をソフトウェア層の抽象的形式としてだけ扱わないよう促す。工学的設計自体に固定的な政治的方向性はない。Edwardsのいう閉じた世界は、冷戦期のコンピュータを軍事的指揮、シミュレーション、閉鎖システムの想像の中へ置く。計算は単なる道具ではなく、敵、環境、決定、統制可能な世界とは何かを形づくる政治的認識論にも参与する。そのためデジタル統治の歴史は、民生用インターネットだけからたどることはできない。リアルタイム情報、指揮統制、システム統合に対する軍事組織の要求も、計算インフラの形成に深く関与した。閉じた世界の閉鎖性は、知識生産の方式にも現れる。複雑な現実はシミュレーション、計算、指揮が可能なシステムへ変換され、意思決定者は「中心から全体を見渡す」想像を得る。今日のデータ・ダッシュボードとリアルタイム監視は、この認識論的衝動の一部を引き継ぐが、その組織構造は冷戦期の軍事システムとは異なる。比較の焦点は、システムの可視化が意思決定者の統制可能性への信念をどう形づくるかに置くべきである。この視点では、ダッシュボード、シミュレーション、リアルタイムの状況認識は、単に意思決定の効率を高めるだけでなく、「完全な情報」や「合理的統制」の基準を形成する。全体の可視化を強調するシステムほど、モデルに入らない地域的知識、曖昧さ、政治的争点をノイズとして扱う可能性がある。計算による統治の認識論的限界は、組織権力と直結している。このシステム視点は責任配分にも影響する。失敗が「情報不足」や「モデルの不完全さ」に帰されると、組織は目標設定そのものを省みるより、さらにデータを拡張しようとすることが多い。閉じた世界は、より大きな計算可能性を絶えず要求する統治の論理でもある。】
- Pickering, Andrew. The Cybernetic Brain: Sketches of Another Future. University of Chicago Press, 2010. (Read: Introduction; selected chapters on British cybernetics).
【注:Pickeringは、イギリスのサイバネティクスの伝統における具体的人物と実験を通じ、サイバネティクスの歴史には、集中管理、予測、最適化に向かう道しかなかったわけではないと強調する。遂行的、適応的な視点によって、未知の環境と絶えず結合する別の機械観を示す。Wiener、Benigerの後に読む場合、本書は「サイバネティクスが必然的に管理社会へ導く」という線形的系譜に歯止めをかける重要な役割を持つ。同じフィードバック、適応、システムの概念が、異なる制度や実践では大きく異なる政治的意味を持ちうるからである。PickeringのThe Cybernetic Brain: Sketches of Another Futureを通じ、「分散型トポロジーと政治的分権は同一ではない」という命題は、修辞から制度分析へと変えられる。工学的設計自体に固定的な政治的方向性はない。PickeringのThe Cybernetic Brain: Sketches of Another Futureでは、この命題は単一の結論を導くのではなく、異なる組織的場面における権力配置の比較を求め、それによって概念の説明範囲と失効条件を判断させる。イギリスのサイバネティクスの異端的伝統を通じ、Pickeringはサイバネティクスが必ずしも集中統制に向かわないことを示す。Ashby、Paskらの実験は、適応、創発、複雑な環境とともに踊ることを重視し、指揮統制とは異なる技術的想像を形成した。この違いによって「サイバネティクス」を単一の政治的系譜とみなすのを防ぎ、同じフィードバック概念が異なる工学実践と社会哲学へ導きうることを示せる。Pickeringの「遂行的」サイバネティクスは、別の技術政治も提供する。環境を完全に表象できると前提するのではなく、フィードバックによってシステムを未知の環境へ絶えず適応させるのである。この姿勢は本来民主的なわけではないが、「予測が正確であるほど統治はよくなる」というデータ主義への重要な反例となる。複雑性はつねに包括的なモデルを増やすことで解決しなければならないわけではなく、開かれた適応関係によって扱うこともできる。】
- Chun, Wendy Hui Kyong. Programmed Visions: Software and Memory. MIT Press, 2011. (Read: Introduction; Chapter 1, "On Software, or the Persistence of Visual Knowledge").
【注:Chunは、ソフトウェアがなぜ不可視でありながら、あたかも機械に直接命令できるものとして経験されるのかを問い、その実行可能性、インターフェース、記憶、統制の言説をメディア史に位置づける。これによって、プロトコルやコードを純粋な論理規則とみなす傾向を修正できる。コードの権力効果はコンパイル、ハードウェア、インターフェース、制度、利用者の実践に依存し、「プログラム可能性」も摩擦のない主権的命令ではない。工学的設計自体に固定的な政治的方向性はない。ChunのProgrammed Visions: Software and Memoryは最終的に、「技術設計、保守労働、統治の正統性のつながり」を、あらかじめ完成した時代診断ではなく、経験的研究によってさらに解きほぐせる問題に変える。Chunはとりわけソフトウェアの二重性を明らかにする。それは透明で実行可能な命令として示されながら、実際の作動には多くの不可視な抽象層と物質的過程を必要とする。この「目に見える統制感」と「不可視の技術的媒介」の間で、ソフトウェアは固有の統治効果を得る。デジタル政治にとってコードの分析は、何が書かれているかだけでなく、それがどうコンパイルされ、配備され、更新され、組織に埋め込まれるかを問うことでもある。ソフトウェアの「実行可能性」は、コードさえ書けば意図が摩擦なく実現するかのような直接性の幻想も生みやすい。実際には実行はつねにOS、ハードウェア、ライブラリ、ネットワーク、利用者環境を経由し、エラー、互換性、更新はシステムの日常である。Chunはこうしてソフトウェアを、抽象的命令から歴史的・物質的条件を持つメディアへと戻す。したがってソフトウェアは媒介の制度として理解すべきである。それは複雑なハードウェアを操作可能なインターフェースへ変え、同時に抽象層を通じて多数の実行条件を隠す。利用者の「クリックすれば起こる」という経験は、命令と結果の間に距離がないからではなく、絶えず保守される技術的連鎖から生じる。この隠された媒介性の中でこそ、ソフトウェアは日常生活への強力な統治効果を得る。】
第2回:アルゴリズム的統治性と生政治
本回は「規則はいかに行為を規定するか」から、「計算はいかに行為の可能性を組織するか」へ移る。Rouvroyのアルゴリズム的統治性は、主体の意味世界の再構築を前提とせず、データ収集、相関分析、確率的予測を通じて、行為が起きる前に分類、リスク、機会の構造を形成する。Cheney-Lippoldのアルゴリズム的アイデンティティは、主体が絶えず更新されるデータ・カテゴリーへ分割されうることをさらに示す。Bucherは「アルゴリズム」を単一で神秘的なブラックボックスと想像することを拒み、コード、訓練データ、プラットフォームの目標、工学実践、利用者行動、制度的環境がともにアルゴリズム権力を構成すると強調する。したがって計算可能な主体とは、機械によって完全に「知られた」人ではなく、異なる制度的場面で再コード化され、集団化され、順位づけられる行為者である。
アルゴリズムの政治的効力も、モデルの性能だけでは説明できない。確率的出力が統治の一部となるのは、融資承認、福祉給付の資格、国境審査、採用選考、コンテンツ配信、保険料設定などの組織的手続きにおいて、行為への帰結を与えられるときである。したがって予測、説明、判断、正当化は区別しなければならない。統計的相関は因果的説明ではなく、モデルが示すリスク・スコアも、法的・倫理的理由を自動的に担うことはできない。Amooreの不確実性と確率判断の分析、Eubanksの福祉自動化の経験的研究、Pasqualeのブラックボックスと説明責任の議論は、いずれも同じ問いを指す。技術的出力が機関内部で、現実的な拘束力を持つ決定へどのように翻訳されるのか、という問いである。
本回はまた、「アルゴリズムの偏り」を漠然とした価値判断から制度的連鎖へと戻す。訓練データ、ラベル体系、目的変数、評価指標、閾値設定、配備環境、人間による再審査、不服申立ての経路は、いずれも不平等を生み出し拡大する環となりうる。分類は世界をただ記述するのではなく、組織に採用されると、機会構造と社会的アイデンティティの形成に参与する。ゆえにアルゴリズム的統治性は、知識の問題であると同時に、行政手続き、組織責任、権利保障の問題でもある。計算可能性の拡大は政治的判断を消し去ったのではなく、多くの場面で、判断を隠し、配分し、その責任を問う仕方を変えたのである。
必読文献:
- Rouvroy, Antoinette. "The End(s) of Critique: Data Behaviourism versus Due Process" In Privacy, Due Process and the Computational Turn, edited by Mireille Hildebrandt and Katja de Vries, 143–167. Routledge, 2013.
【注:本章は本回の概念的核心である。Rouvroyのアルゴリズム的統治性は、主体の意味、動機、意図をまず理解する必要がなく、データ収集、相関分析、予測によって行為以前に介入する統治を強調する。鍵となるのは、相関と因果、予測と正当化、プロファイリングと判断の区別である。アルゴリズムは「あなたが本当は誰なのか」を知る必要はなく、ある分類が行為を支えるのに十分だと制度に確率的に信じさせればよい。こうして初めて、アルゴリズム統治と従来の法的手続きとの真の緊張が理解できる。より細かく言えば、適正手続きとの衝突は「機械に人間性がない」ことではなく、データ行動主義が、理由を説明し告発に応答できる法的・政治的人格としての主体を、確率的な行動輪郭へと変えがちな点にある。相関分析は因果説明を確立せずに予防的介入を支えうるため、批判そのものが困難に直面する。システムがもはやあなたは誰か、なぜそうなのかを問わず、過去のデータに基づいてあなたが遭遇する環境を調整するだけなら、イデオロギーの暴露や誤った表象の訂正を中心とする従来の批判で十分なのだろうか。したがって本章はYeung、Eubanks、Amooreと併読し、規制、福祉行政、安全保障上の判断という各観点から、アルゴリズム的統治性が異なる制度で同じように成立するかを検証すべきである。Rouvroyのデータ行動主義がアルゴリズム的統治性の核心をなすのは、従来の主体の意図、物語、規範判断を迂回し、大量のデータの相関から直接行動パターンを発見しようとするからである。統治は主体が表明する以前に、確率的にその機会空間を形づくりうる。ここで本質的な法理の問題が生じる。相関は予測を支えられるが、理由を自動的には与えない。手続的正義が求めるのは、正確な結果だけでなく、争いうる規範的根拠でもある。】
- Cheney-Lippold, John. "A New Algorithmic Identity: Soft Biopolitics and the Modulation of Control" Theory, Culture & Society 28, no. 6 (2011): 164–181.
【注:Cheney-Lippoldは、デジタル・システムが行動データから動的カテゴリーを生産する仕組みを説明するために、アルゴリズム的アイデンティティを提起する。Deleuzeとの併読に適している。アイデンティティは主体の自己申告や固定的な人口統計属性だけで決まるのではなく、システムが行動に基づいて絶えず再計算する確率的対象にもなる。その限界も明確である。アルゴリズム分類は依然として社会の既存カテゴリーを材料としており、歴史から離れた純粋に数学的な分類ではない。アルゴリズム的アイデンティティと従来の人口分類の連続性は、用語の対応にとどめず、カテゴリーの由来、制度的用途、更新機構によって具体化すべきである。計算可能な主体はモデルに完全に知られた個人ではなく、異なる制度的場面でスコア、確率、カテゴリー、行動履歴へ分割される対象である。分類が福祉、信用、採用、コンテンツ配信に入り込むと、逆に機会構造と主体の行為を形づくる。アルゴリズム的アイデンティティは、安定的属性だったアイデンティティを、更新され続ける確率的分類へ変える。システムは「本当のあなた」が誰かを確認する必要がなく、行動の軌跡から個人を暫定的に操作可能なカテゴリーへ入れ、それに基づいて広告、価格、リスク、可視性を決めればよい。Cheney-Lippoldはこうしてアイデンティティ政治の新たな層を明らかにする。主体は自分がどのカテゴリーに入れられたかを知らない場合があり、カテゴリーの定義や修正に直接参加することはさらに難しい。動的分類にはもう一つ重要な帰結がある。アイデンティティは登録時だけに確認されるのではなく、クリック、移動、消費のたびに再推論されうる。主体が対面するのは固定的なファイルではなく、絶えず変化し、相互に矛盾しうる一連の確率的アイデンティティである。そのため不服申立てはとりわけ困難となる。影響を受ける者は、どの推論がいつ異なる扱いを引き起こしたかさえ知らないかもしれない。アルゴリズム的アイデンティティの政治的困難は、したがって「誤判定される」ことだけでなく、自分を記述するカテゴリーや、それが場面を横断して流通する仕方を主体が知りにくい点にもある。】
- Bucher, Taina. If...Then: Algorithmic Power and Politics. Oxford University Press, 2018. (Read: Chapter 2, "The Multiplicity of Algorithms," pp. 19–40; Chapter 3, "Neither Black nor Box," pp. 41–65).
【注:Bucherの重要な貢献は、アルゴリズム権力を、神秘的で統一された不可視の「ブラックボックス」と想像することを拒む点にある。アルゴリズム権力は、コード、データ、プラットフォームの目標、利用者行動、工学実践、制度の間に分布する。これによってRouvroyの巨視的な統治概念を具体的システムに位置づけられる。私たちが直面するのは単一の行為者としての「アルゴリズム」ではなく、異なるソフトウェア、制度、実践の集まりである。そのため「アルゴリズムの透明性」も定義し直さなければならない。コードの透明性は統治の透明性を自動的には意味しない。予測システムは、完全なアイデンティティ理論を構築せずとも、十分に操作可能な代理変数を作ればよい場合が多い。ここにこそ、アルゴリズム的アイデンティティと従来のアイデンティティ政治が関連しつつ同一視できない理由がある。BucherのIf...Then: Algorithmic Power and Politicsは、こうして「量化、データ生産、アルゴリズム・システムをつなぐ制度的連鎖」に、限定的だが有効な説明尺度を提供する。それは機構を指摘できるが、組織、法、物質的条件の追加分析に取って代わるわけではない。Bucherはアルゴリズムを単一のブラックボックス的主体とみなさず、コード、データ、商業的目標、利用者実践、組織制度によって構成されると強調する。したがってアルゴリズム権力は関係的権力であり、順位づけの論理だけでなく、技術チームの選択、プラットフォーム指標、利用者のシステムへの適応にも存在する。透明性をソースコード公開だけと理解すれば、この複雑な社会技術的アセンブレッジを捉えられない。アルゴリズムをめぐる想像力へのBucherの注目は、権力がシステムの実際の働きだけでなく、人々がそれに可能だと信じることにも存在すると示す。利用者は想像上の推薦規則に合わせて表現を調整し、機関は「アルゴリズムはより客観的だ」という信念から自動化の導入を拡大することがある。アルゴリズムの社会的効力は、それを取り巻く期待と不確実性からも部分的に生じる。】
- Amoore, Louise. Cloud Ethics: Algorithms and the Attributes of Ourselves and Others. Duke University Press, 2020. (Read: Chapter 1, pp. 1–24).
【注:Amooreは、確率的判断が国境、安全保障、リスク統治において行為を生む効力を得る仕組みに注目する。重要なのは、アルゴリズムが「正確に予測」できるかではなく、確率的出力がいつ機関によって現実の決定へ変換され、その変換が責任と不確実性の政治をどう変えるかである。透明性も唯一の解決策ではない。本回では、説明可能性を、より広い適正手続きと制度的な説明責任の中で理解する。AmooreのCloud Ethics: Algorithms and the Attributes of Ourselves and Othersを本回全体の文脈に置くと、「透明性、説明、不服申立て、責任が自動化された決定でどう再配分されるか」は、より具体的な構造を示す。計算可能な主体はモデルに完全に知られた個人ではなく、異なる制度的場面でスコア、確率、カテゴリー、行動履歴へ分割される対象である。分類が福祉、信用、採用、コンテンツ配信に入り込むと、逆に機会構造と主体の行為を形づくる。Amooreはアルゴリズム倫理の焦点を「モデルは正しいか」から、確率的判断がどう行為へ移されるかへと移す。国境、安全保障、リスクのシステムは不確実性の下で作動することが多い。鍵は不確実性の除去ではなく、機関が閾値をどう選び、特定の属性に高い重みを与え、局所的な相関に現実の帰結を持たせるかにある。ゆえに責任をモデルへ移すことはできない。不確実性の制度的な配分自体が政治的決定である。Amooreはこうして倫理の問題を「属性付与」そのものへ進める。システムは部分的な痕跡から人や出来事の潜在的性質を推論し、それに応じて注意、疑い、資源を前もって配分する。確率は中立的な待機状態ではなく、行為の理由になりうる。リスク統治の政治性は、不確実性の代償を誰が負わされ、どの誤りを制度が許容可能とみなすかにある。】
推薦文献:
- Gillespie, Tarleton. "The Relevance of Algorithms" In Media Technologies, 167–194. MIT Press, 2014.
【注:Gillespieは、「関連性」そのものがプラットフォームの最も重要な公共的決定の一つとなる仕方を論じる。検索、推薦、順位づけは情報を計算するだけでなく、何が見られるに値し、何が重要とされるべきかを決める。これはアルゴリズム的統治性に、メディアと公共文化の次元を加える。理論的類似は制度的同一性を自動的には意味しない。予測の正確さ、統計的相関、行政的正当性は異なる層に属する。リスク・スコアが本当に統治となるのは、閾値、人間による再審査、不服申立て、執行手続きの中で帰結を得るときである。「関連性」はしたがって純粋に技術的ではなく、規範的な性質を持つ。プラットフォームのいう関連性は、通常クリック確率、滞在時間、商業的目標、コンテンツの価値を一つの順位づけ過程に圧縮する。アルゴリズムは真理を直接裁定しないが、注意の配分を通じて公共的議題の出現条件を変えられる。Gillespieはこうしてアルゴリズム研究をモデルの内部からプラットフォーム制度へ押し出す。説明すべきなのは、関連性の基準がどう設定され、調整され、正当性を得るかである。推薦システムの公共性は、それが「共通世界」の可視的構造を変える点にも現れる。利用者ごとに高度に個人化された順位づけが与えられると、公共的議題の入口は同じではなくなる。一方、プラットフォームは推薦目標やコンテンツの境界を調整する権限を保持する。ゆえにアルゴリズム的関連性の研究は、順位づけの正確さの測定だけでなく、メディア制度とビジネスモデルを結びつけなければならない。プラットフォームの順位づけは情報処理であると同時に注意の配分であり、関連性が商業目標の関数となるとき、メディア権力はアルゴリズム設計の核心に入る。Gillespieはさらに、その規範性が「利用者の需要」や「関連性」の名で現れ、価値判断を技術化しがちな点を指摘する。何を推薦に値するとし、何をスパムとし、どのコンテンツの順位を下げるかは、いずれも一定の公共秩序を前提とする。こうした分類の背後の制度目標を無視すれば、アルゴリズム研究はプラットフォームの最重要の政治的判断を、純粋な工学的最適化と誤って記述してしまう。】
- Eubanks, Virginia. Automating Inequality. St. Martin's Press, 2018. (Read: Introduction, pp. 1–14; Chapter 1, pp. 17–36).
【注:本書は抽象的なアルゴリズム統治を、福祉、貧困、行政システムへと具体化する。核心は、アルゴリズムに「偏見がある」と証明することではなく、社会資源の配分が技術システムを通じてどう再編され、歴史的不平等がリスク分類にどう入るかを観察する点にある。これによって分析を「プラットフォーム体験」から国家行政と社会政策へ移せる。EubanksのAutomating Inequalityが与えるのは、「組織が確率的出力を現実の帰結を持つ判断へどう翻訳するか」に対する最終的な命名ではなく、競合する説明によって検証できる理論的切り口である。この開放性そのものが研究上の価値をなす。EubanksのAutomating Inequalityはまた、自動化統治がまず退出能力を最も欠く人々に及ぶことを示す。技術システムが資源不足、福祉審査、リスク管理を効率の問題へ包み直すと、貧困自体が疑わしさと計算可能なリスクに変換されやすい。Eubanksの事例は、福祉の自動化システムが人の判断を機械の判断へ単純に置き換えるのではなく、資源削減、行政改革、貧困統治の既存枠組みの中で選別を再編することを示す。技術は審査を速められる一方、歴史的に貧困層へ向けられてきた疑念と懲罰をデータ規則へ固定することもある。したがってアルゴリズムの公平性は福祉国家の制度目標から切り離せない。苛酷な政策を正確に実施するシステムも、深刻な不正義を生みうる。Eubanksの経験的資料は、技術改革に常見される政治的不均衡も示す。行政機関は費用削減と不正防止を目標にシステムを導入する一方、審査される者がデータの誤り、複雑な手続き、自動拒否に伴う時間的費用を負う。効率を機関側だけから計算すれば、弱い立場の人々の追加労働は業績評価から除かれる。ゆえに自動化統治の評価尺度自体を政治化する必要がある。さらに、貧困統治の自動化は個別の誤りを生むだけでなく、国家と福祉申請者の関係を変えうる。対面での説明や個別事例への裁量がデータベース照合とリスク評価へ置き換えられると、申請者は自分の資格を証明する責任をより多く負う。手続きのデジタル化は、こうして行政負担を機関内部から、資源を最も持たない人々へ移す可能性がある。】
- Pasquale, Frank. The Black Box Society. Harvard University Press, 2015. (Read: Introduction; Chapter 2, "The Digital Poorhouse," pp. 35–58).
【注:Pasqualeの焦点は、自動化された決定における情報の非対称性と説明責任にある。そのため「ソースコードの公開」が十分な答えかを検証できる。コードを公開していても、データ、モデルの複雑さ、営業秘密、組織上の権限のために、システムへの責任追及がなお不可能な場合がある。「量化、データ生産、アルゴリズム・システムをつなぐ制度的連鎖」は、PasqualeのThe Black Box Societyを通じて組織と制度の層へ入る。PasqualeのThe Black Box Societyの尺度では、この連鎖は連続性と変形の双方を扱う必要がある。旧制度が新技術を通じて新たな作動方法を得る場合も、新技術が既存制度へ再吸収される場合もある。PasqualeのThe Black Box Societyはまた、この制度的連鎖の分析には対象の違いと制度的境界を保持する必要があり、概念の修辞的な強さを経験的説明の代わりにしてはならないと示唆する。PasqualeのThe Black Box Societyは不可視性を技術的ブラックボックスから制度的ブラックボックスへ広げる。営業秘密、データ支配、専門性の障壁、組織上の権限が、誰にシステムを説明できるかをともに決める。ゆえに説明可能性は、監査、不服申立て、責任帰属と結びついて初めて、実際の説明責任へ転化しうる。Pasqualeのブラックボックスは、技術的複雑性、営業秘密、制度的な情報非対称性を同時に含む。金融のスコアリング、検索順位、プラットフォームの決定の不可視性は、すべて「モデルの説明」で解決できるわけではない。影響を受ける者は、データへのアクセス、不服申立ての経路、交渉力も欠く場合が多い。したがって説明責任は情報開示から組織責任へ拡張し、誰が決定権を持ち、誰が理由を示す義務を負い、誰が誤りを訂正できるかを明らかにしなければならない。】
- Harcourt, Bernard E.. Exposed: Desire and Disobedience in the Digital Age. Harvard University Press, 2015. (Read: Chapter 1, "The Expository Society," pp. 1–28).
【注:Harcourtは曝露、欲望、監視を一緒に論じる。主体の自己提示を「自ら監視されることを望む」と単純に説明するのではなく、プラットフォーム設計、社会規範、可視性への欲望、商業的データ抽出が相互に強化される仕方を分析すべきである。透明性も唯一の解決策ではない。HarcourtのExposed: Desire and Disobedience in the Digital Ageが示す「透明性、説明、不服申立て、責任が自動化された決定でどう再配分されるか」という問いは、テクスト内の概念論争にとどまらない。HarcourtのExposed: Desire and Disobedience in the Digital Ageは、この問いを特定の対象と因果連鎖に限定できるようにする。それを超える類比には新たな証拠が必要であり、用語だけに頼っては成立しない。Harcourtの「曝露」は、監視研究の重要な転回も示す。主体はつねに受動的対象として存在するのではなく、社交、消費、自己提示の中で可視性を能動的に生産する。問題は、その能動性が強制を消すかどうかではなく、欲望が商業的アーキテクチャに捕捉され、非対称的なデータ・アクセス権の中で予測と統治の資源へ変えられる仕方にある。Harcourtの議論によって、「曝露」は単なるプライバシーの喪失ではなく、社会関係となる。主体は共有、検索、位置情報、消費を通じて自らを生産し続け、プラットフォームはその表現を計算可能な資源へ再変換する。したがって欲望と監視は対立する力ではない。デジタル環境が見られ、つながり、表現したいという欲求を満たすからこそ、データ抽出は参加と利便性の形で日常生活へ深く入り込める。曝露社会の逆説は、主体が表現と接続の自由を得ながら、相互対等には観察できないプラットフォームへ生活の痕跡をますます引き渡す点にある。自由と監視は単純な増減の対立ではなく、同じ相互作用のアーキテクチャの中でともに拡大しうる。】
- Beer, David. The Data Gaze: Capitalism, Power and Perception. Sage, 2018. (Read: Introduction and Chapters 1–2).
【注:Beerはデータのまなざしという概念によって、データ化が「見ること」と「見られること」の関係をどう変えるかを論じる。これはアルゴリズム分類と社会的知覚を結びつけるのに適している。生活がますますデータ化されると、私たちを「見る」のは機関だけではなく、人々も指標を通じて自分を理解するようになる。BeerのThe Data Gaze: Capitalism, Power and Perceptionの分析尺度では、「分類、予測、スコアリング、順位づけが計算可能な主体をどう構成するか」は、まず具体的な関係群として現れる。いわゆるアルゴリズムの偏りは価値判断にとどめてはならない。訓練データ、ラベル体系、目的変数、評価指標、配備環境、既存の制度的不平等は、それぞれ異なる問題を生みうる。層ごとに追跡して初めて、偏りの発生箇所を判断できる。データのまなざしは、機関が人を見つめることだけでなく、主体が指標によって自分を捉え直す知覚の機構でもある。歩数、信用スコア、人気度、順位、業績データは生活を比較可能な数量へ翻訳し、逆に行為の目標を形づくる。Beerはこうしてアルゴリズム統治を量化された主体化と結びつける。データは現実を記述するだけでなく、どの現実を追求すべきかの規定にも参与する。指標化された自己は、強制だけから生じるわけではない。人々は順位表、健康データ、社会的フィードバックを進んで使って生活を組織するが、その自発的利用はプラットフォームがあらかじめ定めた測定可能な次元の中で行われる。主体性には自己量化への欲望が含まれると同時に、指標のアーキテクチャによる誘導も働く。Beerの分析により、権力は外的なまなざしにとどまらず、日常的な自己評価の技術的形式へ入る。この自己量化は、データ権力が外部の監視や強制的収集だけではなく、主体の能動的参加によってしばしば実現されることをさらに示す。指標の力は、測定からだけでなく、人々が指標を中心に自分の行為を組み替えることからも生じる。この自己調整の中でこそ、データのまなざしは主体形成へ入り込む。】
- Yeung, Karen. "Algorithmic Regulation: A Critical Interrogation" Regulation & Governance 12, no. 4 (2018): 505–523.
【注:本論文は、制度と法の支配の観点からの修正を与える。アルゴリズムによる規制は技術的効率の向上だけではなく、機械が規則をどう実行し、監視し、動的に調整するかに関わる。RouvroyとLessigの間に、より明確な理論的橋渡しを行える。Yeungの"Algorithmic Regulation: A Critical Interrogation"、BeerのThe Data Gaze: Capitalism, Power and Perception、Seaverの"Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems"は、「組織が確率的出力を現実の帰結を持つ判断へどう翻訳するか」を異なる知の伝統に置く。Yeungの論文は、この問題を過度に抽象的な時代診断から解放する。いわゆるアルゴリズムの偏りは価値判断にとどめてはならない。Yeungの貢献は、理論を反証可能なものに保つ点にある。その制度的視点はさらに、アルゴリズム規制の鍵が単なる自動化の程度ではなく、フィードバックの循環が規則の執行をどう変えるかにあると示す。継続的監視によって統治は事後的処罰からリアルタイム調整へ移れるが、その動的性質は手続的審査や安定した予期を圧縮することもある。したがって効率、適応性と、法の支配が求める予見可能性、説明可能性の間には、現実の制度的緊張が存在する。】
- Seaver, Nick. "Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems" Big Data & Society 4, no. 2 (2017). (Read: entire article).
【注:Seaverは、アルゴリズムをコードから直接読み取れる技術的対象としてだけ扱うことに反対し、技術者の分類、組織文化、利用者モデル、データ実践、保守作業が共同で構成する文化的対象とみなすよう提案する。本回にとってこの点はきわめて重要である。アルゴリズム的統治性が予測と分類に関わるなら、研究者はモデルの数学的作動だけでなく、「関連性」「利用者」「異常」「成功」といったカテゴリーが組織内部でどう定義されるかも問わなければならない。Seaverの"Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems"を軸にすると、「量化、データ生産、アルゴリズム・システムをつなぐ制度的連鎖」は、相互に関連する複数の環へ分解できる。いわゆるアルゴリズムの偏りは価値判断にとどめてはならない。Seaverの視点では、この制度的連鎖は制度的な権限付与と資源配分から切り離して成立しない。技術形態は組織的権力と結びついて初めて、安定した社会的帰結を生む。Seaverの文化人類学的アプローチは、アルゴリズムを「独立して動くコード」から組織生活へ戻す。技術者は利用者をどう理解するか、何を成功の指標とするか、チームは例外的状況をどう扱うか。こうした文化的実践がすべてアルゴリズム・システムの構成に参与する。したがってアルゴリズムの民族誌の意義は、秘密のコード保管庫へ入ることではなく、技術的対象が日々の仕事、物語、制度的慣行を通じて持続的に制作される仕方を観察する点にある。】
- Fourcade, Marion, and Kieran Healy. "Classification Situations: Life-Chances in the Neoliberal Era" Accounting, Organizations and Society 38, no. 8 (2013): 559–572. (Read: entire article).
【注:FourcadeとHealyは、スコアリング、順位づけ、分類を社会階層研究へ戻し、現代の分類システムは中立的な記述ではなく、個人に与えられる機会、価格、資源、制度的扱いを変えると強調する。「分類状況(classification situation)」は、アルゴリズム的アイデンティティとの併読にとくに適している。計算される主体はデータ上の表象であるだけでなく、分類結果に形づくられる現実の機会構造の中にいる。Fourcadeの"Classification Situations: Life-Chances in the Neoliberal Era"、Seaverの"Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems"、Espelandの"A Sociology of Quantification"は、ともに「透明性、説明、不服申立て、責任が自動化された決定でどう再配分されるか」を、巨視的判断から機構の層へ移す。透明性も唯一の解決策ではない。Fourcadeの貢献は、巨視的判断を識別可能な制度的機構へ押し戻す点にある。いわゆるアルゴリズムの偏りは価値判断にとどめてはならない。"Classification Situations: Life-Chances in the Neoliberal Era"は、この透明性、説明、不服申立て、責任の再配分を歴史的条件の中へ戻す。似たデジタル上の現れが必ずしも同じ機構から生じるとは限らず、概念の有効性も差異の保持に依存する。FourcadeとHealyの分類状況は、分類と生活機会を直接結びつける。信用、保険、教育、消費における分類は、社会的差異を記述するだけでなく、金利、機会、待遇、可視性を決定し、フィードバック効果を生む。デジタル化が強めるのは分類という行為自体ではなく、分類がより細かな粒度、より高い頻度、より低い可視性で日常的制度へ組み込まれる能力である。】
- Espeland, Wendy Nelson, and Mitchell L. Stevens. "A Sociology of Quantification" European Journal of Sociology 49, no. 3 (2008): 401–436. (Read: entire article).
【注:EspelandとStevensは量化研究の方法論的基準を与える。数字、指標、順位は複雑性を圧縮できる一方、共通尺度化(commensuration)によって、もともと直接比較できなかった対象を同じ尺度に置く。本論文をアルゴリズム統治に導入するのは、すべてのアルゴリズムが単なる「採点」だというためではない。予測モデルが制度的決定へ入る前に、対象を量化可能にし、指標を比較可能にし、結果を順位づけ可能にする社会的工学がしばしば先行することを見るためである。Espelandの"A Sociology of Quantification"から出発すると、「分類、予測、スコアリング、順位づけが計算可能な主体をどう構成するか」は、より明確な因果の層を得る。透明性は自動化統治の唯一の解決策ではない。共通尺度化が完了すると、政治的論争は指標設計の問題へ書き換えられがちである。大学ランキング、信用スコア、リスク・モデルが権力を持つのは、数字自体に強制力があるからではなく、組織がそれに従って機会と資源を配分し始めるからである。量化の社会学は、したがって重要な道筋を提供する。分類基準が制度に入った後で行動をどう変え、その変化が逆に分類自体を安定させるかを追跡するのである。EspelandとStevensは反応性(reactivity)も強調する。順位づけられ測定される者は、評価体系に合わせて行動を調整する。指標が目標になると、もともと測定しようとした対象を変えてしまいうる。この機構はアルゴリズム統治にとりわけ重要である。予測モデルの配備が人々の戦略を変え、ひいては将来のデータを変えるからである。量化は受動的記録ではなく、制度的場面を再形成できる社会的過程である。指標が行為の目標になると、量化体系は量化される対象自体を変える。これが順位づけとスコアリングが制度的フィードバックを生む根本理由である。共通尺度化と反応性は合わさって、量化の制度が継続的に自己強化する基本機構を構成する。】
- O'Neil, Cathy. Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. Crown, 2016. (Read: Introduction; selected case studies on education, employment, credit, and policing).
【注:O'Neilは一連の制度的事例によって、大規模な採点モデルが、不透明性、閉じたフィードバック回路、影響の大きい場面で不平等を拡大する仕方を示す。これはアルゴリズム理論の終点ではないが、概念の耐久性を試すのにきわめて適している。モデルを入学選考、雇用、信用、警察活動などの制度的過程へ戻して初めて、誤りがどう配分され、不服申立ての経路が存在するか、量的判断がどう事実上の強制力を得るかが見える。透明性は自動化統治の唯一の解決策ではない。O'NeilのいうWMDは、モデルの規模、不透明性、フィードバック性が合わさって被害を拡大する点を強調する。この一般向けの概念は、より厳密なアルゴリズム研究と組み合わせる必要があるが、制度的配備の核心を捉えている。教育、雇用、司法などの高リスク領域へモデルが入ると、誤りは統計上の偏差にとどまらず、人の機会を変え、その結果を将来のデータへ書き戻す。ゆえにモデル評価には、オフラインの正確さだけでなく社会的フィードバックを含めなければならない。このフィードバック性によってWMD概念は、社会科学の遂行性(performativity)と接続できる。モデルの判断は教育資源、信用機会、警察の注目を変え、その結果が次の訓練の「現実のデータ」とされうる。この循環を断たなければ、歴史的不平等はモデルの性能という形で新たな経験的正当性を得る。したがってモデルの評価では予測の正確さだけでなく、配備された予測が新たなデータを生み、既存の差異を自己証明し続けるかも調べる必要がある。WMDの枠組みは説明責任の難題も提起する。モデルが客観的道具とみなされると、機関は不評な決定を「データの結果」と説明し、政策上の選択を曖昧にしがちである。どの変数を選び、どの誤差を許し、人の上書きを認めるかは、なお制度的決定である。自動化は判断を消したのではなく、一部の判断をあらかじめモデルと過程へ書き込んだ。したがって機関はモデルへの責任を、開発者が一度で果たす技術的責任と考えてはならない。配備、更新、閾値調整、人による上書きは、システムの帰結を継続的に変えるため、責任追及は全ライフサイクルを覆うべきである。】
- Kitchin, Rob. The Data Revolution: Big Data, Open Data, Data Infrastructures and Their Consequences. Sage, 2014. (Read: Introduction; chapters on data infrastructures and data assemblages).
【注:Kitchinはビッグデータを「より多くのデータ」という宣伝文句から、データのアセンブレッジへ戻す。収集装置、標準、データベース、分析道具、組織目標、制度的実践が、データの生産のされ方を共同で定める。これはRouvroyの理論で抽象化されがちなインフラの環を補う。いわゆるデータ行動主義が向き合うのは自然に存在するデータではなく、選択、洗浄、形式化、制度化を経たデータ対象である。KitchinのThe Data Revolution: Big Data, Open Data, Data Infrastructures and Their Consequencesを通じ、「量化、データ生産、アルゴリズム・システムをつなぐ制度的連鎖」は、修辞的命題から制度分析へ転換できる。透明性は自動化統治の唯一の解決策ではない。データ・インフラの分析によって、「ビッグデータ」を自然資源のように自ら存在するものと想像せずに済む。データは収集プロトコル、データベース構造、洗浄基準、メタデータ、機関の目標を経て初めて使用可能な対象になる。ゆえに「データ駆動型」の決定は、最初から選択と排除を含む。モデルの出力を理解するには、データがどう生産され、分類され、保守されるかへ遡る必要がある。Kitchinはまた、データの開放とデータの権力を区別する。技術的に開かれていても、形式、計算資源、専門能力、制度上の権限のため、データを実際に使えるのは少数の主体だけという場合がある。オープンデータの政治は、ファイル公開にとどまらず、誰が分析し、結合し、意思決定の資源へ変える能力を持つかを扱わなければならない。データ・インフラ自体が、不平等な能力の配分機構なのである。その統治には標準の互換性と長期保守も含まれる。異なる機関がデータを理解し再利用できるかは、しばしばメタデータ、形式、継続的更新に依存する。こうした可視性の低い仕事が、いわゆる「データ資産」の成立そのものを左右する。】
- Rieder, Bernhard. Engines of Order: A Mechanology of Algorithmic Techniques. Amsterdam University Press, 2020. (Read: Introduction; selected chapters on ranking, recommendation, and algorithmic techniques).
【注:Riederの機械論(mechanology)的アプローチは、過度に広い「アルゴリズム」という名詞から、順位づけ、推薦、分類といった具体的な技術機構とその歴史へ注意を移す。Bucherの後に置けば、より細かな粒度の批判を訓練できる。異なるアルゴリズム技術には異なる入力、目的関数、評価法、組織上の用途があり、すべてを遍在する一つの「アルゴリズム権力」と書くことはできない。透明性も唯一の解決策ではない。透明性は自動化統治の唯一の解決策ではない。Riederの細粒度の方法は、「アルゴリズム」という語の一般化に抵抗するのにとくに適している。順位づけ、分類、推薦は異なる問題設定、技術的伝統、評価指標を持ち、組織内での権力効果も異なる。モデルの型と具体的な制度用途を対応させて初めて、アルゴリズム批判は巨視的診断から検証可能な機構分析へ進める。機械論はまた、最新モデルだけを追うのではなく、アルゴリズム技術を道具の系譜へ戻すよう求める。順位づけ、照合、クラスタリング、推薦はいずれも長い歴史を持ち、異なる工学上の常識を形成してきた。技術的系譜を理解すれば、どの統治効果が機械学習の新能力から生じ、どれが以前からの情報検索、統計分類、データベース管理に由来するかを区別できる。この技術的分解は、統治にもより正確な対象を与える。異なる順位づけ、照合、分類の機構には、異なる透明性、監査、責任の要件を課すべきである。Riederが強調するアルゴリズム技法は、監査の方法も変えられる。抽象的な「アルゴリズム」全体の透明性を求めるより、入力データ、順位関数、フィードバック信号、組織目標を別々に分析できる。この分解は責任を具体的な環へ対応させ、批判を「ブラックボックスは知りえない」と「コードを公開すれば解決する」の二者択一から解放する。】
第2部:抽出、植民地性、インフラ(第3–4週)
第3回:監視、抽出、データの植民地性
本節では、アルゴリズムによる統治を、さらにデータ生産、資源抽出、歴史的不平等のなかに位置づける。Zuboff の監視資本主義は行動データが予測資産に変わる過程を強調し、Couldry と Mejias のデータ植民地主義はデータ関係を、より長い植民地的領有の歴史に置く。Browne は奴隷制、人種化された識別、監視技術の歴史的連続性から出発し、分類システムに差異化された可視性と曝露がどう組み込まれているかを示す。これらは等価ではない。資本蓄積、植民地的領有、人種化された統治、生政治は、それぞれ異なる歴史的メカニズムを持つ。概念を並置する意義は、データ経済について各々が何を説明できるのか、また、どの時点で過剰な類推が差異を覆い隠すのかを検証することにある。
データ抽出もまた、単に「利用者の同意」の後に生じる交換過程ではない。プラットフォームは、本人確認システム、初期設定、追跡インフラ、広告市場、データ仲介業者、ネットワーク効果を通じて、日常の行動を、保存・関連づけ・予測が可能な記録へと絶えず変換する。記録され、識別され、排除される際に各集団が負うリスクは対称ではない。生体認証、信用スコア、ターゲティング広告、行政システムの分類は、既存の社会的差異を技術過程へ持ち込む。Noble、Benjamin、Magnet、Dencik らは、それぞれ情報検索、人種の技術、本人識別、データ正義からこの問題に切り込む。それにより「データの偏り」は単なるモデルの誤差ではなく、社会的分類、資源配分、権力関係の一部として捉えられる。
ここで Mbembe を用いるのは、「アカウント停止はデジタルな死である」といった安易な類推を作るためではない。植民地性、敵意、境界、差異化された曝露を理解する政治的背景を得るためである。データの植民地性という分析が成り立つのは、強制的依存、領有関係、知の権威、法的非対称性、退出費用を具体的に説明できる場合に限られる。したがって本節が目指すのは、デジタル社会に最も衝撃的な歴史的名称を与えることではない。データ化が観察、抽出、分類、統治をどのようにつなぎ、これらのメカニズムが異なる制度的場面で、以前の権力構造をいかに継承または再編するのかを明らかにすることである。
必読文献:
- Zuboff, Shoshana. The Age of Surveillance Capitalism. PublicAffairs, 2019. (Read: Chapter 3, "The Discovery of Behavioral Surplus," pp. 63–96).
【注:Zuboff の「行動余剰」概念は、データ統治と資本蓄積を結びつける。利用者の行動、データ抽出、予測商品、市場化の過程をどう区別するかに注目して読む。したがって、このモデルがデジタル経済全体に適用できるのか、それとも主に Google や広告プラットフォーム型のデータ抽出を説明するのかを検証しなければならない。第3章では、行動余剰の論証の連鎖をすべて分解する必要がある。利用者がサービスを得るために行動の痕跡を残し、企業が目先のサービス改善に必要な範囲を超えるデータを追加の入力へ変え、計算によって将来の行動を予測する商品を生産する。検証すべきなのは、この連鎖を広告型以外のプラットフォームにも広げられるか、監視資本主義が国家、金融、物流、非広告型プラットフォームの差異を過度に平板化していないかである。ゆえに Zuboff はデジタル経済の一般理論としてではなく、強力なモデルとして読むのがよい。どの産業で最も説明力を持ち、どの場面でプラットフォーム資本主義、データ植民地主義、行政の自動化による補完を要するのか。概念の転用が成立するには、強制、所有権、空間の支配、知識の序列、資源の流れといった具体的機制を説明し、古典的な植民地支配とプラットフォームのデータ関係との重要な違いも認めなければならない。行動余剰は、利用者の活動、データ抽出、予測市場を一つの資本蓄積過程に組み込む。その理論的射程はオンライン広告を大きく超え、監視資本主義は一般化されすぎていないかという論争も生んだ。生産的な読み方は、デジタル経済のすべてをこの概念に収めることではなく、行動予測に実際に依存する事業と、インフラの地代、取引手数料、独占価格から主に収益を得るプラットフォームとを具体的に区別することである。】
- Couldry, Nick, and Ulises A. Mejias. The Costs of Connection. Stanford University Press, 2019. (Read: Chapter 2, "Cloud Empire," pp. 37–68; Chapter 3, "The Coloniality of Data Relations"; Chapter 5, pp. 167–208).
【注:Couldry と Mejias は、社会生活が絶えず抽出可能なデータへ変換される過程をデータ植民地主義と呼ぶ。「Cloud Empire」は、クラウド、プラットフォーム、データ抽出、社会関係を結びつけるため、本節にとりわけ適している。議論すべきなのは、データ植民地主義がクラウド・インフラを理解する歴史的枠組みになりうるか、またどの点で古典的植民地主義との類推が過剰になるかである。著者たちはデータ関係を、より長い植民地的な資源領有の歴史に置き、社会生活そのものが商業利用可能なデータの流れへ絶えず翻訳されることを強調する。「クラウド帝国」は、この関係を支えるインフラと企業権力を指す。「植民地主義」という強い語だけに注目してはならない。社会生活の領有、データ抽出の常態化、新たな社会秩序から、どのように類推を組み立てるのかを精読する必要がある。Browne と併読する際には、植民地性が常に人種化、領土の領有、暴力的統治と絡み合ってきた以上、データ抽出だけから出発する一般化された植民モデルは歴史的差異を見落とさないか、と問う。監視の中心的主体は一つではない。国家安全保障、広告産業、雇用者による管理、金融のリスク管理、ソーシャル・プラットフォームはデータや技術を共有しても、法的権限と利害構造は異なる。監視資本主義という一概念ですべての制度差を均してはならない。データ関係を新たな植民地的領有として記述する狙いは、日常生活そのものが絶えず抽出可能な資源に変わることを強調する点にある。資本蓄積と主体の関係を結びつける強みを持つ一方、歴史的植民地主義における暴力、領土、人種の差異を消す危険もある。比較の有効性は、「植民」という語の道徳的な強さを借りるだけでなく、領有の仕組み、資源の流れ、依存構造を指摘できるかにかかっている。】
- Browne, Simone. Dark Matters: On the Surveillance of Blackness. Duke University Press, 2015. (Read: Introduction; Chapters 1–2).
【注:Browne は監視研究の歴史的奥行きを、奴隷制、人種化された識別、黒人の身体への継続的管理にまで広げる。デジタル監視を、インターネット時代に突然現れた新問題として描くことに反対するのである。論点は「すべての監視は同じだ」ということではなく、現代の識別、境界管理、可視性の制度が、以前の人種化された統治技術をどう継承し再編するかにある。Zuboff とデータ植民地主義のあいだに置くと、商業的データ抽出を中心とする説明だけで、監視権力の歴史的差異を十分に説明できるのかを検証する必要が生じる。「見られること」と「データ化されること」も同じではない。多くの統治システムの力は、まさに不可視の後方処理における分類と推論から生まれる。批判はプライバシーの露出だけでなく、分類を通じた機会、リスク配分、知の権威を研究しなければならない。Browne の Dark Matters: On the Surveillance of Blackness において、「植民地的類推の分析上の利点と、越えてはならない歴史的境界」は、したがって単なる概念問題ではない。データ抽出の政治性は非対称な関係から生まれる。一部の主体が収集、関連づけ、推論、再利用の能力を握る一方、別の主体は主に、記録・分類・予測されるものとして現れる。プライバシーはその一層にすぎず、所有権、知の権威、退出条件も同じく重要である。Browne は監視史に、人種と身体の歴史的連続性を再び扱うよう迫る。生体認証、国境検査、データベース分類は新技術に依存するように見えるが、「誰が識別を必要とされ、誰がリスクと見なされ、誰が身分を証明しなければならないか」は、決して技術的に中立な配置ではなかった。デジタル監視史はインターネット企業から始めるのでなく、可視性と人種化された統治の、より長い制度的系譜をたどらなければならない。また、Browne の dark sousveillance は抵抗の次元を保つ。監視される者は受動的に受け入れるだけでなく、識別や記録に対抗し、可視性を再符号化する実践を生み出す。監視技術の歴史は、権力拡大の歴史であるだけでなく、分類される主体が分類制度を絶えず撹乱する歴史でもある。この点は、デジタル・アイデンティティと生体認証の研究にとりわけ重要である。】
- Mbembe, Achille. Necropolitics. Translated by Steven Corcoran. Duke University Press, 2019. Originally published in French as Politiques de l'inimitié in 2016. (Read: Introduction; "Necropolitics"; selected chapters on enmity, borders, and exposure).
【注:Mbembe の死政治が論じるのは、主権、戦争、植民地性、そして生命が死や危害へ差異的に曝される政治的条件である。デジタル化・知能化の問題に用いる際、アカウント停止、順位引き下げ、プラットフォームからの排除を、そのまま「デジタルな死」と呼んではならない。その類推は概念の歴史的強度を弱め、現実の暴力関係を覆い隠す。より厳密な転用は、データ化された国境、自動リスク判断、軍事化された監視、資源配分が、異なる程度の曝露、傷つきやすさ、行動の自由をどう生み出すかを問うことである。さらに、誰がデータ標準、サーバー、モデル、研究課題を決め、調査される共同体が退出、拒否、修正、共同統治の能力を持つのかも問題となる。Mbembe の Necropolitics を本節全体に位置づけると、「データ正義、知の権威、差異化された曝露をいかに同じ研究枠組みに入れるか」という問いは、より具体的な構造を得る。抽出の政治性は非対称な関係に由来する。一部の主体は収集、関連づけ、推論、再利用の能力を握り、別の主体は主に記録・分類・予測される形で現れる。プライバシーはその一層であり、所有権、知の権威、退出条件も重要である。死政治は、主権が死への曝露、戦争化された空間、差異化された生存条件を通じて生命をどう組織するかを論じる。その歴史的背景はデジタル統治をはるかに超える。データの植民地性に導入する際には、アカウント停止などを「デジタルな死」へ誇張せず、デジタル・インフラがいつ国境、警察活動、戦争、資源配分に加わり、現実の生政治・死政治と接続するかを観察すべきである。したがって、ここで Mbembe に適しているのは尺度を補正する役割だ。データ化された統治が死政治の問題圏に入るのは、国境封鎖、軍事攻撃、警察活動のリスク、基礎的資源の差異的配分と結びつく場合である。この強度の高い生政治的文脈に概念を保つことこそ、デジタル批判が「死」の修辞を借りて歴史的重みを失うのを防ぐ。】
推薦文献:
- Benjamin, Ruha. Race After Technology: Abolitionist Tools for the New Jim Code. Polity, 2019. (Read: Introduction and Chapter 1).
【注:アルゴリズムと人種化された権力を結びつけ、情報検索を扱う Noble の視点を補う。一般的知性は無差別な「人類共通の資源」ではない。誰の知識、言語、身体的経験が訓練データになりやすいか自体が政治的問題なのである。理論上の類似は制度上の同一性を自動的に意味しない。したがって Benjamin の Race After Technology: Abolitionist Tools for the New Jim Code による「データ抽出、監視、人種化された分類、植民地性の歴史的関係」の説明は、対応する制度的機制を経験的資料によって確認できる場合に成立する。「New Jim Code」の分析は、技術的中立性そのものを批判の対象とする。不平等は明示的な差別規則として現れるとは限らず、客観的に見える効率、最適化、リスク管理を通して続きうる。ゆえに Benjamin は、アルゴリズムを既存の住宅、教育、治安、雇用の構造に置いて検討するよう求める。技術の社会的帰結は、それがどの制度史へ入り込むかに左右される。廃絶主義的ツールの方向も、単なる公平性の修繕とは異なる。問うのは、既存システムを各集団により均等に作動させることではなく、あるシステムがそもそも不平等な分類と処罰に奉仕していないかである。技術統治は「モデルをどう改善するか」に加えて、どの判断をそもそも自動化に委ねるべきでないかという根本的な制度選択を含む。Benjamin の廃絶主義的志向は、技術改革と制度改革の区別に特に有効である。予測システムが集団間の統計的公平性を実現しても、その用途が処罰、排除、過剰な警察活動の拡大なら、公平性指標はより根本的な正当性に答えられない。技術倫理には「このシステムは存在すべきか」という制度的否定を残す必要がある。】
- Noble, Safiya Umoja. Algorithms of Oppression. NYU Press, 2018. (Read: Introduction, pp. 1–14).
【注:Noble は、アルゴリズムによる順位づけが真空のなかで作動するのでなく、既存の社会的偏見を増幅することを示す。Crawford のデータ/労働の視点とは異なり、情報環境における分類と表象の政治を重視する。監視の中心的主体は一つではない。Noble の Algorithms of Oppression は、「商業的データ化と国家統治が代替し合わず併存する仕組み」に最終的な名前を与えるのでなく、競合する説明によって検証できる理論的入口を提供する。この開放性自体に研究上の価値がある。Noble の Algorithms of Oppression の核心は、検索結果に偏見があると証明することだけでなく、商業的順位づけ、広告の論理、社会的表象が共同で「関連性」を形づくることを示す点にある。検索システムは文化的分類と市場の誘因が重なる場となり、人種化された表象は単なるデータ誤差ではなくなる。Noble は検索エンジンを例に、商業的順位づけが既存の人種・ジェンダーの偏見を「最も関連する」情報へ包み直すことを示す。検索は公共的知識の中立な鏡ではなく、その分類は広告市場、コンテンツの生態系、プラットフォーム設計に左右される。ゆえに表象は文化だけの問題ではない。検索結果が集団や社会的事実を理解する入口になるとき、順位を決める力は認識論的な権力でもある。Noble の事例は情報インフラの累積効果も示す。検索結果は既存の内容に応答すると同時に、その後のクリック、内容制作、公共的認識を左右し、特定の結びつきを強める。過去の行動を関連性の信号とすれば、社会的偏見がフィードバックを通じて技術化されうる。順位づけは情報を処理すると同時に、将来の情報環境を生産する。検索結果の政治には知の権威も関わる。膨大な情報がアルゴリズムで整理されると、利用者は「最初に何が出るか」を世界そのものの自然な序列と受け取りやすい。Noble が示すのは、このインターフェースによる自然化である。商業的・社会的偏見が技術的順位づけを経て客観性の外観を得るため、制度的選択として見抜きにくくなる。】
- Lyon, David. Surveillance Studies: An Overview. Polity, 2007. (Read: Introduction; chapters on surveillance as social sorting).
【注:Lyon は監視研究に「見られること」より広い概念枠組みを与え、情報収集、分類、社会的選別という制度過程としての監視を重視する。プラットフォーム、国家、商業データベースは、個人を絶えず眺める必要があるとは限らない。身分、リスク、消費、資格のカテゴリーによって、誰がどんな機会を得るかを決めるため、本節の基礎文献に適している。「見られること」と「データ化されること」は同じではない。Lyon の Surveillance Studies: An Overview を通じ、「植民地的類推の分析上の利点と越えられない歴史的境界」は組織・制度の次元へ入る。Lyon の Surveillance Studies: An Overview の尺度では、その利点と境界は連続性と変形をともに扱う必要がある。旧制度は新技術から新たな操作方式を得る一方、新技術も既存制度へ再吸収されうる。Lyon の Surveillance Studies: An Overview の価値は、監視を単一の装置でなく社会的組織の関係として理解することにある。識別、分類、可視性、機関の目的が共同で監視の形式を決め、商業プラットフォーム、国家行政、安全保障システムの比較の基礎を作る。Lyon は監視を単なる「覗き見」でなく、社会的分類と組織的実践として捉える。現代の監視は身元記録、データベース、リスク分類に依拠し、人口を異なる待遇と移動権限へ振り分ける。デジタル技術はこの能力を拡張するが、監視と行政組織の長年の結合を継承してもいる。この社会学的伝統に監視研究を置くことは、カメラ、プラットフォーム、プライバシーだけを中心とする技術中心主義的な語りを避ける助けになる。】
- Andrejevic, Mark. Automated Media. Routledge, 2020. (Read: Introduction; selected chapters on automation, surveillance, and decision systems).
【注:Andrejevic は、自動化されたメディアが知覚、分析、反応をますますデータ集約型システムへ委ねる過程を論じる。核心は「人間が機械に置き換わる」と宣言することではなく、機関が監視の規模を人間の処理能力以上へ広げ、自動化された相関と予測で情報環境を管理する仕方を分析することにある。Zuboff と併読すれば、資本蓄積、メディアの自動化、統治能力の関係をより細かく区別できる。Andrejevic の Automated Media が示す「データ正義、知の権威、差異化された曝露を同じ研究枠組みに入れるにはどうするか」という問いは、文献内部の概念論争にとどまらない。したがって、Andrejevic の Automated Media の外にある経験的資料と競合する説明によっても検証されなければならない。自動化されたメディアの中心問題は「機械が自律しているか」ではなく、知覚・分析・反応の連鎖が圧縮された結果、どの判断が公開の説明を経なくなるかである。大規模な自動化は監視と応答の能力を著しく広げる一方、意思決定の根拠を影響される人々から遠ざけうる。Andrejevic が強調するのは、この規模の政治だ。自動化は機関がどれだけ見え、どれほど速く処理し、どんな条件で行動するかを変える。また、政治的行動の時間尺度も変える。システムは人間が理解する前に印を付け、遮断し、推薦できるため、機関は機械に先に反応させ、後から説明する方式へ依存しうる。速度は権力資源となる。膨大な信号をリアルタイムに処理できる組織は、事後にしか異議を申し立てられない個人より構造的に優位にある。自動化の政治は、この時間的非対称性も分析すべきである。速度の優位が機関の強制力と結びつくと、新しい手続き上の問題が生じる。自動判定は一瞬でも、説明と訂正には長い時間がかかる。異議申立て費用の非対称性は、それゆえ自動化された統治を評価する重要な指標となる。】
- Turow, Joseph. The Daily You: How the New Advertising Industry Is Defining Your Identity and Your Worth. Yale University Press, 2011. (Read: Introduction; selected chapters on behavioral targeting and audience classification).
【注:Turow は広告産業の組織変化から出発し、受け手が細分化され、価値を付けられ、継続的に標的化される過程を追う。監視資本主義に重要な産業史的背景を与える文献である。データ抽出は真空のなかで発明されたのではなく、広告市場、受け手の測定、顧客関係管理、差別的価格設定という既存の実践とつながる。ここでの「アイデンティティ」は第一に商業的分類であり、主体の自己認識と必ずしも同じではない点に注意したい。Turow の The Daily You: How the New Advertising Industry Is Defining Your Identity and Your Worth の尺度では、「データ抽出、監視、人種化された分類、植民地性の歴史的関係」は、まず具体的な諸関係として現れる。広告産業の歴史はまた、データの価値が個人の行動に自然に内在するのでなく、市場分類と取引可能な受け手の想像を通じて作られることを示す。「高価値利用者」「離反しやすい顧客」「高リスク集団」はいずれも、組織が特定の目的のために作るカテゴリーである。個別化されたマーケティングと社会的階層化は独立した二つの過程ではなく、同じデータ・インフラ上で強め合いうる。Turow の産業史は「個人化」の別の顔を明らかにする。市場が利用者を細かく分けられるほど、扱いに差をつけられる。個人化された推薦は対応づけの効率を上げるだけでなく、消費者を異なる価格、割引、情報環境へ振り分けることもある。商業的分類と機会の階層化の境界は曖昧になり、広告技術はより広い社会的選別の効果を持ち始める。広告技術は商品情報だけでなく、企業が投資に値すると考える注意と優遇も配分する。商業的評価が社会的差異化へ直接参加するのである。】
- van Dijck, José. "Datafication, Dataism and Dataveillance: Big Data between Scientific Paradigm and Ideology" Surveillance & Society 12, no. 2 (2014): 197–208. (Read: entire article).
【注:van Dijck はデータ化、データ主義、データ監視を区別することで、「すべてはデータ化できる」という命題を、技術的実践であるとともに認識論的な約束を伴う文化的イデオロギーとして捉えさせる。データは自動的に真理の地位を得るわけではなく、プラットフォームや研究機関が大規模な行動データを信頼すること自体に説明が要る、という指摘が重要である。Rouvroy の相関による統治と照らし合えるが、本節ではデータ抽出を正当化する知識文化的条件を重視する。van Dijck の「Datafication, Dataism and Dataveillance: Big Data between Scientific Paradigm and Ideology」、Turow の The Daily You: How the New Advertising Industry Is Defining Your Identity and Your Worth、D'Ignazio の Data Feminism は、「商業的データ化と国家統治が代替し合わず併存する仕組み」を異なる知の伝統に置く。監視の中心的主体は一つではない。van Dijck の論文は、この仕組みを過度に抽象的な時代診断から解放する。データ主義が重要なのは、データ化を技術操作から認識論的信念へ高めるためだ。記録が十分網羅的で、計算が十分複雑なら、社会的行動は直接把握できるかのように見える。この信念への批判は、アルゴリズムによる統治に必要な認識論的次元を与える。データの規模は概念の選択を代替せず、相関も因果、意味、正当性の問いに自ら答えることはできない。】
- D'Ignazio, Catherine, and Lauren F. Klein. Data Feminism. MIT Press, 2020. (Read: Introduction; Chapters 1–3).
【注:D'Ignazio と Klein は、データ実践をフェミニズムの交差性分析に置き、権力、文脈、身体性、抜け落ちた経験を重視する。一般的な「AI の公平性」手引きではなく、誰がデータセットを定義し、誰が記録され、誰が欠落し、誰が分類の帰結を負うのかを研究者に問い返す。Browne、Benjamin、Noble と並べることで、批判は「システムに偏りがあるか」からデータ生産そのものの政治へ進む。「見られること」と「データ化されること」は同じではない。D'Ignazio の Data Feminism を中心に据えると、「植民地的類推の分析上の利点と越えられない歴史的境界」を、相互に関連する諸段階へ分け直せる。D'Ignazio の Data Feminism から見れば、それは制度的権限と資源配分から独立には成り立たない。技術的形式が持続的な社会的帰結を生むのは、組織的権力と結びついたときである。Data Feminism は権力分析をデータ実践に直接組み込み、問いの設定、分類、可視化、解釈の各段階で、誰が代表され、誰が欠落し、誰が帰結を負うかを追う。「女性や少数者のデータを増やすこと」を十分な答えとはせず、既存の権力構造と二分法的分類への挑戦を重視する。データ正義は偏りの修正にとどまらず、知識生産の目的と参加関係の再設計にも関わる。また、文脈と多元性を特に重視する。収集の文脈を離れたデータセットは、具体的な社会関係を自然な事実へ抽象化しやすいが、多様な知の立場はその代償を明らかにする。データ実践の民主化はダウンロード権限の開放だけでなく、出所の記録、不確実性の保持、影響される集団の問題設定への参加を含む。】
- Birhane, Abeba. "Algorithmic Colonization of Africa" SCRIPTed 17, no. 2 (2020). (Read: entire article).
【注:Birhane は、普遍主義、開発主義、効率の言説を通じてデジタル技術がアフリカ社会へ入る際に再生産されうる非対称な権力を、アルゴリズムによる植民地化と呼ぶ。技術、資本、データ、標準は外部主体に握られ、影響を受ける共同体は導入先やデータ源として扱われがちである。Couldry/Mejias のデータ植民地主義と直接対話し、「植民」という概念は修辞にせず、具体的な所有権、知の権威、制度的依存を通じて立証すべきことを示す。Birhane の「Algorithmic Colonization of Africa」、D'Ignazio の Data Feminism、Ricaurte の「Data Epistemologies, the Coloniality of Power, and Resistance」は、「データ正義、知の権威、差異化された曝露を同じ枠組みに入れる」という問題を、巨視的判断から機制の水準へ下ろす。この問題に対する Birhane の貢献は、巨視的判断を識別可能な制度的機制へ戻す点にある。さらに同論文は、その問いを歴史的条件に置き直す。似たデジタルな外観が同じ機制に由来するとは限らず、概念の有効性は差異を保つことに依存する。Birhane は「植民」を比喩から制度関係へ引き戻す。誰がモデル、クラウド・サービス、データ標準を設計し、誰が知識への需要を定め、技術の利益はどこへ流れるのか。それが植民地性を判断する具体的根拠となる。こう捉えれば、アルゴリズムによる植民地化は、あらゆる越境的技術移動ではなく、持続する非対称な依存と、それによる現地の知、自治、公共的能力の侵食を指す。】
- Ricaurte, Paola. "Data Epistemologies, the Coloniality of Power, and Resistance" Television & New Media 20, no. 4 (2019). (Read: entire article).
【注:Ricaurte はデータ化を権力の植民地性と結びつけ、資源抽出だけでなく知識生産を重視する。何が認識でき、どう分類し、誰がデータの結果を社会的現実と定める権限を持つのか。データ植民地主義を経済的比喩にとどめず、知の権威へ議論を進める点で、ラテンアメリカの脱植民地理論をデータ研究に導入する重要な接点となる。Ricaurte の「Data Epistemologies, the Coloniality of Power, and Resistance」からは、「データ抽出、監視、人種化された分類、植民地性の歴史的関係」について、より明確な因果の層を得られる。Ricaurte はさらに、データの問題を知識生産の植民地性に置く。すべての世界が同じ仕方でデータ化されるわけでも、あらゆる知が損失なくデータベースやモデルへ入るわけでもない。インフラが分類、変数、測定可能な対象を選ぶ時点で、すでに認識論的政治が含まれる。抵抗はプライバシーや所有権だけでなく、何を知識と認めるかを誰が規定できるかにも関わる。この批判はデータ主権にも及ぶ。現地共同体が生データだけを提供し、分類標準、モデル設計、解釈権が外部機関へ集中すれば、知識生産は中心・周辺構造を再生産する。脱植民地化には「代表性のあるデータ」を増やすだけでなく、研究課題、インフラ、結果の解釈に関する権力の再配分が要る。したがってデータの脱植民地化はインフラ問題を含む。計算、保存、モデルの能力が外部主体に依存しつづければ、知の自律性は限られる。データ・インフラそのものが知の政治と脱植民地実践の争点となり、知識への権利とインフラへの権利は明確に分離できない。】
- Magnet, Shoshana Amielle. When Biometrics Fail: Gender, Race, and the Technology of Identity. Duke University Press, 2011. (Read: Introduction; selected chapters on fingerprinting, facial recognition, and biometric failure).
【注:Magnet は生体認証を通じ、いわゆる「客観的な身元」がデータベース、センサー、閾値、既存の社会的分類に依存することを示す。システムの失敗は単なる工学的誤差ではない。身体ごとに不均等に分布し、人種、ジェンダー、国籍の差異を識別問題として再符号化することもある。監視やデータの植民地性という巨視的議論に、極めて具体的な技術的対象を与える文献である。監視の中心的主体は一つではない。生体認証は分類と身体の循環を明確に示す。システムはまず既存の人口カテゴリーで訓練・検証し、その後「技術的識別」の形で同じカテゴリーを国境、安全保障、身元制度へ戻す。失敗率は単なる工学上の欠陥ではない。識別に強制的な帰結が伴うと、誤差は既存の社会的差異に沿って不均等に配分される。Magnet が強調するのは、「失敗」はシステム外の偶発事ではないという点だ。生体認証が客観的・自動的・普遍的とされると、誤認は分類体系の問題でなく個人の異常として制度に扱われる。技術的失敗には政治的な配分があり、特定の身体は追加の認証、疑念、自己証明の費用を負わされる。識別精度はこの不均等な帰結とともに評価しなければならない。生体認証の政治的評価は誤りの率と帰結を結びつける必要がある。同じ統計的誤差も、異なる身元制度ではまったく違う社会的費用を生むからだ。技術的精度と制度的正義は、互いに無関係な評価次元ではありえない。身分証、国境、福祉で使われる生体認証の失敗は、単なる使い心地の問題ではなく、移動、資源の受給、合法的身分の証明を左右しうる。性能評価は制度的帰結と結びつけなければならず、娯楽アプリと強制的身元システムでは、同じ誤認率の政治的重みはまったく異なる。】
- Mejias, Ulises A., and Nick Couldry. Data Grab: The New Colonialism of Big Tech and How to Fight Back. University of Chicago Press, 2024. (Read: Introduction; chapters on data territories, colonialism, and resistance).
【注:本書は著者たちのデータ植民地主義の議論を継承・更新し、日常生活の継続的データ化と巨大技術企業のインフラ拡張を、植民地関係の長い歴史に位置づける。2019年の The Costs of Connection より、現代の巨大テック企業と抵抗戦略を明確に扱うため、二冊を別々の回へ分けて同じ役割を重複させるのでなく、同じ本節内での後続論争として読むのに適している。読解の中心は、植民地的類推の分析上の利点と歴史的境界に置く。「見られること」と「データ化されること」は同じではない。Mejias の Data Grab: The New Colonialism of Big Tech and How to Fight Back によって、その利点と越えられない境界を、修辞的命題から制度分析へ変えられる。Data Grab はデータ植民地主義の枠組みを継続しながら、巨大テック企業が接続、利便性、インフラを継続的なデータ領有へ変える仕方へ集中する。政策上の意義は、問題を個人のプライバシー選択から集団的なデータ関係へ広げる点にある。大手プラットフォームから退出すると仕事、交際、公共サービスの機会を失う場合、「同意」だけで正当性のすべてを支えるのは難しい。抵抗には制度的制限と公共的代替が必要となる。Mejias と Couldry はさらに、データ収奪は一度限りの強制的領有によらず、接続が生活の前提となった後で継続して生じると強調する。退出が難しいほど、データ抽出はサービス交換の自然な条件として包装されやすい。本来の制度的問題は、データ最小化、集団的権利、公共インフラを作り、社会参加の代価として無限のデータ譲渡を要求しないようにできるか、という点へ移る。】
- Taylor, Linnet. "What Is Data Justice? The Case for Connecting Digital Rights and Freedoms Globally" Big Data & Society 4, no. 2 (2017). (Read: entire article).
【注:Taylor は、プライバシー、データ保護、より広い社会的正義を結ぶためにデータ正義を提起する。批判から規範的な問いへ移る際に有用である。データ化が可視性、代表性、資源配分、行為能力の不平等を生むなら、制度評価では個人の同意だけを見るわけにはいかない。本節の締めくくりとして、監視と植民地性の診断を、権利、能力、世界的不平等の研究課題へ変えられる。データ正義はデジタル権を個人のプライバシーから構造的配分へ広げる。誰がデータにアクセスでき、誰がデータ化を拒み、誰が自動化の誤りを負い、誰がデータ経済から利益を得るかは、一つの「同意」原則では解決できない。Taylor の枠組みは統治と社会的正義を結び、国際比較のために制度的差異も保持する。世界的なデジタル権のなかにデータ正義を置くことで、欧米のプライバシー法の経験を普遍化せずに済む。各国ではインフラと社会保障、身元制度、開発政策の結びつきが異なり、デジタル権の緊急性も異なる。越境的正義には、一つの法的範型の複製でなく、個人の自由と資源配分の両方を扱うことが必要である。分析単位も個々のデータ主体だけでなく、国家、プラットフォーム、共同体、インフラが共に構成するデータ関係となる。データ正義はまた、通常分離される政策領域をつなぐ。プライバシー保護はデータの支配、反差別は分類の帰結、開発の正義はインフラと能力の分布を問題とする。Taylor はこれらを同じ規範的空間へ入れ、選択肢に恵まれた個人消費者だけをデジタル権の標準主体にしないよう促す。】
第4回:惑星規模の計算とインフラ空間
本節では「クラウド」を、土地、エネルギー、チップ、データセンター、海底ケーブル、衛星、周波数帯、都市、法的取り決めへ戻す。Bratton の The Stack は惑星規模の計算を多層構造として記述し、Easterling の Extrastatecraft はインフラ空間が標準、物流、空間的形式を通じて準統治的な効果を生む過程を分析する。Star の研究は、インフラが順調に動くときには背景へ退き、故障、衝突、排除のなかで再び姿を現すことを示す。プラットフォームの権力は、インターフェース、アルゴリズム、事業モデルだけでなく、接続の条件、物質的資源、維持管理の体系にも存在する。
インフラ化は権威と領土の関係も変える。Sassen による領土・権威・権利の再編成、Graham と Marvin の差異化された都市ネットワーク分析、Plantin らによるプラットフォーム研究とインフラ研究の接続は、「プラットフォーム」と「インフラ」の境界が不安定になっていることを示す。あるプラットフォームは API、クラウド・サービス、本人確認体系、決済のインターフェースを通じ、他の組織が簡単に迂回できない基盤になる。一方、従来のインフラも、データ化、サービス化、段階的アクセスというプラットフォームの論理を取り入れる。政治的問題は、国家が単純にプラットフォームへ置き換わるかではなく、異なる権威が複数の尺度をまたぐネットワークでどう再配分されるかにある。
Varoufakis と Durand の技術封建主義という命題は、このインフラ的背景に置いてはじめて検証可能になる。クラウド資本、無形資産、プラットフォームの入口、地代、退出障壁が、従来の資本・労働関係とは異なる依存構造をなすかは、プラットフォーム資本主義、独占資本主義、地代理論と項目ごとに比較する必要がある。Morozov ら反対論の重要性は、まさにこの概念に追加的な説明力を示すよう迫る点にある。「封建主義」が独占、地代、市場支配力を言い換えるだけなら、理論的必要性は限られる。プラットフォームのインフラ支配が市場参入、所有権、依存の形を実際に変えるなら、どの制度的水準で変化するかをさらに説明しなければならない。
必読文献:
- Bratton, Benjamin H. The Stack: On Software and Sovereignty. 10th anniversary ed., with a new preface by the author. MIT Press, 2026. Originally published 2016. (Read: new preface; "The Nomos of the Cloud"; "Platform and Stack, Model and Machine"; "Cloud Layer").
【注:Bratton のスタックは、単なる「クラウド・コンピューティング国家論」ではない。「Nomos of the Cloud」は惑星規模の計算が従来の政治地理へ突きつける課題を論じ、「Platform and Stack」は計算の構造が政治モデルになる仕方を説明し、「Cloud Layer」はデータセンター、プラットフォーム、インフラ、国家統治を同じ構造に置く。Google や Amazon を直接国家と呼ぶよりも、主権が各層と各組織のあいだにどう分布するかを精読する。MIT Press の2026年刊行10周年新版を用い、原著の惑星規模の計算の構造的記述を振り返ると同時に、新序文で十年後の有効性と盲点を検証する。六層モデルが示唆的なのは、現実が Earth、Cloud、City、Address、Interface、User に整然と分かれるためではない。エネルギーと鉱物がクラウド層を支え、アドレス体系が対象を計算可能な秩序へ入れ、インターフェースが制度規則を操作可能な選択肢にし、「利用者」が人間、組織、センサー、機械エージェントにもなりうるという、層をまたぐ関係を追えるからである。これは「モデルのモデル」であり、経験的地図ではない。2026年版では、十年間の主権、クラウド・プラットフォーム、AI エージェント、地政学の変化のうち、何がスタックの説明力を強め、何が階層の想像の修正を要するかを問い返したい。スタックは尺度横断的な構造言語を提供する。六層は単なる技術層でなく、計算が領土、身元、主権を再編する様子を切り取る分析面である。データセンター、都市、利用者インターフェースを同じ惑星的システムに置ける強みがある一方、構造的抽象が地域差を覆い隠す危険もある。現実の完全な地図より、比較の枠組みとして用いるのが適切である。】
- Easterling, Keller. Extrastatecraft: The Power of Infrastructure Space. Verso, 2014. (Read: Chapter 2, "Zone: The Infrastructure Commons," pp. 25–70).
【注:Easterling が論じるのは、単に「特区は法の外にある」ということではない。インフラ空間が空間商品、プロトコル、標準、物流、制度的取り決めによって、従来の国家計画を超える統治効果を生む仕方である。デジタル・プラットフォームへの適用では類推の限界を保つ必要がある。Amazon のクラウドと自由貿易区は同じ空間ではないが、どちらも規則が従来の法文でなくインフラや標準を通じて作用しうることを示す。プラットフォームを国家、帝国、領土と呼ぶには慎重さが要る。比較可能なのはインフラ化された規制という次元であり、法的地位、空間構造、経済関係は別々に扱うべきだ。インフラ空間の政治性は、明示的な法的命令によらず、技術規格、空間的ひな型、運用標準で行動を組織する点にこそある。自由貿易区、物流網、データセンターは「国家の外」の真空ではなく、国家政策、民間資本、国際標準が共同で生産することが多い。Easterling は準統治的効力を分析する空間的言語を提供する。さらに重要なのは、インフラ空間が複製可能性を通じて力を発揮することだ。特区、工業団地、物流モデルが標準的空間商品として包装されると、あらかじめ設定された統治論理とともに地域を越えて広まる。API、クラウド構造、利用規約も似た複製可能性を持つが、空間的・法的機制は異なる。比較は、規則がインフラとともに広がる水準にとどめるべきである。この空間政治は、インフラの権力がしばしば「初期設定の環境」として働くことを示す。空間やシステムに埋め込まれた規則は、明示的命令より見分けにくい。核心は「法の外」ではなく、規則が空間、標準、運用という形式で持続的効力を得る仕方にある。】
- Star, Susan Leigh. "The Ethnography of Infrastructure" American Behavioral Scientist 43, no. 3 (1999): 377–391.
【注:Star は本節に不可欠な方法論的基礎を与える。インフラが見えにくいのは、仕事と生活の背景条件になるためであり、故障、衝突、排除の際に突然見えるようになる。クラウド研究では、インターフェースやソフトウェアだけでなく、電力、ネットワーク、サーバー室、維持労働、標準を問わなければならない。地代、ネットワーク効果、独占を既存の資本主義理論が十分説明できるなら「封建」は修辞にすぎない。新たな依存、入口支配、非市場的配分があるなら、なぜ新しい社会構成概念が必要かを説明すべきである。ゆえに Star の「The Ethnography of Infrastructure」における「地代、利潤、入口支配、退出費用の政治経済的差異」は単なる概念問題ではない。Star は維持、修理、例外的経験を研究の中心に置く点でも重要だ。安定して動くインフラは大多数の利用者にとって背景へ退くが、維持する者、排除される者、故障に遭う者には常に見える。断線、サービス停止、互換性の失敗、権限拒否において、社会的差異が最も明確になることが多い。関係的である以上、同じシステムが誰にでも同じように「背景」となるわけではない。安定したブロードバンドは都市中間層には不可視でも、遠隔地、障害のある利用者、保守労働者には明示的な制約でありつづける。Star の民族誌は差異の視点を内包する。「誰のインフラか」には具体的に答える必要がある。維持の視点は革新の語りも変える。システムの信頼性を支えるのは、当初の設計よりも長期にわたる反復的で目立たない修理労働であることが多い。障害ログや保守現場は、宣伝資料より現実の権力関係をよく示すことがある。】
- Plantin, Jean-Christophe, Carl Lagoze, Paul N. Edwards, and Christian Sandvig. "Infrastructure Studies Meet Platform Studies in the Age of Google and Facebook" New Media & Society 20, no. 1 (2018): 293–310.
【注:本論文は本節の方法論的宣言とも言える。プラットフォームがインフラ的性格を獲得し、従来のインフラがプラットフォームの論理で再編されるため、両者は完全には分離できなくなっている。Star と Bratton をつなぎ、「クラウド」を技術層だけでなく社会組織の形式として捉えさせる。惑星規模の概念は具体的な場所を消してしまいやすい。巨視的モデルが説明力を持つには、鉱物、チップ、データセンター、ケーブル、都市、労働力、法域といった節点に落とし込み、地域ごとの費用負担が対称かを検証しなければならない。Plantin の「Infrastructure Studies Meet Platform Studies in the Age of Google and Facebook」を全体の文脈に置くと、「惑星的モデルと具体的場所、労働、維持との尺度上の緊張」は具体的構造を得る。プラットフォームのインフラ化とインフラのプラットフォーム化は、Google や Facebook が任意のサービスから社会的依存の節点へ移る過程を的確に捉える。インフラ化したプラットフォームでは安定性、普遍的アクセス、公共的責任が切実となり、従来のインフラにプラットフォームの論理が入れば、データ化、商業化、インターフェース化が進みうる。双方が公共施設と民間プラットフォームの境界を曖昧にする。この双方向の変化は公共的責任も書き換える。インフラには安定性、普遍性、長期利用可能性が期待されるが、企業はサービス、価格、規則を素早く変えられる。私的プラットフォームが教育、仕事、公共的交流に不可欠な節点となるとき、企業の柔軟な統治と社会インフラとしての責任の矛盾が表面化する。】
推薦文献:
- Sassen, Saskia. Territory, Authority, Rights: From Medieval to Global Assemblages. Princeton University Press, 2006. (Read: Chapter 7, "Digital Networks, Global Cities," pp. 323–377).
【注:Sassen が重要なのは、グローバル化を国家主権の単純な消滅と捉えず、領土、権威、権利が国際的ネットワークでどう再編されるかを論じる点にある。「プラットフォーム国家」より正確な枠組みであり、異なる権威が共存して特定の空間的節点で重なり合う。デジタル・インフラは生来的な「超国家空間」ではなく、新旧の統治尺度が再結合する場所なのだ。グローバルなアッサンブラージュは「国家の退場」という語りの説明力を失わせる。世界的デジタル網は法的承認、都市の節点、金融制度、越境規則を必要とし、一部の国家能力を弱める一方、別の能力を再編・強化しうる。権威は国家からプラットフォームへ単純に移るのではなく、複数の尺度のあいだで再編成される。また脱領土化と再領土化はしばしば同時に起こる。越境データ流は場所を離れるように見えるが、サーバー立地、税、司法管轄、人材集積は特定の都市や国家へ再び固定する。世界的ネットワークは空間の消失ではなく、異なる尺度の空間関係の並べ替えである。デジタル・グローバル化は領土を取り消すのでなく、領土、権威、権利を新しい尺度の組み合わせで同じインフラ網へ入れる。デジタル網の越境性には、新たな節点集中と法域間競争が常に伴う。世界性は具体的場所を通じて生産される。権威の再編は、企業と国家がゼロサム関係ではないことも意味する。国家は税制、データ現地化、安全保障政策でクラウドを形づくり、プラットフォームは越境的構造で国家の規制費用を変えられる。Sassen の視点は「誰が誰を代替するか」に現実を単純化せず、この共同生産関係を分析する助けとなる。】
- Graham, Stephen, and Simon Marvin. Splintering Urbanism. Routledge, 2001. (Read: Chapter 1, pp. 1–36).
【注:都市インフラの差異的供給から、ネットワークが空間的分断を生む仕方を理解する。Easterling と併読すれば、インフラは公共性を必然的に生むのでなく、価格、優先度、接続方式、空間的選択によって「サービス階級」を作りうると分析できる。プラットフォームを国家、帝国、領土と呼ぶには慎重さが要る。分裂する都市主義は、インフラが本来的に普遍主義へ向かうわけではないことを示す。ネットワーク型サービスは価格、技術標準、空間投資で接続を階層化し、高価値地域へ高速性、冗長性、安定性を与え、周辺地域には中断と不足を負わせうる。デジタル・インフラに応用すれば、「接続」自体が新たな社会的差異になる過程を分析できる。分裂は接続の有無だけでなく、品質、遅延、帯域、信頼性、支払可能性にも生じる。仕事、教育、公共サービスの前提になるほど、技術的差異は社会的機会の差異へ変わる。インフラの政治は「普及率」という単一指標を超えなければならない。接続品質の階層化は、やがて教育、雇用、公共参加の機会に沈殿し、インフラの公平性は明確な分配の政治となる。研究では接続の有無だけでなく、サービス品質と空間差を含める必要がある。インフラ差は最終的に社会的機会の差となる。良質な接続、冗長性の保障、維持資源の配分は、名目的普及率よりもデジタル都市の階級的・地域的分断をよく示す。ネットワークの公平性は設計と投資判断に入るべきだ。また Splintering Urbanism は、市場化が普遍的サービスの論理を収益性の論理へ書き換えることを示す。高価値の利用者と地域は優先投資を得やすく、低収益空間は長く保守不足になる。同じ商業的選択だけに依存するデジタル網も、既存の都市の階級差と地域差をさらに固定しかねない。】
- Mattern, Shannon. Code and Clay, Data and Dirt: Five Thousand Years of Urban Media. University of Minnesota Press, 2017. (Read: Introduction; selected chapters on urban information infrastructures).
【注:Mattern は都市メディア史を文字、地図、建築、物流、データベース、センサー体系まで延ばし、スマートシティを突然到来したデジタル革命と扱うことを拒む。鍵となる方法は、コードと粘土、データと土を並置することだ。情報インフラは常に物質的媒体、空間制度、維持労働に依拠する。本節では、Bratton の巨視的スタックを歴史化し、都市の尺度で補正できる。Mattern の Code and Clay, Data and Dirt: Five Thousand Years of Urban Media を通じ、「地代、利潤、入口支配、退出費用の政治経済的差異」は組織・制度の次元へ入る。都市は文字、音、石材、地図、センサー、データベースが長く共同で構成してきたメディア環境であり、デジタル化を突然の革命と描くスマートシティ論に抗する。都市計算は既存の道路、配管網、行政記録、社会的記憶に依拠せざるをえない。インフラは旧都市を覆う不可視の網ではなく、歴史的な物質層と絶えず重なり、摩擦を起こす。長期の都市メディア史は、古いインフラを時代遅れの背景とする「データ都市」にも反対する。センサーは道路に付着し、データベースは記録の分類を引き継ぎ、都市インターフェースは建築と行政境界に制約される。デジタル化は白紙からの再設計ではなく、既存の物質、制度、歴史との重層化なのである。この視点は、スマートシティの言説にある技術的断絶感を大きく和らげる。また、メディア概念を画面から都市そのものへ広げる。街路表示、記録体系、建材、センサーが共同で情報を組織する以上、スマートシティは長いメディア化の歴史の新たな層であり、都市が初めて「計算可能」になる瞬間ではない。】
- Hu, Tung-Hui. A Prehistory of the Cloud. MIT Press, 2015. (Read: Introduction; selected chapters on cloud infrastructure, sovereignty, and labor).
【注:Hu はクラウドの前史を通じ、ネットワークを場所も摩擦もない空間として想像する言葉を解体する。クラウドを以前のネットワーク、軍事、メインフレーム、データセンター、労働関係へつなぎ、自然な技術層ではなく歴史的に形成されたインフラと政治的想像として示す。Bratton と併読し、構造モデルとメディア考古学が同じ対象をどう扱うか比較できる。惑星規模の概念は具体的場所を消しやすい。Hu の A Prehistory of the Cloud が示す「惑星的モデルと具体的場所、労働、維持との尺度上の緊張」は、文献内部の概念論争にとどまらず、Hu の A Prehistory of the Cloud の外にある経験的資料と競合する説明にも検証されなければならない。Hu の A Prehistory of the Cloud はこの緊張を特定の対象と因果連鎖に限定する。その範囲を超える類推には、新しい証拠が必要であり、用語だけでは済まない。クラウドの比喩の前史をたどると、無場所性、無重量性、無限拡張の想像が、軍事、ネットワーク、データセンターの基盤を覆い隠すことがわかる。クラウドの「仮想性」は物質的事実でなく、インターフェースの効果である。データセンターの集中、端末の分散、経路の階層化は、クラウドが空間を抽象化すると同時に具体的な地理的依存を再編することを示す。比喩には政治心理的効果もある。利用者は保存と計算を、企業が建設して料金を取るインフラより自然環境として理解しやすくなる。サービスは空気のようにどこでも得られるように見えるが、背後では少数の巨大事業者に依存する。Hu の歴史化は、この自然化を技術産業の文化的産物として再び露わにする。】
- Parks, Lisa. Cultures in Orbit: Satellites and the Televisual. Duke University Press, 2005. (Read: Introduction; selected chapters on satellite infrastructures and global visibility).
【注:Parks はメディア・インフラを軌道空間へ延ばし、衛星が地表、信号、画像、地政学をどうつなぐかを研究する。惑星規模の計算に対する意義は、「惑星的」が修辞としての「グローバル」ではなく、軌道、周波数帯、地上局、国家の許可、資本投入で構成される垂直空間だと示す点にある。クラウドもこうした具体的物質層に位置づける必要がある。Parks の Cultures in Orbit: Satellites and the Televisual の尺度では、「クラウド、プラットフォーム、インフラが土地、エネルギー、標準、主権をどう結ぶか」は具体的な諸関係として現れる。技術封建主義の成立は、独占、地代、プラットフォーム資本主義と比較すべきだ。「封建」が市場支配力の言い換えなら追加的説明力は限られ、インフラ支配が依存、参入、収益構造を変えるなら、どの制度的水準で変わるかを示す必要がある。衛星メディアは、インフラの垂直次元をデジタル政治に持ち込む。軌道位置、地上局、周波数配分、遠隔観測能力が通信条件を共同で構成するため、「全世界のカバー」は具体的な地政学的配置に依存する。Parks は惑星尺度を巨視的修辞から、追跡できる技術対象と資源競争へ変える。衛星放送から衛星インターネットまで、軌道インフラは通信主権に垂直空間があることを示す。国境だけでは信号範囲を説明できないが、打上げ能力、周波数調整、地上受信はなお制度に管理される。惑星計算の「世界性」は、希少な位置、技術的許可、資本投入を通じて具体的に作られる。「全世界のカバー」は境界の不在でなく、地表の境界が軌道、周波数、地上局の制度的配置へ延びることなのだ。】
- Lehdonvirta, Vili. Cloud Empires: How Digital Platforms Are Overtaking the State and How We Can Regain Control. MIT Press, 2022. (Read: Introduction; selected chapters on platform governance and institutional power).
【注:Lehdonvirta は巨大プラットフォームを、規則制定、執行、紛争処理の能力を持つ制度組織として比較し、越境デジタル市場で伝統的統治の一部を担える理由を問う。書名の「国家を追い越す」は事実描写ではなく検証すべき命題である。価値は「プラットフォームはもう国家だ」と一括するのでなく、国家主権、私的統治、市場インフラ、利用者の依存を正確に区別する点にある。Lehdonvirta の Cloud Empires: How Digital Platforms Are Overtaking the State and How We Can Regain Control、Parks の Cultures in Orbit: Satellites and the Televisual、Varoufakis の Technofeudalism: What Killed Capitalism は、「プラットフォームのインフラ化とインフラのプラットフォーム化」という双方向過程を異なる知の伝統へ置く。プラットフォームを国家、帝国、領土と呼ぶには慎重さが要る。Lehdonvirta の Cloud Empires: How Digital Platforms Are Overtaking the State and How We Can Regain Control はこの過程を抽象的すぎる時代判断から解放する。技術封建主義の成立は、独占、地代、プラットフォーム資本主義と比較すべきだ。「封建」が市場支配力の言い換えなら追加的説明力は限られ、インフラ支配が依存、参入、収益構造を変えるなら、具体的な制度水準を示す必要がある。帝国に似た統治構造と記述する重点は、修辞的な国家類推ではなく、規則制定、紛争処理、本人管理、制裁の能力にある。従来の国家と異なり、権威の源は契約とインフラ依存である。領土主権はなくても越境的規模を持ちうる。比較の最も有益な点は、異なる統治能力の源泉と限界を明らかにすることである。】
- Varoufakis, Yanis. Technofeudalism: What Killed Capitalism. Bodley Head, 2023. (Read: Chapter 3, "Cloud Capital," pp. 58–91; Chapter 4, "The Rise of the Cloudalists and the Demise of Profit," pp. 92–115; Chapter 7, "Escape from Technofeudalism," pp. 175–214).
【注:Varoufakis の中心命題は、クラウド資本が市場競争と利潤の従来の関係を変え、プラットフォームの入口、データ、クラウド・インフラを通じて新しい依存構造を作るというものだ。有益な問いは「技術封建主義」という語を受け入れることではなく、地代、ネットワーク効果、独占、利用者依存を説明できるかである。独占資本やプラットフォーム資本主義でなお説明できるなら、封建主義概念は何を加えるのか。逆に、従来の資本理論では説明しにくい権力関係を示すのか。精読ではクラウド資本と通常のデジタル資本を区別したい。「企業が巨大だから封建的だ」という論証ではない。行動を形づくり、取引を組織し、生態系の参加者からクラウド地代を取るインフラを蓄積することで、市場自体が私的プラットフォーム内部でますます編成されるという議論だ。第3・4章は連続して読む。前者がクラウド資本を定義し、後者がクラウド資本家と利潤/地代関係の変化を提起する。第7章は時代診断だけでなく、退出と政治的再編に向き合う必要を示す。Amazon Marketplace、アプリストア、クラウド・サービスなどを比較し、同じ構造なのか検証する。命題が説明しようとするのは、プラットフォームが資本主義市場の企業にとどまらず、クラウド・インフラと利用者行動を通じて取引環境を形成し、依存から地代を得ることだ。争点は「封建」が目立つ語かではなく、利潤、地代、市場、プラットフォーム権力の関係が構造的に変わったかである。従来の独占理論がすでに収益を説明できるなら、技術封建主義の追加的説明力を明示的に立証しなければならない。】
- Durand, Cédric. How Silicon Valley Unleashed Techno-Feudalism: The Making of the Digital Economy. Translated by David Broder. Verso, 2024. Originally published as Technoféodalisme: Critique de l'économie numérique in 2020. (Read: Chapter 2, "On Digital Domination"; Chapter 3, "The Rentier Class of the Intangible"; Chapter 4, "The "Techno-Feudal" Hypothesis").
【注:Durand の構成は Varoufakis との併読に適している。第2章はデジタルな支配、第3章は無形資産と金利生活者階級を扱い、第4章ではじめて技術封建主義仮説を体系的に示す。支配機制、蓄積形態、社会構成仮説という三水準へ分解できる。地代をそのまま封建的地代と同一視せず、マルクス主義の地代理論、独占資本主義、プラットフォーム資本主義との差を比較したい。似た語彙は異なる道筋を隠す。Durand はデジタルな支配、無形資産、知的財産、レントの論理が蓄積を再編する仕方を重視し、資本主義の内的発展と潜在的な形態転換の境界に技術封建主義を置く。第2〜4章が段階的に進む以上、「封建仮説」へ直接飛んではならない。まず支配と無形資産のレントが立証されたかを判断し、その後で社会構成を名づける根拠になるかを論じる。二冊を並べる目的は勝者を選ぶことではなく、「封建性」の判準を比較することだ。Durand の How Silicon Valley Unleashed Techno-Feudalism: The Making of the Digital Economy、Varoufakis の Technofeudalism: What Killed Capitalism、Thorhauge の The Economic Lives of Platforms: Rethinking the Political Economy of Digital Markets は、「惑星的モデルと具体的場所、労働、維持との尺度上の緊張」を巨視的判断から機制へ下ろす。惑星規模の概念は具体的場所を消しやすい。Durand の議論は時代命名としてより、デジタル支配はどう生じ、無形資産はどう地代を生み、それらは新しい社会構成をなすに足るか、という三問に分けて扱いたい。「封建主義」は既定の結論でなく、検証すべき政治経済的仮説となる。プラットフォーム依存、知的財産、独占地代の関係は、別々に分析する必要がある。】
- Thorhauge, Anne Mette, Andreas Lindegaard Gregersen, Eva Iris Otto, Jacob Ørmen, and Morten Axel Pedersen, eds. The Economic Lives of Platforms: Rethinking the Political Economy of Digital Markets. Bristol University Press, 2024. (Read: "Rethinking the Political Economy of Digital Markets," pp. 1–16; "Modes of Connectivity," pp. 17–36).
【注:この論集は技術封建主義論争に、より細かな比較資料を与える。包括的な社会構成概念からでなく、接続、市場設計、データ、媒介関係を通じてプラットフォームが経済活動をどう組織するかを見る。Varoufakis、Durand の後に読むことで、「クラウド地主」「地代」といった大きな語を研究可能な機制へ分解できる。どの入口を支配し、どう価格を決め、需給を組織し、どこに実際の退出費用があるのか。経済的生命は「市場仲介」だけでは括れない。接続方式、価格設定、データ蓄積、規則統治が価値の創造と配分を共同で決め、プラットフォームの種類にも大きな差がある。論集の意義は、政治経済学を単一の事業モデルから比較分析へ進め、少数の巨大企業を経済全体の代わりにしない点にある。接続様式は、誰が取引に入り、どんな身分で参加し、どの関係がデータ化されるかも決める。価格とネットワーク効果だけでは、接続様式自体が制度的配置であることを見落とす。接続を設計された関係と捉えると、市場構造が技術的インターフェースへ書き込まれる仕方を分析できる。政治経済学は、財務諸表や市場占有率に権力を探すだけでなく、インターフェースが経済関係を制度化する方法を研究すべきだ。接続関係自体が制度設計であり、取引以前の技術的前提にとどまらない。これは通常の市場仲介者との重要な違いである。参加資格、取引単位、見える情報を定義する時点で、技術的インターフェースのなかに市場関係はあらかじめ組織されている。権力は市場形成の前から作用する。】
- Mezzadra, Sandro, Niccolò Cuppini, Mattia Frapporti, and Maurilio Pirone, eds. Capitalism in the Platform Age: Emerging Assemblages of Labour and Welfare in Urban Spaces. Springer, 2024. (Read: "Introduction. The Platform Age," pp. 3–14; "Operations of Platforms," pp. 15–31; "The Politics of Platforms," pp. 81–95).
【注:このオープンアクセス論集は、古典的デジタル労働論から近年のプラットフォーム労働研究へ進む手がかりとなる。プラットフォームを労働、都市空間、福祉、社会的協力を同時に組織するインフラと捉え、価値増殖、アルゴリズムによる管理、労働者の主体性、代替案を一つの枠組みに置く。Scholz、Woodcock と併読し、プラットフォーム協同組合主義が所有権を変えるだけか、労働組織と社会的再生産をもっと広く変えるのか検証できる。プラットフォームを国家、帝国、領土と呼ぶには慎重さが要る。データセンター、都市、国境、供給網、保守現場では、巨視的構造が実際にどう実行され、各地域の環境・労働費用が対称かを確認できる。Capitalism in the Platform Age は都市、福祉、労働再生産の交差点へ研究を進める。プラットフォームは独立したアプリでなく、交通、配送、ケア、公共サービスを再編し、労働市場と都市空間をともにプラットフォーム化しうる。分析では企業評価額やデータ抽出だけでなく、地方制度の書き換えと社会的再生産費用の再配分を追うべきだ。都市の次元は、プラットフォーム資本主義と地方国家が代替し合うだけではないことを示す。配送、配車、短期賃貸、ケアは、交通規則、住宅政策、労働法、福祉制度と摩擦を起こす。拡張は地方制度の差異を利用しつつ、地方政府に規制対象の再定義を促す。福祉制度への影響は特に重要である。仕事が臨時の課題や複数プラットフォームからの収入へ分割されるほど、安定雇用に基づく保険、課税、労働保護は適用しにくくなる。企業組織の革新だけでなく、福祉国家に労働の身分と社会権の再定義を迫るのである。都市はその衝突が最初に現れる制度的尺度となる。】
- Morozov, Evgeny. "Critique of Techno-Feudal Reason" New Left Review, nos. 133/134 (2022). (Read: entire essay).
【注:Morozov の技術封建主義批判は、本節内の反対論に適している。デジタル独占、知的財産、プラットフォーム地代、金融化を、資本主義が封建主義へ席を譲った証拠と早急に解釈することを疑い、競争、利潤、国家政策の継続的な作用を再検討させる。二冊の技術封建主義論に議論を引っ張られず、その概念を反証・比較可能な仮説にできる。「Critique of Techno-Feudal Reason」は、「地代、利潤、入口支配、退出費用の政治経済的差異」を修辞的命題から制度分析へ変える。その差異は単一の結論へ向かわず、組織ごとの権力配置を比較し、概念の説明範囲と失効条件を判断するよう求める。現代資本主義を早急に「封建主義」と宣言すると、資本、競争、金融、国家がなお作動する機制を覆い隠しかねないという、必要な反論である。概念の新しさは歴史的時代区分の証拠を代替しない。有益な論争は、同じ現象、すなわちプラットフォーム地代、独占、インフラ依存に各理論がどれだけ正確な因果説明を与えられるかを比較する。また歴史的類推は政治戦略も左右する。権力を封建領主の関係と理解すれば解決は「資本主義的競争の回復」に傾きうる。資本集中と私有インフラが問題なら、反独占、公的所有、労働政治の優先度は異なる。概念選択は名づけ遊びでなく、政策的想像を組織するのである。】
- Mosco, Vincent. To the Cloud: Big Data in a Turbulent World. Paradigm Publishers, 2014. (Read: Introduction; chapters on cloud computing, big data, and political economy).
【注:Mosco は政治経済学とコミュニケーション研究から、クラウドの産業構造、労働、データ、イデオロギーを脱神秘化する。スタックや技術封建主義の論述に先立つ歴史的節点として有用だ。クラウド計算は最初から単なる技術の外注ではなく、企業集中、データセンター、エネルギー、労働、国家政策と絡み合っていた。惑星規模の概念は具体的場所を消しやすい。概念の適用範囲を最大化するより、理論の反証可能性を保ちたい。経験的資料が核心的機制を支える場合にのみ、枠組みを延長すべきである。「惑星的モデルと具体的場所、労働、維持との尺度上の緊張」は、Mosco の To the Cloud: Big Data in a Turbulent World の外にある資料と競合説明にも検証されなければならない。脱神秘化とは、クラウドが場所をなくしたのでなく、言葉によってサーバー、データセンター、電力、労働を利用者の経験から隠したと示すことだ。イデオロギー的効果と物質的構造を同時に研究し、軽く遍在的なサービスとして現れるほど、固定された土地、エネルギー、資本投入を追う必要がある。集中には規模の経済もある。大規模センターは資本、電力調達、運用技術から費用優位を得るため、随時利用の便利さと背後の高度な集中が同時に拡大する。脱神秘化は言葉だけでなく産業構造の問題であり、クラウドが遍在的な公共環境に見えるほど、所有権は覆い隠されやすい。Mosco はさらに世界的通信政治へ戻す。センターとサービスの集中は少数企業に産業横断的な基盤の地位を与え、既存メディア、政府、一般企業もその顧客となる。生じるのは「インフラ依存」であり、政治的問題はプライバシーだけでなく、サービス中断、価格、技術標準の支配権にもある。】
第3部:媒介、組織、労働(第5–6週)
第5回:再帰的公共、媒介、デジタル政治
本節は支配と抽出から、公共性の制度的条件へ移る。Kelty の再帰的公共は「オンラインで議論する集団」ではなく、自らの技術的・制度的基盤の維持、変更、再生産に参加できる公共の形式である。自由ソフトウェア共同体の理論的意義はコードの公開だけにない。ライセンス、協働規範、討議手順、バージョン管理、インフラがともに、再帰的に変更できる公共生活を構成するからである。Warner の公共と対抗的公共の分析は、公共があらかじめ存在する固定集団ではなく、伝達、応答、再伝達によって絶えず呼び集められることを強調する。Fraser はさらに、参加の不平等と対抗的公共を公共圏理論の内部に置く。
Kornbluh の Immediacy, or The Style of Too Late Capitalism は、この議論にメディア形式の緊張を与える。即時性、透明性、没入、「媒介のなさ」は、デジタル文化の自然な価値とされがちだ。しかし政治組織は、手続き、代表、遅延、説明、記憶、責任の連鎖にこそ依存する。Tufekci、Dean、Gerbaudo らのネットワーク運動研究も、低費用の動員と高速な伝達が、持続的な組織能力へ自動的に変わるわけではないと示す。接続は参加規模を急速に広げる一方、意思決定、資源蓄積、長期的調整を脆弱にもする。
したがってデジタル政治の核心は、「中央集権化」か「分散化」かという抽象的選択ではなく、参加者がどの媒介を理解し、責任を問い、変更できるかを判断することにある。公共が自前のサーバーを持つかは唯一の基準ではない。より重要なのは、規則制定、資源配分、可視性の構造、紛争解決手続きへ影響できるかである。メディア条件と組織形式の両方が政治的に問えるときにこそ、公共性はプラットフォームが提供する一時的な注意の空間へ退化せずに済む。再帰的公共はこうして、メディア技術、制度設計、集団的自治をつなぐ理論的結節点となる。
必読文献:
- Kelty, Christopher M. Two Bits: The Cultural Significance of Free Software. Duke University Press, 2008. (Read: Chapter 1, "A Recursive Public," pp. 27–63).
【注:本節の中心文献である。Kelty の再帰的公共は公共メディアを利用するだけでなく、そのインフラ的・制度的条件を変更、維持、再生産できる公共の形式を指す。ライセンス、コード、討議、維持、アイデンティティが共同で公共性を作る仕方に注目する。プラットフォーム時代と比較すると、コード、サーバー、アカウント、規則への実効的支配を失った後、「公共性」に何が残るかが問われる。逆に、公共性の持続には技術インフラへの一定の自治権が必要なのか。人数や伝達速度で直接測ることはできない。高い可視性があっても、記録の保存、規則の変更、インフラの所有、退出条件の決定ができない集団はありうる。これらの制度次元は表現権と区別すべきだ。Kelty の Two Bits: The Cultural Significance of Free Software は、「公共性が自らのメディア条件を維持・変更する能力を持つか」を制度関係へ具体化する。公共性は情報流量の同義語ではなく、持続する規則、記憶、手続き、メディア条件を必要とする。プラットフォームは可視性を広げるが、公共を支えるインターフェース、データ、伝達構造を私的統治の下へ置くこともある。再帰的公共は通常のオンライン討論集団でなく、交流と協働の条件そのものを変更できる公共だ。自由ソフトウェア共同体が再帰的なのは、成員がコードやライセンスを使うだけでなく、その制度・技術的基盤を維持するためである。変更できない専有プラットフォーム上でしか公共が形成できないとき、組織的自治はどう制限されるのか。ここに鋭い問題がある。再帰性は「自らを修復する」能力も意味する。ソフトウェア、ライセンス、通信経路に問題が生じた際、外部プラットフォームの決定を待つだけでなく、共通規則を通じて修正できる。技術的自治と民主的自治は完全には重ならないが、維持する権利において重要な接続を持つ。】
- Kornbluh, Anna. Immediacy, or The Style of Too Late Capitalism. Verso, 2024. (Read: "Circulation," pp. 24–43; "Writing," pp. 65–112; "Antitheory," pp. 148–190).
【注:Kornbluh の本は技術を直接論じるものではない。だからこそ、プラットフォーム時代の公共性がなぜ「直接」「即時」「没入」「無媒介」を求めるのか、という広い問いを導入できる。彼女はこの文化形式を即時性と呼び、表象、距離、媒介への許容度を下げると指摘する。再帰的公共にとって重要な緊張が生じる。公共性には規則、手続き、討議、維持が要るのに、即時性文化はそれらを障害と感じやすい。政治的伝達が即時反応を求めるほど、組織に必要な遅延、説明、制度手続きはどうなるか。「Circulation」「Writing」「Antitheory」は連続して読む。即時性は単一のメディア現象でなく、流通、一人称表現、没入、透明性、反理論的態度を結ぶ後期資本主義のスタイルである。論争的な強さ自体も批判の対象としたい。媒介の減少は常に政治的後退なのか。周縁的主体が直接の発言を求めるとき、表象・代表の制度的価値をどう論じるべきか。Warner、Fraser、Kelty の間に置くのは、媒介の一面的美化を避けるためであり、必要な媒介、責任を問える媒介、独占的媒介を区別したい。媒介は本来的に反民主的ではない。可視化し、責任を問い、組み直せるかが問題だ。即時性批判は文化形式と政治組織を結ぶ。直接性、没入、一人称、即時の流通が価値となると、媒介は疎外や遅滞と理解されやすい。しかし公共政治は、代表、手続き、説明、組織という除去できない媒介に依存する。デジタル伝達の速度には二面性がある。応答能力を広げると同時に、持続的判断を作る距離と時間を侵食しうる。】
- Warner, Michael. "Publics and Counterpublics" Public Culture 14, no. 1 (2002): 49–90.
【注:Warner は、公共がメディア以前の固定集団ではなく、伝達、応答、再伝達を通じて絶えず呼び集められることを強調する。推薦、再投稿、グループチャット、購読、アルゴリズムによるタイムラインが公共形成を変えるプラットフォーム時代に特に重要だ。Kelty と併読して、技術的条件が公共性の生産へどう参加するかを論じられる。低費用の動員と組織能力は同じ指標ではない。急速な集合は争点を増幅しても、信頼、資源、役割分担、紛争処理を必ずしも蓄積しない。動員、組織、制度を区別しつづけたい。「Publics and Counterpublics」は、公共、対抗的公共、ネットワーク化された公共、再帰的公共の違いを説明する、限定的だが有効な尺度を与える。機制を示すが、組織、法、物質条件の追加分析を代替しない。公共は先在する人々ではなく、文章の流通と見知らぬ者同士の持続的応答によって召喚される。公共性はメディアに依存し、伝達のリズム、呼びかけ方、入手可能性が「誰が公共をなすか」を変える。タイムラインはさらにこの過程へ介入し、参加者の行為だけでなく順位づけと推薦も形成を左右する。また対抗的公共は、主流と異なる表現や社会的リズムで自らを組織する。プラットフォームは形成の伝達費用を下げる一方、統一的な内容形式と指標ですべての表現を測る。標準化と対抗的公共の差異の緊張は、ネットワーク上の公共性を理解するうえで見逃せない。】
- Fraser, Nancy. "Rethinking the Public Sphere: A Contribution to the Critique of Actually Existing Democracy" Social Text, no. 25/26 (1990): 56–80.
【注:Fraser は従属的対抗公共、参加の対等性、アイデンティティの差異を通じて古典的公共圏理論を修正する。統一的公共圏を唯一の理想にせず、複数の公共、対抗的公共、周縁的公共が共存する仕方を考えられる。本節の文脈に「Rethinking the Public Sphere: A Contribution to the Critique of Actually Existing Democracy」を置くと、「表現、インフラ所有、民主的統治を混同してはならない」という論点は具体的構造を得る。公共性は情報流量の同義語でなく、持続する規則、記憶、手続き、メディア条件を必要とする。プラットフォームは可視性を広げるが、公共を支えるインターフェース、データ、伝達構造を私的統治の下へ置くこともある。対抗的公共の概念は「一つの共通公共圏」という規範的想像も修正する。社会的不平等のため、討議へ入る資源、言語、可視性は常に不均等であり、周縁集団はまず自前の対抗公共空間を必要としがちだ。デジタル網は分化の可能性を広げると同時に、公共同士の出会いを順位づけと商業論理で形づくる。参加の対等性の強調は、形式的開放だけでは平等を保障できないと示す。資源、地位、表現習慣は誰が聞かれやすいかを左右する。発言の敷居を下げても、注意の競争が新たな不平等を生みうる。複数公共の存在は、実際の参加条件の差とともに検討すべきだ。対抗的公共の価値は、全員に同じ言語と表現規範への変換を求めず、公共性の内部に差異を保つ点にある。】
推薦文献:
- Gramsci, Antonio. Selections from the Prison Notebooks. Edited and translated by Quintin Hoare and Geoffrey Nowell Smith. International Publishers, 1971. Originally published in Italian selections from the prison notebooks, 1948–1951. (Read: "The Intellectuals," pp. 5–23).
【注:プログラマーを単に「有機的知識人」と呼ぶためでなく、知識、生産、組織の結びつきを捉え直すために読む。Gramsci は知識人が特定の社会勢力から生じ、文化、組織、ヘゲモニー形成で果たす役割を論じる。デジタル環境では、ソフトウェア技術者、プロトコル開発者、管理者、データ労働者、研究者の知識上の位置を別々に調べるべきで、「技術を知る人」が自然に特定の政治階級を代表するとは想定しない。公共性は人数や伝達速度では直接測れない。Gramsci の Selections from the Prison Notebooks によって、「自らのメディア条件を維持・変更する能力があるか」を巨視的判断から機制分析へ移せる。概念の成立は、制度関係を経験的に識別できるかにかかる。有機的知識人は「賢い人」へ特別な地位を授けるのでなく、知的労働と社会集団、組織的実践の結びつきを示す。デジタル社会では、技術者、研究者、技術共同体、労働組織が共通の世界観と組織能力の形成へ参加する仕方を分析するのに適している。専門性自体は政治的立場を決めない。知識がどう組織され、どの社会勢力に奉仕し、集団的実践へ変わるかが鍵だ。技術的権威と政治的代表も区別すべきである。システムの知識を持つだけで、他集団の公共利益を定める資格が自動的に得られるわけではない。専門知は組織、交渉、共同行為に入ってこそ集団的政治能力の一部となる。AI 専門家による統治では特に重要だ。また知識人と組織には双方向の関係がある。前者は社会的経験を組織し、後者は知的活動へ持続的な社会的位置を与える。短期的なオピニオンリーダーだけに依存する公共性には、この長期的機能を形成しにくい。技術共同体、労働組織、公共研究機関がどう知を蓄積し、組織的記憶を伝えるかが「新しい有機的知識人」の問題の核心である。】
- Habermas, Jürgen. The Structural Transformation of the Public Sphere. Translated by Thomas Burger. MIT Press, 1991. Originally published 1962. (Read: Part V, pp. 141–180).
【注:公共圏理論の歴史的基準として読む。プラットフォームを Habermas のマスメディアと同一視するのでなく、公共圏が必要とする制度、メディア、社会的条件と、商業化によるその変化を理解する。媒介は本来的に反民主的ではない。見えるか、責任を問えるか、再構成できるかが問題だ。歴史分析はなお重要な基準を与える。公共性は情報流通だけでなく、特定の制度空間、メディア形式、社会的資格に依存する。プラットフォーム時代も「誰でも発言できる」に還元できない。事業モデル、所有権、注意の配分、不平等が討議をどう組織し、どの制度が批判的交流を保てるかが重要である。商業化はデジタル環境で新しいインフラ形式を取る。従来のマスメディアは主に所有権と広告を通じて討議へ影響したが、プラットフォームは身元、順位、伝達、フィードバックまで支配する。理論は内容生産から配信構造へ広がり、私的技術システムが交流条件をどう規定するかを論じる必要がある。「誰が伝達インフラを支配するか」は避けられず、所有権と技術構造を同時に論じるべきだ。物質的条件は新聞、喫茶店、放送から、サーバー、インターフェース、推薦へ延び、メディア構造は公共圏の制度の一部となる。公共圏を再論することは古いメディア秩序の復元でなく、新しいインフラが対等な参加、争点形成、公共的判断へ及ぼす影響の分析である。規範枠組みは断片化されたネット公共を直接包摂できなくても、参加条件は平等か、議題は私益の支配を離れられるか、理由は公開の争論に付せるか、という尺度を与える。インフラと結びつければ、歴史的社会モデルを復元せずとも批判的効力を保てる。】
- Benkler, Yochai. The Wealth of Networks. Yale University Press, 2006. (Read: Chapter 3, "Peer Production and Sharing," pp. 59–90).
【注:Benkler は対等な共同生産を制度・経済の両面から分析する。Kelty と併読して、コモンズに基づくピア生産が企業や市場と異なる組織形式を提供するか、その公共的成果がプラットフォームにどう再吸収されるかを論じられる。低費用の動員と組織能力は同じ指標ではない。Benkler の The Wealth of Networks を通じ、公共、対抗的公共、ネットワーク化された公共、再帰的公共の違いは組織・制度の次元へ入る。政治的可能性はネットワーク自体からでなく、生産協力と排他的所有権の関係の変化から生じる。公開ライセンス、モジュール型協働、共有資源は参加の敷居を下げるが、統治規則、維持労働、持続可能な物質的基盤をなお必要とする。ここでプラットフォーム化との対照が鮮明になる。開かれた協働はプラットフォームの無償労働と等しくない。「非市場的協力」と「企業への無償貢献」も区別すべきだ。前者には共有規則、共通資源、参加者が生産過程へ影響できる制度が通常あり、後者ではインフラと収益が企業の一方的支配下にありうる。どちらもネットワーク協働を使うが、所有権と統治構造はまったく異なる。オープンソース・モデルでは特に重要である。重みを無料ダウンロードできても、訓練、統治、導入を単一企業が決めるなら、コモンズ型ピア生産との間には重要な制度的距離が残る。】
- Coleman, Gabriella. Coding Freedom. Princeton University Press, 2012. (Read: Chapter 3, pp. 61–84).
【注:Coleman の民族誌は自由ソフトウェアとハッカー文化を、抽象的な「技術的自由」から具体的な共同体へ移す。技術的能力、倫理、政治的アイデンティティ、組織文化の関係に注目する。Coleman の Coding Freedom が示す「表現、インフラ所有、民主的統治を混同してはならない」という論点は、文献内部の概念論争にとどまらず、Coleman の Coding Freedom の外にある経験的資料と競合説明にも検証されなければならない。Coleman の Coding Freedom はこの区別を特定の対象と因果連鎖に限定する。その外への類推には新たな証拠が必要で、用語だけでは済まない。Coleman の Coding Freedom は自由ソフトウェア倫理を具体的な実践に戻す。技術力、ユーモア、法意識、ハッカー倫理、組織の境界が絡み合う。公共性は公開ライセンスだけでなく、成員がコード、規則、共通のアイデンティティへ持続的に関わることからも生まれる。民族誌が特に重要なのは、政治がライセンス本文から自動的に導かれるわけではないと示すためだ。ユーモア、技芸、評判、法意識、日常的協働が共同で倫理を形づくる。技術的自治は実践と組織によって作られる文化形式であり、コードを開けば自然に現れる政治的属性ではない。また共同体の政治性は、壮大な宣言より、法の細部や日常規範として現れることが多い。著作権、匿名性、言論の自由、協働義務の具体的理解が行動を形づくる。自由ソフトウェア文化は制度化された倫理であり、安定性は単一のイデオロギーの共有でなく長期の実践に由来する。】
- Tufekci, Zeynep. Twitter and Tear Gas. Yale University Press, 2017. (Read: Introduction and Chapters 1–2).
【注:Tufekci は、ネットワーク・メディアが動員費用を下げても、長期の組織能力を必ずしも高めないことを論じる。Kornbluh と併読したい。即時性は高速伝達に有利だが、政治組織には記憶、信頼、資源蓄積、手続きが要る。公共性は人数や速度では直接測れない。Tufekci の Twitter and Tear Gas の尺度では、「自らのメディア条件を維持・変更する能力があるか」は具体的な諸関係として現れる。Tufekci の Twitter and Tear Gas による説明が成立するのは、対応する制度的機制を経験的に識別できる場合である。Tufekci の Twitter and Tear Gas を通じ、問いは厳密になる。どの機制が本当に変化し、どれが媒体の変更にすぎず、誰が帰結を負うのか。Tufekci の Twitter and Tear Gas はさらに、対象差と制度的境界を保ち、概念の修辞的強さで経験的説明を代替しないよう促す。Tufekci の Twitter and Tear Gas は、運動が極短時間で大きな伝達規模を得ても、交渉、資源管理、内部紛争処理の能力を同時には作れないことを示す。「能力不足」は動員の否定でなく、組織基盤と伝達速度の非対称性の記述である。ネットワーク化された抗議では、デジタル・メディアが費用を大幅に下げ、規模と可視性を急速にもたらすが、交渉、決定、紛争処理の能力は必ずしも伴わない。従来の組織がゆっくり蓄えた経験こそ、急速な動員に欠ける部分かもしれない。運動の「脆弱性」は伝達の失敗でなく、組織能力と伝達能力の非対称な成長から生じうる。】
- Dean, Jodi. Crowds and Party. Verso, 2016. (Read: Chapter 3, "The Distributive Party," pp. 119–156).
【注:Dean は「接続が多いほど政治組織は強い」という前提を疑う。ネットワーク上の伝達は個人化された多数の集合を生んでも、持続する組織を必ずしも生まない。Kelty と併読し、再帰的公共とネットワーク上の群衆を区別できる。Dean の Crowds and Party、Tufekci の Twitter and Tear Gas、Gerbaudo の Tweets and the Streets: Social Media and Contemporary Activism は、「即時性・伝達速度と、組織に必要な媒介・遅延との衝突」を異なる知の伝統に置く。媒介は本来的に反民主的ではない。可視化し、責任を問い、再構成できるかが問題である。Dean の Crowds and Party はこの衝突を抽象的すぎる時代判断から解放する。Dean の Crowds and Party の貢献は、理論の反証可能性を保つ点にある。Dean の Crowds and Party が強調するのは動員の単純な否定でなく、一時的な群衆の集合と、決定、規律、継続行動を担う組織形式との構造的差異だ。デジタル・メディアは前者を増幅しても、後者を自動的には生まない。組織の問題を強調すると、「ネット接続」は自然に民主的だという後光を失う。急速な伝達は共通感情と短期集合を作れても、決定、資源配分、長期責任を担う組織を必ずしも形成しない。群衆と党の緊張は、デジタル行動主義の理解に欠かせない。政治能力には表現とともに、持続的な組織形式が必要なのである。】
- Gerbaudo, Paolo. Tweets and the Streets: Social Media and Contemporary Activism. Pluto Press, 2012. (Read: Chapter 1, pp. 1–20).
【注:Gerbaudo の指導者なきネットワーク論は、ソーシャルメディアが政治的可視性と指導権を再配分する仕方を分析する。運動に「民主主義がない」と証明するのでなく、インターフェースが新たな中心を生みうると示す。低費用の動員と組織能力は同じ指標ではない。Gerbaudo の Tweets and the Streets: Social Media and Contemporary Activism を中心に、公共、対抗的公共、ネットワーク化された公共、再帰的公共の差を関連する諸段階へ分け直せる。「指導者なきネットワーク」の修正は特に重要だ。ソーシャルメディアは指導を消さず、アカウントの可視性、感情の編成、行動の呼びかけを通じて新しいソフトな指導を作る。形の上では分散していても、注意と議題設定は高度に集中しうる。自己説明だけで分散度を判断せず、見えにくい編集、調整、代表の機制を分析する必要がある。「集会の振付」は、オンライン伝達と現地集会の間に具体的な組織労働があることを示す。場所、時間、標語を誰が選び、情報流を誰が保つかが、「自発的」な運動を形づくる。組織は消えたのでなく、一部の労働が編集、再投稿、感情的動員として見えにくくなったのだ。ソフトな指導は新しい代表問題にも役立つ。高可視性のアカウントは行動を有効に調整する一方、正式な委任なしに運動を解釈する権威を得ることもある。伝達上の優位を疑えない代表権へ自動転換しない、新たな説明責任が必要だ。指導と水平的参加は二者択一でなく、組織設計の問題である。】
- Milan, Stefania. Social Movements and Their Technologies. Palgrave Macmillan, 2013. (Read: Chapter 2, pp. 31–54).
【注:社会運動が通信・メディアのインフラを自ら作る仕方を論じる。再帰的公共の問いを政治的行為者へ広げ、「メディアを所有すること」と「使うこと」に大きな差がありうると示す。Milan の Social Movements and Their Technologies、Gerbaudo の Tweets and the Streets: Social Media and Contemporary Activism、Juris の Networking Futures は、「表現、インフラ所有、民主的統治を混同できない」という判断を巨視的水準から機制へ下ろす。歴史的類推は機制の水準でこそ有益だ。アルゴリズム、プラットフォーム、モデルが登場しただけで時代交代を宣言すれば、連続と断絶を区別する理論の力は失われる。Milan の Social Movements and Their Technologies は、その巨視的判断を識別可能な制度機制へ戻す点で貢献する。この区別は、Milan の Social Movements and Their Technologies の外にある経験的資料と競合説明にも検証されなければならない。運動の技術の要点は、どの媒体を選ぶかだけでなく、伝達条件に対する相対的自治を作れるかにある。自前のサーバー、暗号化通信、独立メディア、技術ボランティアは、運動自身のインフラ能力となりうる。Milan はメディア利用を統治の問題へ進める。政治組織が技術条件をどこまで支配するかは、行動の持続性へ直接影響する。】
- Juris, Jeffrey S. Networking Futures. Duke University Press, 2008. (Read: Chapter 1, pp. 1–26).
【注:グローバル化への抵抗運動を事例に、「ネットワーク型組織」が実際の行動から生じる様子を観察し、純粋な技術構造への還元を避ける。公共性は人数や伝達速度では直接測れない。Juris の Networking Futures からは、「自らのメディア条件を維持・変更する能力があるか」について明確な因果の層を得られる。Juris の Networking Futures の説明は、対応する制度機制を経験的に識別できる場合に成り立つ。Juris の Networking Futures を中心に追うべきなのは、すべてを覆う一語ではなく、機制同士の接続である。Juris の Networking Futures の民族誌は、ネットワーク組織が技術的位相の政治形式への直接投影ではないことを示す。会議、信頼、資源、行動の伝統、組織間交渉が共同でネットワークを保ち、水平主義にも実際の調整と非公式な指導が常に伴う。反グローバル化運動の民族誌は、ネットワーク化が技術構造であると同時に組織規範と政治文化でもあると示す。水平的協議、自治、多様性は会議、メーリングリスト、行動網で反復実践され、調整費用と非公式権力も伴う。民主性は節点の接続から自動発生する属性でなく、維持すべき制度的成果なのだ。水平主義の困難も示される。非公式ネットワークは権力を自動的に消さず、技術習熟、資源、言語、移動能力がなお中心を作る。組織分析では正式構造と非公式階層をともに見る必要がある。水平主義は達成済みの組織的事実より、持続的実践の規範として扱うのがよい。】
- Papacharissi, Zizi. Affective Publics: Sentiment, Technology, and Politics. Oxford University Press, 2015. (Read: Introduction; selected chapters on affective publics and networked storytelling).
【注:Papacharissi はネットワーク上の公共性における情動、感情表現、物語構造を中心に置く。公共は合理的論証と情報交換だけでなく、情動のリズム、タグ、可視性、物語の流通によって一時的に組織される。Warner と Fraser の後に読み、過度に制度的・規範的な傾向を補える。同時に、ここでは公共形成を扱い、次節の Berlant は不安定な生活における感情的愛着を扱うという違いも保つ。媒介は本来的に反民主的ではなく、見えるか、責任を問えるか、再構成できるかが問題だ。情動的公共は情動を公共的理性の単なる対極から解放する。タグ、画像、物語の流通は、分散した個人を感覚を共有する一時的公共へ組織し、感情の強度自体が政治的接続になる。それが持続的組織へ転換できるか、可視性と相互作用の最適化が特定感情を選択的に増幅するかが問われる。情動は必ずしも浅薄な政治へ向かわない。記憶、アイデンティティ、共通の関心を保つ一方、短期の注意機制に消耗させられもする。形成と制度化を区別するのに適した枠組みであり、前者は接続の成立を説明し、後者には組織論の補足が要る。持続性は結局、組織、記憶、制度的支えに依存し、感情の強さだけで長期の政治能力を代替できない。流通には明確なプラットフォーム依存もある。短文、視覚的衝撃、反応指標で広がりやすい感情があり、順位づけは高反応の内容をさらに増幅しうる。情動的公共は自発的な心理的集合ではなく、メディア形式とアルゴリズムの可視性によって形づくられる。デジタルな政治的感情の理解には、このインフラ条件を含める必要がある。】
- boyd, danah. "Social Network Sites as Networked Publics: Affordances, Dynamics, and Implications" In A Networked Self: Identity, Community, and Culture on Social Network Sites, edited by Zizi Papacharissi, 39–58. Routledge, 2011. (Read: entire chapter).
【注:boyd はネットワーク化された公共を用い、持続性、複製可能性、拡張可能性、検索可能性が公共的相互作用をどう変えるかを論じる。Facebook や Twitter が民主的公共圏を自動生成すると宣言するのでなく、メディアのアフォーダンスが発言の時間性、可視性、受け手の境界、再伝達を変える仕方を説明する中範囲の概念を与える点が重要だ。Warner の循環的伝達と併読すると有効である。低費用の動員と組織能力は同じ指標ではない。「Social Network Sites as Networked Publics: Affordances, Dynamics, and Implications」によって、公共、対抗的公共、ネットワーク化された公共、再帰的公共の違いを修辞的命題から制度分析へ移せる。枠組みは、アフォーダンスが公共的交流を形づくることを強調する。持続性、複製可能性、拡張可能性、検索可能性は、表現の時間、範囲、検索されうる性質を変える。ソーシャルネットワークは従来の公共圏の単なるオンライン版ではない。保存される可視性、複製できる内容、曖昧な受け手の境界により、身元管理、文脈崩壊、公共参加が同時に生じる。受け手の想像も変わる。発言者が向き合うのは明確な集団でなく、潜在的で見えず、アルゴリズムで拡大可能な受け手であることが多く、自己提示とリスク判断を変える。文脈崩壊は個人のプライバシー管理の下手さではなく、複数の社会的場面が一つの画面へ技術的に圧縮されることだ。ネット公共の社会的緊張は、こうしてインターフェース設計へ入る。】
- Gillespie, Tarleton. Custodians of the Internet: Platforms, Content Moderation, and the Hidden Decisions That Shape Social Media. Yale University Press, 2018. (Read: Chapter 2, "The Myth of the Neutral Platform," pp. 24–44; Chapter 5, "The Human Labor of Moderation," pp. 111–140).
【注:本書は政策と技術規則による内容統治を分析すると同時に、モデレーションが大量の隠れた人間労働に依存すると強調する。「アルゴリズムが自動統治する」という想像を修正し、実際にはアルゴリズム、規則、人間の審査、商業的リスク管理が共同で統治することを示す。内容統治は準公共制度の性格を持つが、公共機関に通常求められる民主的委任と手続き上の制約を欠く。規則制定、自動検出、人間の審査が可視性の境界を共同で決め、企業は規則解釈とインフラの双方を握る。Gillespie は「プラットフォームの責任」を、削除するかどうかだけでなく、統治の正当性そのものへ向ける。内容管理は私的プラットフォームの制度的逆説も露わにする。中立的仲介者と称しながら、公共秩序に似た規則を作りつづけなければならない。審査規模のため自動システムと世界規模の人間労働を併用し、規則解釈は継続的な組織過程となる。論争は「言論の自由対検閲」という一つの対立に還元できない。モデレーションは内容政策、労働制度、インフラ統治の問題であり、三者が公共性の実際の境界を定める。私企業であると同時に事実上の交流インフラでもあるという身分の緊張が、論争の根にある。また規模は「規則と事例」の恒常的摩擦を生む。統一政策は言語、文化、政治的状況の差に直面し、自動検出は文脈を完全には理解できない。世界的な一貫性と現地判断の間で調整しつづけることになる。研究の焦点は個々の争われる決定だけでなく、この巨大制度が実行可能な分類と例外をどう生産するかに置くべきだ。例外の扱いにも規則の政治性は現れる。複雑な文脈を統一政策へ圧縮しようとするほど、多くの境界判断が要る。その判断こそ、プラットフォームが事実上の公共権力を行使する場所なのである。】
第6回:プラットフォーム労働、不安定性と新たな階級関係
本回では、プラットフォームの政治経済学を労働過程と生活時間に即して考える。プラットフォームは、仕事の割当て、評価、動的価格設定、タスクの細分化、本人確認、自動的なアカウント停止を通じて雇用関係を再編する。「上司」の権力は、常に目に見える管理者が直接行使するものではなく、インターフェース、指標、モデルとして現れうる。Scholz、Woodcock、Rosenblat、Aloisi/De Stefanoは、協同組合主義、労働者の闘争、運転手の経験、労働法という異なる角度から、アルゴリズム管理が労働関係をなくしたわけではないことを示す。情報の非対称性と責任の分散によって、労働統制はむしろ捉えにくくなる場合がある。
Terranovaのフリー・レイバー、Casilliのマイクロタスク研究、GrayとSuriによるゴースト・ワークの記述は、デジタル労働の範囲を、ユーザー参加、コンテンツ・モデレーション、データのアノテーション、学習支援へと広げる。いわゆる「自動化」は、空間的に分散し、語りのなかでは見えなくされた大量の人間労働に依存することが多い。そのため、AIシステムの作動はモデルと計算資源だけでは理解できない。プラットフォーム化は仕事のリスクを労働者個人へさらに転嫁し、待機、受注、評価の維持、機器費用の負担、いつでも働ける状態を保つことを労働過程の一部にする。階級関係は、賃金水準だけでなく、情報に関する権利、退出の条件、データ所有権、時間の支配にも現れる。
BerlantのCruel Optimismは、この経験の構造に感情と時間の次元を与える。不安定労働が持続するのは、外からの強制だけによるのではない。「柔軟性」「自由」「起業」「自己実現」といった展望が、人々の人生の物語を支える愛着の対象になりうるからでもある。したがって、プラットフォーム協同組合、労働組合、団体交渉、公共インフラが向き合うべき問題は所有権にとどまらない。リスクをどう配分し、生活時間をどう組織し、労働者が持続可能な共同の政治的主体を形成できるかも問われる。制度的な代案が「プラットフォームの所有者を替える」という限定的な想像力を超えるには、これらの次元に踏み込む必要がある。
必読文献:
- Berlant, Lauren. Cruel Optimism. Duke University Press, 2011. (Read: Chapter 1, "Cruel Optimism," pp. 23–50; Chapter 6, "After the Good Life, an Impasse," pp. 191–222; Chapter 7, "On the Desire for the Political," pp. 223–264).
[注:Berlantはプラットフォーム資本主義の直接的な理論を示すわけではないが、プラットフォーム労働がなぜ「希望」に結びつけられ続けるのかを理解する強力な概念を提供する。残酷な楽観主義とは、自らの充実した生を妨げる対象に主体が愛着を抱く関係である。第6章はこの関係を、不安定な現在、仕事と生活の時間のなかに置き、第7章は政治的欲望と行動の困難をさらに論じる。プラットフォーム労働の後で読むことで、技術的な解決策が見落としがちな問いが浮かぶ。なぜ労働者は、安全感と将来の計画を絶えず損なう「柔軟で、自律的で、起業家的な」働き方に依存し続けるのか。制度的な代案には、プラットフォームの所有権を変えるだけでなく、希望、生活のリズム、不安定性によって維持されるこの依存への対処も必要である。第1章の概念規定は、第6章の行き詰まりと第7章の政治的欲望とを併せて読む必要がある。Berlantが問うのは、有害な対象を手放せない理由だけではない。「よい生」を支えるインフラが機能しなくなった後、主体はいかに長期化した不安定な現在を即興的に生き延びるのか、また、なぜ政治的欲望は自動的に集団的な構想へ結晶しないのかである。これはプラットフォーム労働にとりわけ重要である。柔軟性はプラットフォームの言説であると同時に、ケア、移住、教育、失業という条件下では労働者の現実の必要でもありうる。愛着の物質的な合理性と構造的な代償の両方を認めてこそ、残酷な楽観主義を「労働者は洗脳されている」という単純な説明に還元せずに済む。Berlantの概念が提供するのはプラットフォーム労働の経済学ではなく、依存関係を理解する感情の政治学である。主体は、安定した生活を得る条件を損ない続ける対象であっても、「柔軟な仕事」「自主的な起業」「よい生」の約束に愛着を抱くことがある。不安定な現在は、賃金や契約だけでなく、時間感覚、希望、日々の適応によっても支えられる。この依存を見落とす労働制度改革では、不安定な形態が魅力を保ち続ける理由を説明しにくい。]
- Terranova, Tiziana. "Free Labor: Producing Culture for the Digital Economy" Social Text 18, no. 2 (2000): 33–58.
[注:Terranovaのフリー・レイバーは、デジタル文化における自発的な参加、社会的相互作用、資本の価値増殖の関係を論じる。デジタル労働を「賃金が支払われるオンラインの仕事」だけと捉えるべきではない。ユーザーの創作、交流、データ生産そのものが、資本による収奪の条件になりうる。アルゴリズム管理は管理者の消滅を意味しない。ルールの設計者、顧客対応、運営、リスク管理のチーム、資本の所有者は存在し続け、一部の命令がインターフェース、評価、自動停止を通じて実行されるにすぎない。そのため、責任をたどる際には「システムが決めた」という組織の言説を突き抜ける必要がある。「無償」の参加、賃金雇用、マイクロタスク労働は分析上区別しなければならない。さもなければ、デジタル労働という概念は強制の程度の違いを捉えられなくなる。アルゴリズム管理は管理関係をなくしたのではなく、一部の命令を仕事の割当て、評価、動的価格設定、自動停止へと変えた。インターフェースの非人格性は責任を分散させやすいが、労働者が受ける統制を弱めるとは限らない。フリー・レイバー概念の最も強い点は、自発的な参加と資本の価値増殖が同時に起こりうることを示す点にある。ユーザーは創作、交流、ゲームを楽しんでいるからといって価値を生産していないわけではない。逆に、すべてのオンライン活動を労働と呼べば、賃金関係や強制の程度の差が消えてしまう。Terranovaが提供するのは分析の連続体であり、生活全体を労働化する単純な公式ではない。さらに、資本の価値増殖と文化的自律性がゼロサムではないことも示される。デジタル文化は実際に共同体、美学、知識を生み出しながら、同時にプラットフォームによってトラフィックやデータへ変換されうる。批判の役割は参加者にとっての自律的な意味を否定することではなく、参加者が十分に統御できないインフラのなかで価値がいかに捕捉されるかを説明することである。フリー・レイバーの分析力はまさにこの二重性にある。参加は現実の自律的実践でありながら、プラットフォームの蓄積を無償で支える条件にもなりうる。]
- Scholz, Trebor. Uberworked and Underpaid: How Workers Are Disrupting the Digital Economy. Polity Press, 2016. (Read: Chapter 4, "Crowd Fleecing"; Conclusion, especially the sections on "Tactical Refusal, Defection, and Withdrawal from Data Labor" and platform cooperativism).
[注:本書の中心は「労働者版Uberを開発する」ことではなく、プラットフォーム経済のリスクを誰が負い、データを誰が管理し、プラットフォームを誰が所有するのかを問い直すことにある。第4章はクラウド・フリーシングを論じ、結論は拒否、離脱、撤退をプラットフォーム協同組合主義へ結びつける。この読み方により、協同組合主義を単なる起業に還元せず、労働者所有、団体交渉、インフラの統治に位置づけ直せる。自動化について特に避けるべきなのは代替への物神崇拝である。機械が一つの作業を代替しても、一つの職業や労働全体がなくなるわけではない。プラットフォーム協同組合主義の制度的核心は「既存のプラットフォームを複製して所有者だけ替える」ことではなく、統制権、剰余の配分、データ統治を再編することにある。協同組合が外部のクラウドサービス、資本調達、閉鎖的なアルゴリズムに依存し続ければ、その自治もインフラによって制約されうる。したがってScholzの問題意識は、所有権を超えた統治能力と技術的主権へ向かう。プラットフォーム協同組合主義には規模の問題への対処も必要である。協同所有は収益と決定権を変えられるが、ネットワーク効果、資金調達、ソフトウェアの保守、ユーザー獲得によって、協同プラットフォームも既存の巨大企業に依存せざるをえない場合がある。代案の実現可能性は、共有技術、公共調達、協同組合間の連合など、より広い制度的な生態系を築けるかにかかっている。協同組合主義の難点は企業所有から、技術スタック、資金調達、市場への接続へと広がる。プラットフォームの自治には、より十全な制度的生態系が必要である。労働者所有がインフラ能力と結びついてこそ、従来のプラットフォームとは実質的に異なる組織形態が生まれうる。協同組合の技術的条件を中立な外部資源とみなすことはできない。]
- Woodcock, Jamie. The Fight against Platform Capitalism: An Inquiry into the Global Struggles of the Gig Economy. University of Westminster Press, 2021. (Read: Chapter 3, pp. 45–67).
[注:Woodcockは、プラットフォーム労働における集団行動を、労働者の実践へと戻す。作業停止、接続の切断、組織化、労働組合、プラットフォームを横断する協働である。この視点はScholzの制度的代案の路線と緊張関係をなす。協同プラットフォームが存在しても、労働者には実際の組織能力と交渉力の構築が必要である。協同組合主義も技術的な万能薬ではない。WoodcockのThe Fight against Platform Capitalism: An Inquiry into the Global Struggles of the Gig Economyを本回の文脈に置くと、協同組合、労働組合、労働法、コモンズの統治がそれぞれ権力のどの層を変えるかという問いが、より具体的になる。アルゴリズム管理は管理関係をなくしたのではなく、一部の命令を仕事の割当て、評価、動的価格設定、自動停止に変えた。インターフェースの非人格性は責任を分散させやすいが、労働者が受ける統制が弱まるとは限らない。Woodcockは、代替的なビジネスモデルの議論だけでなく、労働者の組織実践にプラットフォームの対立を位置づけ直す。ストライキ、一斉ログオフ、労働組合、プラットフォームを横断するネットワークは、労働関係がタスク化・個人化されても集団行動が消えないことを示す。アルゴリズム管理は、かえって共通の不満の対象を生むこともある。したがって、プラットフォーム労働政治の核心は依然として組織能力、すなわち分散した経験を持続可能な集団の力へ変える方法にある。組織化には、労働過程を媒介する技術への対処も必要である。個別化された仕事の割当てと動的価格設定は共通の経験を弱めうるが、同じインターフェースが、労働者に似た不公正を発見させ、新しい組織化の道具を生み出す場合もある。技術は管理の媒体であると同時に、集団行動の対象や資源にもなりうる。これらの実践は、プラットフォーム労働者がアルゴリズムの受動的な対象ではないことを示す。労働者もまた、アルゴリズムのルールをめぐる知識、戦略、集団的な対抗策を形成する。]
推薦文献:
- Srnicek, Nick. Platform Capitalism. Polity, 2017. (Read: Chapter 2, pp. 35–63).
[注:ここではプラットフォームを、データ抽出と市場の仲介を担う経済組織の形態として理解する。この枠組みは、本回の「計算資源の集中」に、資本主義の規模の経済とネットワーク効果という前史を与える。ただし、プラットフォーム資本主義をそのままAI資本主義と同一視してはならない。理論上の類似は、制度上の同一性を自動的に意味しない。Srnicekは、データ抽出、ネットワーク効果、市場仲介によって形成される特定のビジネスモデルとしてプラットフォームを捉え、その集中化を理解する経済的基礎を提供する。プラットフォームは均質ではない。広告、クラウド、産業、リーン・プラットフォームは、それぞれ異なる収益構造と資産構成に依存する。これらの類型の区別は、プラットフォーム労働と計算資源の集中を分析する際に概念の一般化を避ける前提となる。この類型論は同時に、なぜ「プラットフォーム」を単一の階級関係として扱えないかも説明する。固定資産、労働契約、データへの依存度が異なり、集中化の経路も異なるためである。この枠組みのもう一つの強みは、データを抽象的な「情報社会」ではなく企業間競争に位置づける点にある。企業がデータを蓄積するのは、サービスの改善、ユーザーの囲い込み、モデルの学習、参入障壁の形成に役立つからである。計算資源とデータの集中は、技術的規模だけでなく、具体的な収益モデルと結びつけて考える必要がある。プラットフォームの集中は、ネットワーク効果に加えて、データ、資本、インフラ間の正のフィードバックから生じる。あらゆる代替政策は、どの層に作用するのかを明示しなければならない。この意味で類型論は政策分析の前提でもあり、集中の仕組みの違いには、異なる規制と公共的代案が必要となる。経済的な類型の区別は、一律の「プラットフォーム統治」構想に先行する。Srnicekの類型論は、拡張に必要な資本集約度の違いも示す。軽資産型は外注とリスク転嫁に依存し、クラウド型は巨額の固定資本を必要とする。どちらもプラットフォームと呼ばれるが、労働編成と競争障壁はまったく異なる。プラットフォーム資本主義とAIインフラをつなぐ際には、この資本構造の区別が特に重要である。そうしなければ「プラットフォーム」という語が、計算資源産業の重資産的な特質を覆い隠してしまう。]
- Huws, Ursula. Labor in the Global Digital Economy: The Cybertariat Comes of Age. Monthly Review Press, 2014. (Read: Chapter 1, pp. 11–38).
[注:Huwsは、世界的な分業と商品化からデジタル労働を理解する。デジタル労働はソフトウェア技術者の仕事だけではなく、サービス、外注、プラットフォーム、知識労働がともにつくる新たな世界的労働構造である。アルゴリズム管理は管理者の消滅を意味しない。Labor in the Global Digital Economy: The Cybertariat Comes of Ageが提供するのは、「不安定な生活への感情的な依存と労働制度との関係」への決定的な名称ではなく、競合する説明によって検証できる理論的な切り口である。その開放性自体が研究上の価値をなす。Huwsはデジタル労働を世界的な商品化と分業に位置づけ、「知識労働者」をデジタル経済の唯一の主体とする見方に反対する。顧客対応、外注、物流、製造、事務労働はすべて、情報化を通じて再標準化され、遠隔的に組織される。サイバータリアートという概念は、高技能と低技能、中心と周辺をつなぎ、デジタル化の階級構造を世界的な労働連鎖のなかで捉えることを求める。世界的な視点は、「デジタル労働=遠隔のホワイトカラー労働」という偏見も修正する。高度なソフトウェア産業と低賃金のバックオフィス業務は、同じ価値連鎖に属することが多く、前者の自動化された外観が後者の標準化された労働に支えられる場合もある。階級分析では、空間的に分離しながら機能的に依存し合うこれらの仕事を同時に扱わなければならない。そのため、ソフトウェアのインターフェースと供給網の双方を見て、可視的な知識労働と隠された支援労働を同じ構造へ戻す必要がある。Huwsは家庭と社会的再生産もデジタル労働の視野に含める。遠隔化と柔軟化はケアの負担を消さず、むしろ仕事の時間をさらに家庭へ浸透させることがある。したがって、プラットフォーム労働の自由な時間割は、ジェンダー化された再生産責任と併せて分析する必要がある。同じ「柔軟性」も、労働者によっては自律を意味し、別の労働者には仕事と生活の境界の持続的な崩壊を意味しうる。]
- Casilli, Antonio A. Waiting for Robots: The Hired Hands of Automation. Foreword by Sarah T. Roberts. Translated by Saskia Brown. University of Chicago Press, 2025. Originally published in French as En attendant les robots: Enquête sur le travail du clic in 2019. (Read: Introduction; selected chapters on micro-work, clickwork, data labor, and automation).
[注:焦点は「自動化の背後にも労働がある」ことにある。マイクロワーク、データのアノテーション、コンテンツ・モデレーションは、AIの学習と運用が、高度に分散し、見えにくい人間労働に依存することが多いと示す。ここから「プラットフォーム労働」と「AI労働」を結びつけられる。2019年のフランス語の研究を、生成AIの登場後の「自動化」言説と直接対話させるため、2025年の英訳を用いる。ただし本書の核心は隠された労働にあり、大規模モデルへの事後的な論評として誤って位置づけるべきではない。CasilliのWaiting for Robots: The Hired Hands of Automationは、「自動化の語りがマイクロタスク、アノテーション、保守労働をどう隠すか」を組織と制度の分析へ導く。最大の貢献は、いわゆる「自動化」を、多数の人間による微細な労働の組織過程へと分解する点にある。クリック、アノテーション、検証、審査は、機械知能の外側に偶然付け加わる補修ではなく、モデル学習とサービス運用の必要条件であることが多い。したがって、自動化は労働の単純な消滅ではなく、空間、契約、可視性に関する労働の再配置である。Casilliの分析は、人間労働が自動化システムのフィードバック回路に組み込まれることも示す。モデルには継続的なアノテーション、検証、修正、内容判断が必要であり、プラットフォームはクラウドソーシングを通じてこれらの工程を世界の労働市場へ分散させる。ヒューマン・イン・ザ・ループは一時的な移行段階ではなく、大規模なAIサービスの長期的な組織形態となりうる。こうして自動化と新たな労働市場はともに成長する。微細な労働の世界的分布は、従来の労働法における雇用主の特定も難しくする。労働者は多層的な請負業者、クラウドソーシングのプラットフォーム、最終的なAI企業と向き合い、責任の連鎖は分断される。そのため、自動化の倫理には、供給網の責任と労働基準をつなぎ直す必要がある。]
- Gray, Mary L., and Siddharth Suri. Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass. Houghton Mifflin Harcourt, 2019. (Read: Chapters 1–3).
[注:本書は、AIの背後にあるマイクロタスク、コンテンツ・モデレーション、プラットフォーム化された人間労働を示す。CasilliやCrawfordと併読すれば、「自動化」を労働の再配分と不可視化として捉え直せる。協同組合主義も技術的な万能薬ではない。GrayのGhost Work: How to Stop Silicon Valley from Building a New Global Underclassが照らす「協同組合、労働組合、労働法、コモンズの統治がそれぞれ権力のどの層を変えるか」という問いは、テクスト内部の概念論争にとどまらない。ゴースト・ワークが明らかにするのは、自動化の語りにおけるインフラの逆説である。システムが人手を必要としないと説明されるほど、その「知能」を維持する人間の判断は隠されやすくなる。世界的なマイクロタスク市場は、労働を断片的でオンデマンドの、組織化しにくい単位へ分割し、技術的効率を新たな労働の不安定性の上に築く。このことは、AI労働政策が技術職だけを対象にしてはならないことも意味する。マイクロタスク労働者の賃金、労働安全、異議申立ての経路、国境を越える労働権もAI統治の一部である。GrayとSuriは、ゴースト・ワークの組織上の逆説も示す。労働者は個人アカウントへ分散されながら、経験の交換、詐欺の見分け、相互扶助のために非公式の共同体を必要とする。プラットフォームが安定した労働組織を提供しないため、労働者自身が協働ネットワークをつくる。デジタル労働の集団性は消えず、正式な職場から、より脆弱で制度的な承認を得にくい空間へ移る。この組織形態は「オンデマンド」が需要の自然発生を意味しないことも示す。プラットフォームはタスクの分解、価格設定、品質管理によって労働市場を積極的に形成する。労働者の代替可能性は、技術設計と契約設計によって共同でつくられる。ここに、労働の不可視性と技術神話との直接的な制度上のつながりがある。]
- Rosenblat, Alex. Uberland: How Algorithms Are Rewriting the Rules of Work. University of California Press, 2018. (Read: Chapters 2–4).
[注:Rosenblatは運転手の経験から、アルゴリズムによる配車、価格、評価、情報の非対称性を分析する。アルゴリズム管理についての具体的な民族誌により、プラットフォーム労働は抽象的な資本と労働の関係だけでは捉えられなくなる。Uberland: How Algorithms Are Rewriting the Rules of Workの分析尺度では、「プラットフォーム労働のタスク化、アルゴリズム管理、リスク転嫁、集団行動」はまず具体的な関係として現れる。自動化は労働を隠すことで実現されることが多い。アノテーション、モデレーション、マイクロタスク、保守作業が世界的な労働市場へ分割され、「機械が作業を完了する」という語りが新たな労働編成とリスク転嫁を覆い隠す。Uberのアルゴリズム管理は管理をなくすのではなく、命令を価格表示、タスクの割当て、評価、情報の非対称性へ分散させる。運転手には形式上の選択の自由があるが、実際の労働では、プラットフォームが設計した誘因の環境に絶えず応答する。したがってRosenblatの民族誌は、「自律」と「統制」が同じ労働制度でどう共存するかを具体的に説明する。情報の非対称性が管理の道具になる仕組みも明らかにする。プラットフォームは需要、価格、運転手の実績を同時に把握する一方、運転手は部分的な情報しか見られず、それに基づいて選択する。アルゴリズム管理の力は、命令の自動化だけでなく、市場全体の情報の独占と、情報開示のタイミングの統制にも由来する。この点は従来の現場監督との重要な違いである。労働者が向き合うのは、自分を観察できるが自分からは観察しにくいプラットフォームである。この情報構造により、「自律的な労働」は明確な非対称性を帯びる。選択権は存在するものの、プラットフォームのより大きな情報優位を前提としている。]
- Aloisi, Antonio, and Valerio De Stefano. Your Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labour. Bloomsbury, 2022. (Read: chapters on algorithmic management and worker protection).
[注:本書は労働法と制度設計からアルゴリズム管理を扱う。「上司」は一人の人間とは限らず、配車・配分モデル、業績評価、自動停止、ルールのシステムとして現れうる。この視点は政治経済学の議論を労働制度へ接続する。AloisiのYour Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labour、RosenblatのUberland: How Algorithms Are Rewriting the Rules of Work、IraniのChasing Innovation: Making Entrepreneurial Citizens in Modern Indiaは、「不安定な生活への感情的な依存と労働制度との関係」を異なる知的伝統に置く。アルゴリズム管理は管理者の消滅を意味しない。AloisiのYour Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labourは、この関係を抽象的すぎる時代診断から解き放つ。自動化は労働を隠すことで実現されることが多い。Your Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labourのこの関係への貢献は、理論を反証に開かれたものに保つ点にある。AloisiとDe Stefanoは労働法からアルゴリズム管理に接近し、「上司はアルゴリズムだ」という通俗的な表現に制度的な内容を与え直す。仕事の割当て、業績管理、価格設定、停止は自動化できるが、労働関係における統制権は依然として企業が設計し、その法的帰結を負う。核心は、責任をソフトウェアへ渡すことではなく、従来の雇用区分、情報を得る権利、集団交渉が、この非人格的な管理にどう対応するかにある。]
- Irani, Lilly. Chasing Innovation: Making Entrepreneurial Citizens in Modern India. Princeton University Press, 2019. (Read: chapters on entrepreneurial citizenship, design and development).
[注:Iraniは、「イノベーション」が統治と経済政策の道徳的言語になる過程と、起業家的主体の形成を研究する。Berlantと併読すると、労働者に「革新者」「起業家」、自己管理者であることが求められるとき、権力は希望と自己実現の言葉を通じて労働関係に入るのかと問える。Chasing Innovation: Making Entrepreneurial Citizens in Modern Indiaを軸にすれば、「自動化の語りがマイクロタスク、アノテーション、保守労働をどう隠すか」を、相互に関連する過程へ分解できる。自動化は労働を隠すことで実現されることが多い。Chasing Innovation: Making Entrepreneurial Citizens in Modern Indiaから見ると、この不可視化は制度的な権限付与と資源配分から切り離せない。技術的形態は組織権力と結びついて初めて持続的な社会的帰結を生む。Iraniは、イノベーションが経済的標語から主体形成の仕組みへ転じる過程を研究する。デザイン思考、起業家教育、開発事業は、社会問題を起業によって解決できる機会として捉え直すよう個人を促し、構造的リスクを自己責任へ変える。プラットフォーム労働と並べると、起業家的市民性は「不安定性」が自律、柔軟性、革新として再包装される理由を説明する。「起業家的市民」は失敗の帰責方法も変える。革新が個人の責任になると、事業の失敗はインフラ、資本、政策条件の帰結よりも、能力不足として説明されやすくなる。プラットフォーム経済の自己ブランド化、評価、フリーランス文化はこの主体化を引き継ぎ、労働リスクを個人の選択として語り直す。]
- Schor, Juliet B. After the Gig: How the Sharing Economy Got Hijacked and How to Win It Back. University of California Press, 2020. (Read: Introduction; selected chapters on platform business models, inequality, and alternatives).
[注:Schorは共有経済のプラットフォームに関する実証研究を通じて、ビジネスモデル、労働への参加形態、分配の帰結を区別し、ギグ・エコノミーのあらゆる経験を一つの搾取モデルに押し込めることを避ける。ScholzとWoodcockの後に読むと特に有益である。協同組合主義や労働者の闘争は重要だが、労働者の依存度、所得構成、労働動機は同一ではなく、代案も単一の組織モデルには頼れない。SchorのAfter the Gig: How the Sharing Economy Got Hijacked and How to Win It Back、IraniのChasing Innovation: Making Entrepreneurial Citizens in Modern India、van Doornの「Platform Labor: On the Gendered and Racialized Exploitation of Low-Income Service Work in the 'On-Demand' Economy」は、「協同組合、労働組合、労働法、コモンズの統治がそれぞれ権力のどの層を変えるか」を、巨視的な判断から仕組みの分析へ移す。協同組合主義も技術的な万能薬ではない。After the Gig: How the Sharing Economy Got Hijacked and How to Win It Backの貢献は、まさにこの問いを識別可能な制度的仕組みへ戻す点にある。自動化は労働を隠すことで実現されることが多い。Schorは、共有経済の初期の社会的展望と、その後のプラットフォームの商業化との分岐を区別し、効率、共同体、副収入の語りが、ベンチャー資本と規模拡張によって再編される過程を示す。実証研究が示すのは、プラットフォーム経済には単一の道筋がないということである。労働条件、所有権、共同体のルールによって異なる結果が生まれる。問いは、共有を宣伝文句から実際の資源統治へ変えるには、どのような制度が必要かへと移る。]
- van Doorn, Niels. "Platform Labor: On the Gendered and Racialized Exploitation of Low-Income Service Work in the 'On-Demand' Economy." Information, Communication & Society 20, no. 6 (2017): 898–914. (Read: entire article).
[注:van Doornは、オンデマンドのサービス労働が抽象的な「柔軟労働」ではなく、高度にジェンダー化・人種化され、都市のサービス関係に依存する労働であると強調する。既存の低賃金サービス業、移民労働、社会的再生産とプラットフォーム資本主義を結びつけることで、インターフェースやアルゴリズムによる配分だけを出発点とする分析を修正する。「Platform Labor: On the Gendered and Racialized Exploitation of Low-Income Service Work in the 'On-Demand' Economy」から、タスク化、アルゴリズム管理、リスク転嫁、集団行動の因果関係をより明確に区別できる。van Doornは、ジェンダー化・人種化された低所得のサービス労働を中心に据え、運転手や高技能のフリーランサーでギグ・エコノミー全体を代表させる傾向を修正する。清掃、ケア、配送は、もともと社会的再生産と移民労働の不平等な構造に位置しており、プラットフォーム化は評価、マッチング、支払いをさらにデジタル化する。アルゴリズムはこれらの差異を創出したのではないが、再符号化し、拡大しうる。オンデマンド経済のジェンダーと人種の構造は、技術的な媒介が労働市場の歴史的階層を消さないことも示す。顧客評価、地理的マッチング、プラットフォームの分類は、既存の偏見を一見個別的な取引結果へ変換しうる。したがって労働の公正には、アルゴリズム設計だけでなく、移民資格、ケア責任、社会保障などプラットフォーム外の制度も関わる。プラットフォーム化は社会的偏見を、雇用主の直接的な判断から、分散した顧客評価とシステム順位へ移し、差別責任の特定を難しくしうる。アルゴリズム化は不平等からアイデンティティを取り去るのではなく、個別的に見えるデータのフィードバックを通じて不平等を作動させ続ける場合がある。]
- Graham, Mark, Isis Hjorth, and Vili Lehdonvirta. "Digital Labour and Development: Impacts of Global Digital Labour Platforms and the Gig Economy on Worker Livelihoods" Transfer 23, no. 2 (2017): 135–162. (Read: entire article).
[注:Graham、Hjorth、Lehdonvirtaは、デジタル労働プラットフォームを世界的な開発と越境的労働市場に位置づける。オンライン外注が仕事の機会を増やす一方、交渉力の非対称性、供給過剰、地域間の賃金差を生む過程を検討する。本回の視野をUber型の地域的なギグ労働から世界的な遠隔労働へ広げ、プラットフォーム労働の内部に異なる空間尺度があることを示す。アルゴリズム管理は管理者の消滅を意味しない。三者は世界的なデジタル労働市場から、オンライン外注が開発と地理的不平等に及ぼす影響を考察する。インターネットは越境取引の費用を下げたが、賃金格差、言語の序列、地域への依存をなくしてはいない。プラットフォームは労働需要を世界化する一方、労働者をそれぞれ異なる社会保障制度に残す。デジタル労働は国境を越えるとともに、国境を利用する。この世界市場は新たな代替可能性も生む。雇用主は複数の国の間でタスクを迅速に移せるが、労働者は地域の生活費、通貨、法規に制約される。遠隔プラットフォームは非対称な空間的自由を形成する。資本とタスクは動きやすいが、労働者の社会的再生産は依然として強く地域に根ざす。したがって越境的デジタル労働の空間政治は「地理の消滅」ではなく、地理的差異が費用上の優位と交渉力の格差へ変換されることである。その「開発効果」には二面性がある。周辺地域に新たな収入をもたらす一方、その低賃金と弱い社会保障を利用して世界的な費用差を維持することもある。開発上の価値は雇用数だけでなく、交渉力、技能の蓄積、所得の安定性、価値連鎖の上位へ移動できるかによって評価する必要がある。さらに、プラットフォームの評価と世界的競争は開発問題を個人化する。労働者には技能と評判による境遇改善が求められ、為替、地域賃金、プラットフォームの交渉力といった構造条件は背景に退く。デジタル開発政策は、これらを再び視野に入れなければならない。]
- Karatani, Kojin. The Structure of World History: From Modes of Production to Modes of Exchange. Translated by Michael K. Bourdaghs. Duke University Press, 2014. Originally published 2010. (Read: Chapter 12, "Toward a World Republic," especially the discussion of Mode D, pp. 285–308).
[注:Karataniの交換様式Dはデジタル通貨の提案ではなく、歴史の動態における交換の再考である。資本主義を「生産様式」だけで理解する方法を超え、資本、国家、互酬性を交換形態のより長い歴史へ位置づける。本回に置く意義は、Web3、DAO、プラットフォーム協同組合主義に、より高次の制度的な問いを突きつける点にある。交換様式そのものが変わらないなら、技術的な接点だけを変えて資本と労働の関係を変えられるのか。The Structure of World History: From Modes of Production to Modes of Exchangeを通じて、「自動化の語りがマイクロタスク、アノテーション、保守労働をどう隠すか」を、修辞的な命題から制度分析へ移せる。様式Dの規範的な力は、資本主義の代案を所有形態の単独の変更へ縮減することを拒む点にある。交換、国家、互酬関係は入れ子状に結びつき、デジタル協同のいかなる構想も、公共財政、社会的再生産、地域を越えた調整を扱わなければならない。Karataniは「分散型組織」に、技術的統治を超える世界史的な尺度を突きつける。この交換様式の枠組みは、デジタル左派に国家の問題の再考も迫る。資本、国家、ネーションは別々に取り外せる環ではなく、市場の拡大は国家の再分配や保護とともに働くことが多い。DAOや協同プラットフォームによって既存組織を置き換える構想も、税、公共サービス、共同体間の調整を扱わなければ、社会的再生産の全体的な機能を担いにくい。この点からも、様式Dを技術的ユートピアに還元できないことがわかる。真に難しいのは、資本と国家の相補的構造を超えつつ、公共財、再分配、地域間調整を担う制度をつくることである。技術はその一要素にすぎない。]
- Ostrom, Elinor. Governing the Commons. Cambridge University Press, 1990. (Read: Chapter 1, pp. 1–28).
[注:Ostromは実証研究によって、共有資源の統治を単に「コモンズの悲劇」と捉える見方に挑む。DAOと併読する際の焦点は、ルールがどう生まれ、実施され、修正されるか、またどのような参加者の構成が共有資源を維持できるかにある。コードはルールを実装できるが、信頼、参加、責任を自動的には生み出さない。協同組合主義も技術的な万能薬ではない。概念の適用範囲を最大化するより、理論を反証可能に保つべきである。実証資料が核心の仕組みを支持する場合にのみ、この枠組みを拡張すべきである。Governing the Commonsは最終的に、「協同組合、労働組合、労働法、コモンズの統治がそれぞれ権力のどの層を変えるか」を、完成済みの時代診断ではなく、実証研究によってさらに分解できる問いにする。Ostromの制度分析は、「コモンズ=無規則な共有」という想像を修正するのに特に有効である。安定した共有資源の統治は、境界、参加型のルール、監視、段階的制裁、紛争解決に依存し、それらを具体的な共同体へ適応させる必要がある。デジタル・コモンズは自然資源と複製費用が異なるが、その統治問題も開放的なアクセスだけでは解決しない。Ostromの設計原則をデジタル資源へ機械的に複製してはならない。ソフトウェアとデータの複製は非競合的であり、構成員の境界も世界に広がりうるが、サーバー、保守、統治に向ける注意力は依然として希少である。移しうるのは制度の多様性と自己組織化の能力に関する実証主義であり、すべてのオープンな事業をそのままコモンズと呼ぶことではない。ここから問いは制度設計へ戻る。資源が複製可能でも、構成員による統治、保守義務、紛争解決の必要は消えない。デジタル資源の複製可能性は枯渇の問題を弱める一方、保守と統治の注意力の希少性を強める。コード審査、サーバー費用、共同体の管理、紛争処理には継続的な投入が必要である。Ostromの実証主義が教えるのは、共有資源の要点は「開放」だけでなく、資源の特性に合う長期的な統治を築けるかにあるということだ。]
第4部:一般的知性と技術の未来(第7–8週)
第7回:一般的知性と計算資源の集中への抵抗
本回は、Marxの一般的知性の問題を現代のAIインフラへつなぎ直す。ただし、それは「マルクスは大規模モデルを予言した」という主張には基づかない。Grundrisseの「機械論断片」が扱うのは、社会的知識、固定資本、労働時間、価値形態の矛盾である。科学的知識と社会的協働がますます機械体系へ入り込むと、直接労働と生産力の関係は変化する。これに対してVirnoは、一般的知性を生きた公共の言語的・認知的能力として強調し直し、機械に凝固した知識の蓄積とは同じでないと指摘する。両者の緊張関係は、現代の計算資源の集中を理解するために、「AIは人間に似ているか」よりも生産的な問いをもたらす。
生成AIは、社会が共同で生産する文章、コード、画像、フィードバック、判断を、大規模に学習資源へ変換する。同時に、半導体、データセンター、エネルギー、資本投下、専門労働に依存し、参入障壁の高い固定資本を形成する。PasquinelliによるAIの社会史、Crawfordによる物質的・労働的条件の追跡、知識資本主義、認知資本主義、デジタル労働の研究は、いわゆる「知能」が社会関係から独立した純粋な能力ではなく、知識、労働、インフラ、所有権の配置が組み合わさった結果であると示す。したがって計算資源の集中は、まず経験的に分析できる産業と制度の問題であり、先験的な政治的結論ではない。
抵抗の尺度も「オープンソース」や「誰もがGPUを持つこと」へ縮減できない。重みの公開は利用可能性を改善しうるが、学習資源、クラウド基盤、エネルギー費用、統治権を変えるとは限らない。公共の計算資源は参入障壁を下げられるが、データ、保守、責任、配分の問いは残る。Virnoの退出、コモンズ、非国家的公共性が示すのは既成の解答ではなく、一般的知性を再編する政治的な手がかりである。社会に共通する認知能力は、それに見合う制度とインフラを持ち、モデル、データ、計算資源を私的固定資本としてのみ存在する状態から変えられるだろうか。
必読文献:
- Marx, Karl. Grundrisse: Foundations of the Critique of Political Economy. Translated by Martin Nicolaus. Penguin Books, 1973. Originally written 1857–58; first published 1939–1941. (Read: "Fragment on Machines," pp. 690–712).
[注:「マルクスはAIを予言した」という遡及的な読みを退けなければならない。Marxが扱うのは、社会的知識、一般的知性、固定資本、労働時間、価値の歴史的矛盾である。科学的知識と社会的協働が機械体系にますます入り込むと、生産における労働時間の位置は変わる。今日のAIは、この問題の新たな歴史的場面と捉えられるが、大規模言語モデルをMarxの「機械体系」と直接同一視することはできない。要点は、社会的知性が私有の計算インフラへ凝固していくとき、価値、労働、所有の矛盾が新たな形で現れるのかを問うことにある。「機械論断片」は、固定資本と生産力についてのMarxの分析へ戻して読む必要がある。機械体系が科学的知識と社会的協働を吸収することで、労働者はますます生産過程の監督者・調整者となり、富の尺度としての労働時間と生産力の社会化との緊張が生じる。テクストにおける一般的知性の位置は、後のアウトノミア的解釈とは同一でなく、この差を保つ必要がある。AIへ移す際に最も有益なのは、知識が固定資本へ入る形態の比較である。モデル、ソフトウェア、半導体、データセンターは新たな固定資本の組合せをなすのか。学習データと生きた協働能力のうち、固定化できない部分は何か。これはまず、社会的知識、共通の能力、生産関係の理論的問題である。現代への適用には、これらの能力がどう固定資本化されるかの説明が必要であり、モデルのパラメーターを人類の共通精神として擬人化してはならない。Marxの「機械論断片」がまず扱うのは、固定資本、社会的知識、労働時間の矛盾であって、AIの予言ではない。一般的知性の機械体系への編入は、生産が社会的に蓄積された科学と協働能力にますます依存する一方、価値形態はなお労働時間を尺度にすることを意味する。今日の基盤モデルはこの矛盾の新たな場面になりうるが、モデル、データセンター、GPUは特定の歴史的形態であり、逆にMarxの概念を定義するものではない。]
- Virno, Paolo. A Grammar of the Multitude: For an Analysis of Contemporary Forms of Life. Translated by Isabella Bertoletti, James Cascaito, and Andrea Casson. Semiotext(e), 2004. Originally published 2001. (Read: Day One, "Common Places and 'General Intellect'"; Day Two, "Intellect as Score" and "Reason of State and Exit"; Day Three, "The Emotional Tonalities of the Multitude").
[注:VirnoによるMarxの一般的知性の重要な展開は、それを機械、固定資本、技術体系に凝固した「知識の総量」とだけ捉えることを拒む点にある。「Common Places and 'General Intellect'」と「Intellect as Score」では、一般的知性はまず、社会が共有する言語、知識、判断、協働の能力として、生きた労働の一般的条件として現れる。すると現代AIの核心は、機械が人間の知能を得たかではなく、すでに社会に存在する認知能力が、プラットフォーム、モデル、インフラによってどう再編されるかとなる。「Reason of State and Exit」はこれを政治の次元へ進める。一般的知性が資本に取り込まれても、なぜ解放は自動的に生じないのか。Virnoの「退出」は、既存の国家・企業組織の掌握だけを目標とせず、公共の協働条件を再編する政治を求める。今日、計算資源の集中への抵抗を、より民主的な巨大技術企業の単なる複製と理解してはならない。真の問いは、社会がデータ、モデル、計算、知識のインフラを再び自らのものにし、一般的知性を計算資本の原料にするのではなく、計算資源を一般的知性に役立てられるかにある。ここで「計算資源のマルチチュード」は、問題を提起する概念として保ち、Virno本人の用語としてはならない。一般的知性、インフラの所有、自律的組織の間に、新たな政治的接続が成立するかを検討するための概念である。とりわけ「Intellect as Score」の比喩を深く読む必要がある。楽譜は完成済みの演奏ではなく、具体的行為のなかで初めて実現する公共の能力である。こうして一般的知性は固定資本に保存された対象だけでなく、発話、協働、学習、判断の共通条件となる。「Reason of State and Exit」は、国家と非国家的公共性との関係へ政治的問題を進める。退出は私的な逃走でも、技術共同体による規制からの逃避でもなく、協働を再編して古い関係を支えなくすることである。これをオープンなモデル、公共の計算資源、連合型インフラへ適用する際には、各制度が本当に共通の能力の組織権を変えるのか、それとも道具の価格を下げるだけなのかを一つずつ問う必要がある。Virnoは一般的知性を固定資本から、生きた言語、判断、習慣、協働能力へ引き戻す。それは資本によって生産力として組織されうると同時に、非国家的公共性を形成する可能性を含む。「退出」は制度から単に逃れることではなく、協働条件の再編である。現代の計算資源の問題で真に政治的な焦点となるのは、機械が知能を持つかではなく、共通の認知能力が自らのデータ、モデル、計算インフラを築けるかである。]
- Pasquinelli, Matteo. The Eye of the Master: A Social History of Artificial Intelligence. Verso, 2023. (Read: Chapter 4, "The Origins of Marx's General Intellect"; Chapter 7, "The Automation of Pattern Recognition"; Conclusion, "The Automation of General Intelligence").
[注:Pasquinelliの貢献は、AIを社会から独立した計算上の奇跡とみなさず、分業とパターン認識の歴史へ戻す点にある。第4章はMarxの一般的知性をたどり直し、第7章はパターン認識と自動化を論じ、結論はそれらの歴史を一般的知能の自動化へと接続する。本回では、「モデルの学習」を純技術的操作から労働と知識の組織化の問題へ移せる。AIが自動化するのは、一人の技能、一つの職業、それとも社会がすでに形成した集団的認知能力なのか。本書の社会史は、知能を分業とともに理解することを求める。機械知能は、人間の労働から規則、パターン、統計的秩序を抽出し、自動化の形で労働を再編することが多い。第4章が一般的知性へ戻るのは、MarxがすでにAIを説明したと宣言するためでなく、知識がいかに生産装置になるかを問うためである。第7章のパターン認識の歴史化は、深層学習が「データから自ら学ぶ」という神話を解体する。結論の一般的知能は、まず社会的・制度的な問題となる。どの一般的能力が形式化され、誰がパターンを決め、誰が機械の認識するパターンへの適応を強いられるのか。分業とパターン認識の歴史からAIを再構成することで、機械学習は純粋数学から突然出現した知能とは理解されなくなる。自動化は社会的実践の判断規則をまず抽出・形式化し、その実行を機械へ委ねることが多い。こうして一般的知能の歴史は、組織、測定、管理の歴史と結びつく。AIが「自動化」するのは抽象的な精神ではなく、すでに社会化された認知操作である。]
- Hardt, Michael, and Antonio Negri. Multitude: War and Democracy in the Age of Empire. Penguin Press, 2004. (Read: Part 2, selected sections on immaterial labor, biopolitical production, and the common).
[注:HardtとNegriは、マルチチュード、非物質的労働、生政治的生産、共を結びつける。生産が言語、知識、感情、協働にますます依存するとき、政治的主体は均質な産業労働者の像だけを前提としなくてよい、というのである。MarxとVirnoの後に読むと、より明確な政治組織の構想が得られるが、プラットフォーム労働研究による実証的な修正も必要である。社会的協働が広く動員されても自律的主体が自動的に成立するわけではなく、資本は所有権、プラットフォーム、インフラを通じてその協働を捕捉できる。本回の文脈でMultitude: War and Democracy in the Age of Empireを読むと、理論上のマルチチュードや共から持続的な制度を築く難しさが具体化する。一般的知性をモデルのパラメーター、データセット、インターネットと直接同一視してはならない。それはまず、社会化された言語、知識、判断、協働の能力を指す。これらは固定資本に取り込まれても、機械そのものと完全には重ならない。HardtとNegriのマルチチュードは、共通の生産能力と差異をもつ主体の結合を強調し、均質な人民や従来の労働者階級とは異なる政治形態を探る。この開放性はデジタルな協働に接続しやすいが、ネットワークの比喩によって過度に美化されやすくもある。真の問いは、分散した協働が制度的な持続性、資源、意思決定の構造をどう得るかである。そうでなければ「マルチチュード」は接続を描く政治的修辞にとどまりうる。デジタル・ネットワークとの親和性は技術楽観主義を誘いやすい。共同生産はネットワーク接続だけで共通の政治を生まず、異なる主体も自然に安定した意志を形成するわけではない。難題は、差異を消さずに意思決定と資源配分の制度を築くことにある。この点がマルチチュード論を統治の問題へ導く。]
推薦文献:
- Vercellone, Carlo. "From Formal Subsumption to General Intellect" Historical Materialism 15, no. 1 (2007): 13–36.
[注:Vercelloneは、知識、協働、社会的知性が資本蓄積に入る過程を論じる。第6回と第7回をつなぐ議論であり、労働編成の問題は最終的に、一般的知性と知識コモンズを誰が所有するかへ向かう。理論上の類似は制度上の同一性を自動的には意味しない。計算資源の集中は産業構造とインフラの問題である。半導体、エネルギー、資本、クラウドの規模要件は、誰がモデルを学習・展開・維持できるかを決めるため、重みの公開だけでは私的インフラへの依存はなくならない。したがって「From Formal Subsumption to General Intellect」による、一般的知性、固定資本、認知労働、計算資源の集中の関係の説明は、対応する制度的仕組みが実証資料によって識別できる場合に成立する。Vercelloneは一般的知性を、資本による社会的知識の包摂と結びつけ、価値生産が工場の境界を超えた協働と知識にますます依存すると強調する。認知資本主義論は、レント、知的財産権、社会化された生産を同じ矛盾に置く。AI時代に提供するのは「知識が労働に取って代わった」という結論ではなく、完全に囲い込みにくい共通の知識に、資本が希少性と統制を再構築する仕組みの分析である。この議論は知識経済の分析にレントを戻す。共通の知識を完全に私有化しにくい場合、資本は特許、プラットフォームへの入口、インフラの所有を通じて収益を得る傾向を強めうる。AIモデルによる知識の封入とAPIアクセスもこの観点で研究できるが、それがレントにあたるかは、競争構造、所有権、価格の仕組みに即して判断する必要がある。形式的・実質的包摂と一般的知性との接続は、資本が労働を直接組織することから、社会化された知識の条件を統御することへ移る過程の説明でもある。この時代区分に議論の余地があっても、重要な問いは残る。生産が広範な社会的協働に依存するとき、どの制度が分散した協働を私的利益へ戻すのか。プラットフォームとAIインフラは、この問いに新たな実証対象を与える。]
- Moulier Boutang, Yann. Cognitive Capitalism. Translated by Ed Emery. Polity, 2011. Originally published in French as Le capitalisme cognitif in 2007. (Read: Introduction; selected chapters on knowledge, externalities, and cognitive labor).
[注:Moulier Boutangの認知資本主義という命題は、知識、革新、協働ネットワーク、正の外部性を価値生産の中心に据え、従来の工場の境界に生産過程が収まりにくくなる理由を説明する。一般的知性と密接に関わるが、同義語ではない。MarxとVirnoの問いは社会的知性と資本の関係にあり、認知資本主義は資本主義の特定段階を指す実証的・理論的な名称である。併読により、「知識が主要な生産力になる」とは何を意味するのかを検証できる。計算資源の集中は産業構造とインフラの問題である。Cognitive Capitalismが示すのは、「社会に共通する能力が所有権、プラットフォーム、インフラによってどう再編されるか」への最終的な命名ではなく、競合する説明によって検証できる理論的切り口であり、その開放性自体に研究上の価値がある。Moulier Boutangは、蓄積における知識、革新、正の外部性の突出した位置を認知資本主義として描き、工場外の協働を価値分析へ含める。「認知」という概念で物質的生産を覆い隠してはならないが、ネットワーク、特許、教育、知識の流通が資本間競争の重要条件になる理由を説明できる。AIと結びつける際には、モデルの認知資源と、それを支えるエネルギー、ハードウェア、労働の双方を見る必要がある。認知資本主義は労働の測定問題にも直面する。知識生産は協働、教育、社会的交流に大きく依存し、個々の労働者の寄与は分離しにくくなるが、資本はなお賃金、知的財産権、企業の境界を通じて所得を配分する。モデル学習はこの矛盾を拡大する。学習材料は広範な社会的生産に由来する一方、利益は少数のインフラ所有者へ集中するからである。]
- Gorz, André. The Immaterial: Knowledge, Value and Capital. Translated by Chris Turner. Seagull Books, 2010. Originally published in French as L'Immatériel in 2003. (Read: selected chapters on knowledge, value, and the crisis of measurability).
[注:Gorzは、知識、能力、生きた社会的経験を従来の労働時間だけでは十分に測れない一方、資本がそれらを占有可能な資源へ変え続ける矛盾に注目する。「知識」を抽象的情報から主体の能力と社会関係へ戻し、価値尺度の危機を問うため、一般的知性の系譜に置くのが有効である。AIに適用する際には、モデル学習が知識そのものを商品化するのか、それとも知識の表現、データ化された痕跡、アクセス権を資産化するのかを問うべきである。The Immaterial: Knowledge, Value and Capitalは、「開放、相互運用性、公共所有、自律的統治は同じ解決策ではない」という区別を組織と制度の次元へ導く。計算資源の集中は産業構造とインフラの問題である。Gorzの知識価値論は、認知資本主義の核心的な矛盾を、複製可能な知識と希少性の製造の間に置く。デジタル知識を複製しやすくなるほど、資本は知的財産権、プラットフォームの閉鎖、サービス化によって排他性を再構築する必要がある。基盤モデルとAPI経済では、吸収された公共の知識が管理されたアクセスとして市場へ戻るという、新たな形が現れる。知識の測定不可能性の強調によって、価値の問題を個人の労働時間だけで説明することはできなくなる。知識は長期の教育、公共研究、社会的交流に由来することが多く、複製費用はゼロに近づく。資本は特許、閉鎖的プラットフォーム、ブランドによって希少性をつくり直さなければならない。生成AIによる公共の文章の大量吸収は、共通の知識と私有のアクセス権の矛盾をいっそう可視化する。この議論はモデルのアクセス権にも使える。資本は知識内容自体を所有せずとも、計算資源、インターフェース、サービス条件を管理することで、共通の知識への接近を再び希少な商品にできる。占有は内容の所有権からインフラの所有権へ移る。]
- Roggero, Gigi. The Production of Living Knowledge: The Crisis of the University and the Transformation of Labor in Europe and North America. Translated by Enda Brophy. Temple University Press, 2011. (Read: Introduction; selected chapters on living knowledge, the university, and cognitive labor).
[注:Roggeroは大学と知識労働を中心に、生きた知識の生産を論じる。認知労働は抽象的な「創造階級」ではなく、教育制度、労働市場、協働ネットワーク、対立のなかで形成される。一般的知性を大きな哲学的命題から、大学、研究、訓練、知的財産制度へ戻す視点であり、生成AIが教育や研究へ入る際の、知識の共同生産と制度的占有の矛盾を問うのに適している。The Production of Living Knowledge: The Crisis of the University and the Transformation of Labor in Europe and North Americaが示す、理論上のマルチチュードや共から持続的な制度を築く困難は、テクスト内部の概念論争を超える。計算資源の集中は産業構造とインフラの問題である。Roggeroは「生きた知識」を大学の危機と労働の変容に位置づけ、知識は正式な研究機関だけでなく、社会的協働、学習、ネットワークのなかで絶えず生まれると強調する。大学は公共の知識を生産すると同時に、その商品化と数量化にも参加する。基盤モデルが学術的・公共的文章を大量に取り込むと、この矛盾はさらに強まる。社会的知識の開放性と私的な計算インフラの間に、新たな占有関係が生まれるからである。Roggeroにとって大学は、知識資本主義の外にある純粋な公共空間ではなく、労働変容の実験場である。研究、教育、学習は、業績測定、プロジェクト制、競争的資金調達にますます従う。AIとつなげると、学術知識が一方ではオープンな文章としてモデルに吸収され、他方では出版、プラットフォーム、計算資源によって再び階層化される過程を観察できる。]
- Beller, Jonathan. The Message Is Murder: Substrates of Computational Capital. Pluto Press, 2018. (Read: Introduction; selected chapters on computational capital, attention, and racialized extraction).
[注:Bellerは計算を、資本、視覚文化、金融化、人種化された暴力の歴史に位置づける。計算は機械内部の演算だけでなく、より広い社会的な価値収奪に埋め込まれている。この視点は一般的知性と必要な緊張関係をつくる。社会的知識を解放可能な共通能力としてのみ捉えると、その生産自体が差異化された暴力、注意の捕捉、世界的分業のなかにあることを見落としかねない。The Message Is Murder: Substrates of Computational Capitalの尺度では、一般的知性、固定資本、認知労働、計算資源の集中の関係は、まず具体的なつながりとして現れる。Bellerは計算資本主義を、知覚、注意、インフラがともに価値増殖に参加する体系と理解し、デジタル労働の境界を広げる。強い主張には実証資料による制約が必要だが、重要な問題を捉えている。計算は中立な裏方の道具ではなく、資本が視覚、時間、社会関係を組織する一般的な基盤になりうる。基層という概念は、視覚文化、計算、物質的基礎を接続する。画面上の注意活動は、半導体、ネットワーク、エネルギー、労働の体系のなかで初めて資本化できる。したがって計算資本は、純粋な情報経済でも従来の産業資本の単なる延長でもなく、複数層のインフラが価値捕捉を共同で組織する形態である。計算の「無形性」は、ここで再び物質化される。注意、画像、データの価値増殖は、常に基底の設備、ネットワーク、労働に依存する。計算資本の分析では、知覚の政治と物質的インフラの二重の視野が必要である。どちらかを欠けば、計算資本は再び抽象化される。]
- Tronti, Mario. Workers and Capital. Translated by David Broder. Verso, 2019. Originally published 1966. (Read: "Lenin in England," pp. 78–88).
[注:Trontiの意義は方法論にある。資本が何を望むかより先に、労働者の闘争が資本をどう変えさせるかを分析する。今日のプラットフォーム経済には、これを研究方法として用いるべきであり、「AIは労働者の闘争によって生まれた」という単純な証明にしてはならない。Workers and Capital、BellerのThe Message Is Murder: Substrates of Computational Capital、DoctorowのThe Internet Con: How to Seize the Means of Computationは、「社会に共通する能力が所有権、プラットフォーム、インフラによってどう再編されるか」を異なる知的伝統に置く。Workers and Capitalはこの問いを過度に抽象的な時代診断から解き放つ。Workers and Capitalの貢献は、まさに理論を反証可能に保つことにある。Trontiの方法は、技術変化を資本の革新戦略だけから説明できないと教える。労働者の拒否、移動、組織化、生産上の対立も、新たな管理技術と労働編成を資本に探させうる。自動化に適用するなら、具体的な闘争と投資決定の歴史的な関係を追うべきで、自動化そのものを労働者闘争の直接の成果として美化すべきではない。Trontiの労働者主義は、政治的知識が闘争のなかの調査から生まれることも強調する。労働者の利益を先験的に決めるのではない。プラットフォームとAIの労働では、運転手、アノテーター、モデレーター、技術者の具体的な対立から出発し、職場で技術がどう読み替えられるかを理解する必要がある。理論は労働者を技術変化の受動的な被影響者としてのみ位置づけてはならない。]
- Doctorow, Cory. The Internet Con: How to Seize the Means of Computation. Verso, 2023. (Read: Introduction; chapters on interoperability, switching costs, and antitrust).
[注:Doctorowは「計算の手段を奪取する」という主張を所有権の標語にとどめず、相互運用性、対抗的互換性、反独占、退出権という具体的な制度手段へ落とし込む。本回にとって貴重な中間的政策の層を提供する。一般的知性がプラットフォームのインフラに囲い込まれるなら、抵抗はまずロックインを弱め、相互運用と移行の能力を回復する必要があるのか。同時に、これらの改革がデータセンター、計算資源、資本の所有に届くのか、それとも主に市場競争を改善するのかも問うべきである。The Internet Con: How to Seize the Means of Computationにより、開放、相互運用性、公共所有、自律的統治が同じ解決策ではないことを、相互に関連する要素へ分解できる。「計算の手段を奪取する」とは、競争政策、相互運用性、ユーザーの権利をインフラ政治の中心へ戻すことである。技術的な開放が退出費用、データ移行、市場支配を変えないなら、選択肢は表面的になりがちだ。この改革路線は、より根本的な公共所有や協同統治と生産的に比較できる。相互運用性には明確な権力上の意味がある。ユーザーがデータ、社会的つながり、コンテンツを持ち出し、サービス間を接続できれば、囲い込みは弱まる。ただし、相互運用には安全、プライバシー、標準の統治に関する複雑な費用もある。「接続口を開くこと」は無条件の答えではなく、公共的ルールに支えられるべき競争と自治の仕組みである。その政治的意義は、競争を「プラットフォームの数を増やすこと」から、ユーザーが退出と移行の制度的能力を実際に持つかへ進める点にある。]
- Tarnoff, Ben. Internet for the People: The Fight for Our Digital Future. Verso, 2022. (Read: Introduction; chapters on privatization, public ownership, and democratic control).
[注:Tarnoffはインターネット私有化の歴史を公共的統御の問題へ接続し、今日のプラットフォーム権力を一部の悪い企業やアルゴリズムだけに帰す説明に反対する。その構想はインフラ政治に近く、所有権、公共投資、民主的統治、脱商品化を論じる。Doctorowと併読すると、改革の尺度をめぐる争点が現れる。相互運用性と競争政策で十分なのか、それともインターネットの一部の層を公共インフラとして捉え直す必要があるのか。Internet for the People: The Fight for Our Digital Future、DoctorowのThe Internet Con: How to Seize the Means of Computation、SchneiderのGovernable Spaces: Democratic Design for Online Lifeは、理論上のマルチチュードや共から持続的な制度を築く難しさを、巨視的判断から具体的な仕組みへ移す。Internet for the People: The Fight for Our Digital Futureの貢献は、その大きな判断を識別できる制度的仕組みへ戻す点にある。Tarnoffは私有化を歴史的過程とみなし、「ネットワークは当然に市場に属する」という常識を崩す。公共資金、商業化、広告モデルがともに今日の構造を形づくった以上、脱商業化にも資金調達と公共制度が必要であり、企業倫理の改善を訴えるだけでは足りない。こうしてインターネットの公共性は、所有と制度設計の問題になる。]
- Schneider, Nathan. Governable Spaces: Democratic Design for Online Life. University of California Press, 2024. (Read: Introduction; selected chapters on governance, ownership, and democratic design).
[注:Schneiderはオンライン共同体の統治実践から出発する。デジタル空間は本来無政府的なものでも、企業だけが管理できるものでもなく、ルール、構成員資格、所有権、意思決定手続は再設計できる。Virnoの退出やKeltyの再帰的公共との遠い共鳴もあるが、ここでの重点は制度構築にある。持続的な統治構造を欠くと、プラットフォーム権力からの「退出」は、別の私的プラットフォームへの移動にすぎなくなりやすい。Governable Spaces: Democratic Design for Online Lifeは、一般的知性、固定資本、認知労働、計算資源の集中を結ぶ因果関係を明確にする。Schneiderはオンライン空間を、憲制的に組織できる統治空間と捉える。抽象的な民主主義の称賛ではなく、ルールの形成、構成員の代表、権力の制約が焦点となる。民主化には実施可能な制度手続が必要であり、投票機能を画面へ組み込むだけでは足りない。統治可能な空間の核心は共同体の規則をつくるだけでなく、構成員が規則を変更できる制度的地位を得ることにある。プラットフォーム民主主義を世論調査や投票プラグインと同一視してはならない。サーバー、コード、最終的な執行権が外部の所有者に握られたままなら、参加手続に実質的拘束力がない場合もある。基準は参加ボタンの有無ではなく、構成員がルール、インフラ、執行権を持続的に制約できるかである。統治権がインフラ所有者を拘束して初めて、デジタル参加は諮問から制度的権力へ変わる。民主的設計は最終的に執行権へ届かなければならない。]
- Söderberg, Johan. Hacking Capitalism: The Free and Open Source Software Movement. Routledge, 2008. (Read: Introduction; selected chapters on free software, labor, and political economy).
[注:Söderbergは労働と政治経済学から自由ソフトウェア・オープンソースを説明し、従来の企業境界を超える協働生産の可能性と、資本が無償の貢献を再吸収する過程の双方を見る。本回を貫く問い、すなわちコードの公開は共同所有と同じなのかを検証できる。クラウドの計算資源、展開経路、データが集中管理されたままなら、ソフトウェア・コモンズの政治的効力はどこで弱まるのか。自由ソフトウェア運動は、技術的生産手段の統御をライセンス、協働規範、共有コードによって制度化できると示す。同時に、それは所有権に挑戦しうる一方、低費用の革新モデルとして資本に取り込まれもする。開放性は資本の外の純粋な空間ではなく、争奪が続く生産関係である。この運動は非国家的な制度革新も示す。コピーレフトは法的手段を逆用して共有を守り、公共資源の再私有化・再閉鎖を防ぐ。「資本からの退出」は法からの退出を意味するとは限らず、自律的実践はしばしば既存の法で持続可能な境界を築く。この点はデータとモデルのコモンズに直接示唆を与える。コモンズは制度的境界の廃止を意味しない。適切な法的設計が公共の生産を守る条件になりうる。公共モデルとオープンデータでも、ライセンスと統治の仕組みが再閉鎖を防ぎうる。開放性の持続自体が制度の成果なのである。また、開かれた生産も固有の物質的・制度的依存を形成する。事業にはホスティング、保守、資金、貢献者の統治が必要であり、それらが少数企業へ再集中すれば、コードの開放とインフラへの依存は共存しうる。公開モデルでも、重み、学習データ、計算資源、展開プラットフォームの所有関係をそれぞれ区別する必要がある。]
- Lazzarato, Maurizio. "Immaterial Labor" In Radical Thought in Italy: A Potential Politics, edited by Paolo Virno and Michael Hardt, 133–147. University of Minnesota Press, 1996. Originally published in Italian in 1994. (Read: entire chapter).
[注:Lazzaratoの非物質的労働という概念は、ポストフォーディズムにおける情報、文化的内容、コミュニケーション、主体性の生産の重要性を描く。一般的知性の系譜に大きな影響を与えたが、今日では二重に読む必要がある。一方では生産が社会的コミュニケーションへ広がる傾向を捉え、他方では「非物質的」という語が半導体、鉱物、物流、アノテーション、身体労働を隠しやすい。第6回の労働研究と併読することで、この限界が明らかになる。「Immaterial Labor」を通じて、開放、相互運用性、公共所有、自律的統治の区別を、修辞から制度分析へ移せる。この概念は言語、感情、知識、協働が価値生産に入る傾向を捉えるが、境界には慎重さが必要である。認知や伝達の活動が物質的労働から離れるわけではなく、機器製造、物流、エネルギー、再生産労働に依存する。両者を結び直してこそ、「非物質的」という語でデジタル経済の物質的基礎を隠すことを避けられる。歴史的にこの概念は、ポストフォーディズム、伝達、主体生産をつなぎ、労働が文化的内容と社会関係をますます直接生産することを強調した。問題は、設備、物流、身体労働を覆い隠しやすい点にある。物質的生産の退場宣言ではなく、労働内容と統制方法の変化として読む方が適切である。そうして初めて、現代のデジタル労働の多層的構造に接続できる。最も有益なのは名称自体でなく、価値生産が伝達、記号、主体の関係にますます依存すると示した点である。デジタル資本主義の問題は、これらの活動と物質的基礎がどう再結合されるかにあり、一方が他方に取って代わるかではない。]
- Dyer-Witheford, Nick, Atle Mikkola Kjøsen, and James Steinhoff. Inhuman Power: Artificial Intelligence and the Future of Capitalism. Pluto Press, 2019. (Read: Introduction; selected chapters on AI, labor, and capital).
[注:本書はマルクス主義政治経済学からAIの自動化と資本の関係を論じ、「将来の機械は意識を持つか」だけでAI問題を定義することを拒む。機械学習が生産、管理、労働市場へ入ると、資本の統制能力がどう再編されるかが焦点となる。一般的知性を現実のAI政治経済へ接続し、次回のAI哲学と未来の想像力に実証的な境界を設ける。Inhuman PowerはAIを独立した技術的主体ではなく、資本主義的自動化の歴史的論理に置く。機械学習は資本の予測、調整、監督の能力を広げるとともに、既存の分業と所有構造に依存する。「非人間的な力」は、まず社会関係によって組織された技術的能力である。Inhuman PowerはAIの自律性と資本の自律性も並べて問う。機械体系が行為能力を獲得したように見えても、目標、展開、拡張は資本蓄積の論理に形づくられている。技術の自律性と資本の自律性は別の概念である。区別することで、企業の投資・管理決定を「AI自身が加速を要求する」と擬人化することを避けられる。この区別は機械の自律を組織問題へ戻す。技術能力の拡大は、展開の方向と利益配分を決める資本の権限を消さない。AI政治経済では、技術体系の能力増大と、資本の組織形態が与える社会的方向を区別する必要がある。Inhuman Powerの資本主義的枠組みは、AIを自律的行為者とする哲学的想像も修正する。アルゴリズムの目標は機械内部から自然に生まれず、投資、企業戦略、競争、管理上の必要から与えられる。機械の能力が個々の設計者の予想を超える複雑性を示しても、その社会的な展開には明確な制度的主体がある。システムの複雑さを理由に、政治的責任を擬人化して移してはならない。]
第8回:人工知能、自動化とポストヒューマンの未来
最終回では、AIを哲学、工学、政治経済学、未来の想像力の交差点に置く。Turingの機械知能に関する古典的な問い、Russellの人間と機械の目標関係、Bostromの超知能リスクの枠組み、Brassier、Land、Negarestaniによる理性、技術、加速への異なる考察は、同じ種類の議論ではない。それぞれ操作的テスト、工学的制御、思考実験、規範哲学、形而上学的推論に関わり、一つの証拠基準で互いに代替できない。AGIも、安定した定義と日程をすでに持つ実証的対象として扱うべきではなく、能力、自律性、汎用性、導入条件、制度的権力という別々の問いへ分解する必要がある。
Haraway、Hayles、Braidotti、Huiは、ポストヒューマニズム、身体性、関係性、技術の多元性から「人間」の境界を問い直す。ポストヒューマンは人間の強化や意識のアップロードと同じではなく、必ずしも技術楽観主義を意味しない。まず、理性、主体性、政治的資格を人間個人の内部に固定する従来の想像を問いに付す。他方、Crawford、Bender/Hanna、Narayanan/Kapoor、Haoは、知能をめぐる壮大な語りをデータ、労働、エネルギー、企業組織、産業競争、市場宣伝へ繰り返し戻し、「未来」を現実の制度へ引き戻す。
したがって本回が問うのは、未来の語り自体がどう政治的な力になるかである。産業のロードマップ、リスク言説、加速主義、解放的な技術の想像力は、何が実現可能で、必要で、不可避とみなされるかを先取りして規定する。AI批判は、すべての能力を誇大宣伝へ還元しても、一部のシステムに顕著な能力があるからといって産業の一括した「知能」という命名を受け入れてもならない。現実の能力、推測的リスク、政治経済の構造、規範的な未来は、それぞれ別に論証する必要がある。これらの層の間に明確な関連を築いてこそ、自動化とポストヒューマンの議論は、技術崇拝と技術恐怖の二項的な揺れを超えられる。
必読文献:
- Crawford, Kate. Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press, 2021. (Read: Introduction; Chapter 2, "Labor"; Chapter 3, "Data").
[注:Crawfordの重要な役割は、一般的知性を惑星的な物質性へ戻すことにある。AIが依存するのはアルゴリズムとモデルだけでなく、鉱物、エネルギー、データセンター、アノテーション労働、コンテンツ審査、世界的供給網でもある。第2章はAIの労働を再び見えるようにし、第3章はデータの資源化を示す。MarxやVirnoと併読すると、具体的な問いが生まれる。AIが社会的知性を私有化すると同時に、大量の物質的労働と生態的費用をモデルの背後へ隠すなら、なぜ「知能」はいつも脱身体的な形で現れるのか。Crawfordの役割は一般的知性の直接的な拡張ではなく、AIの現実条件を捉え直すことにある。Atlas of AIはEarth、Labor、Dataなどの尺度を通じて、AIを抽出の産業体系として捉える。鉱物、エネルギー、空間は計算の前提であり、アノテーションと保守は自動化の語りが隠す労働であり、データセットは分類の歴史をモデルへ運ぶ。Haoの組織調査とBender/Hannaの誇大宣伝批判と併読すると、「AIとは何か」を抽象的な知能の問いから、技術能力、産業基盤、制度的権力の複合的対象へ書き換えられる。最終回では現実の能力判断と哲学的可能性を区別しなければならない。あるシステムが今日できないことは、機械に永久に不可能だという論理的証明にはならず、逆に超知能が概念上可能でも、現行の導入の政治的優先度を自動的には証明しない。Crawfordの地図的方法はAIを、鉱物、エネルギー、労働、データ、分類、国家権力の惑星的な集合へ分解する。AIは宙に浮くソフトウェア能力ではなく、きわめて物質的・制度的な体系である。労働者、鉱山、データセンターを分析に戻すと、リスクの時間尺度が変わる。政治的帰結は未来の超知能の登場時に始まるのではなく、すでに資源採掘、労働編成、分類制度に存在する。]
- Bender, Emily M., and Alex Hanna. The AI Con: How to Fight Big Tech's Hype and Create the Future We Want. Harper, 2025. (Read: Introduction; selected chapters on AI hype, synthetic media, labor, and institutional power).
[注:BenderとHannaは、「AI」をまず産業言説、制度的選択、具体的な技術体系へ分解し、機械学習製品を統一的意志のある知的主体として擬人化することに反対する。焦点は誇大宣伝による注意の転換である。社会が遠い未来の機械支配を論じる間に、データ、労働、資源、ビジネスモデル、すでに起きている自動化の害を見落としうる。2025年のテクストとして、生成AI以後の新たな技術の語りへ議論を戻し、Crawfordの物質的政治を補完する。AIは高度に異質な総称である。予測モデル、生成モデル、ロボット、最適化システム、仮説上のAGIを同一の能力判断で扱うべきではない。判断は具体的な技術対象に対応させ、「AI」という総称で分類を代替してはならない。産業言説への批判は、具体的な技術対象、組織制度、観察可能な帰結に結びつけ、技術神話を別の壮大な擬人化の語りへ置き換えることを避ける必要がある。BenderとHannaはAIを産業の語りと権力装置として解体し、モデルの限界だけでなく、能力の誇張が投資、市場集中、政策課題にどう役立つかに注目する。技術リスク論と比べ、この視点は現存する制度的帰結へ注意を向ける。AIを定義できるのは誰か、神話から利益を得るのは誰か、導入費用を負うのは誰か。産業の誇大宣伝が政治的効果を持つのは、資源配分を変えられるからである。汎用知能と自動代替の壮大な語りは、投資、人材、規制の優先順位、公共の注意に影響し、現在の一部の問題を周辺化する。誇大宣伝の批判は技術変化の否定ではなく、能力の主張と社会的決定の間に、より厳格な証拠関係を求めることである。脱神話化は技術の重要性を下げるのではなく、判断を検証可能な能力、制度、資源関係へ戻す。]
- Narayanan, Arvind, and Sayash Kapoor. AI Snake Oil: What Artificial Intelligence Can Do, What It Can't, and How to Tell the Difference. Princeton University Press, 2024. (Read: Chapter 1, "Introduction"; Chapter 2, "How Predictive AI Goes Wrong"; Chapter 4, "The Long Road to Generative AI").
[注:NarayananとKapoorは、技術的認識論の「脱魔術化」を提供する。予測AI、生成AI、より広いAI言説を区別し、能力の境界と社会的な導入条件を強調する。本回ではこの修正が重要である。批判理論は、実際の技術能力と産業宣伝の「AI」を区別できなければならない。AI Snake Oil: What Artificial Intelligence Can Do, What It Can't, and How to Tell the Differenceは、「資源、労働、生態、組織史がAGI未来主義をどう制約するか」に、限定的だが有効な説明尺度を与える。仕組みを示せても、組織、法、物質条件の追加分析に取って代わるものではない。異なる技術対象には異なる証拠基準が必要であり、予測の失敗、生成モデルの能力、AGIの想像は互いに代替できない。すべてのシステムを同一の知能とみなす技術批判は、かえって産業マーケティングの曖昧さを再生産する。脱神秘化はまず分類の仕事である。AI Snake Oilの重要な方法は、能力の主張を使用場面に一つずつ対応させることにある。ベンチマークの成績は、採用、教育、医療の決定への適合性を自動的には証明せず、文章の流暢さも事実的な信頼性を証明しない。評価は抽象的能力から、導入条件、誤りの費用、代替手段へ移らなければならない。この区別は読書会の最終段階に適している。哲学的未来主義と現実のシステム能力が、一つの「AI」概念のなかで互いに置き換えられることを防げるからである。]
- Hao, Karen. Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI. Penguin Press, 2025. (Read: Introduction; selected chapters on OpenAI's scaling strategy, data labor, compute infrastructure, and global environmental impacts).
[注:Haoは長期の調査報道によってOpenAIとAI競争を再構成し、モデルの拡張を計算資源、資本、データ労働、世界的供給網、環境紛争へつなぐ。抽象的なAI哲学ではなく、組織史と世界各地の事例から、「スケール化」が産業戦略になる過程を示す。最終回の必読文献に置く意義は、AGI、超知能、ポストヒューマンの未来に関する哲学的推論を、現実の組織と資源政治によって検証することにある。誰がシステムを築き、誰が費用を負い、誰が未来を定義できるのか。ポストヒューマン批判も物質的政治からの逃避になってはならない。人間中心主義への疑問は重要だが、鉱物、労働、エネルギー、生態的費用、知的財産権は具体的な制度によって配分される。関係的主体の理論は、これらの構造と併読する必要がある。本回の文脈でEmpire of AI: Dreams and Nightmares in Sam Altman's OpenAIを読むと、リスク統治、解放の想像力、民主的制度の間で技術の未来がどう争われるかが具体化する。AIはまず、きわめて異質な対象の集合である。予測モデル、生成モデル、ロボット、最適化システム、仮説上のAGIは能力とリスクの構造が異なり、一つの「知能」概念にまとめると経験的判断が急速に曖昧になる。HaoはOpenAIの歴史を通じて、AGIの理想、資金調達、組織統治、計算資源競争を一つの語りに置く。価値は企業の内部事情だけでなく、「人類全体に利益をもたらす」という規範的展望が、商業拡大、人材競争、インフラ投資のなかで制度的圧力にさらされる過程を示す点にある。AI未来主義は組織から離れた観念史ではなく、企業統治と資本構造に密接に結びつく。企業史は、AI統治が企業内の使命宣言だけに依存できないことも示す。組織構造、取締役会の権限、投資家との関係、計算資源は、使命が商業判断を制約できる程度を変える。公益が企業の自主規制としてしか存在しなければ、規模拡大の競争のなかで再解釈されやすい。]
推薦文献:
- Turing, Alan M. "Computing Machinery and Intelligence" Mind 59, no. 236 (1950): 433–460. (Read: entire article).
[注:Turingの論文で最も重要なのは「機械は考えられるか」という一文ではなく、その問いを自ら書き換えたことである。模倣ゲーム、反論の一覧、学習機械を用い、どの行動基準なら機械知能を操作可能な研究対象にできるかを論じる。2020年代のAI言説の後で読み直すと、知能は自明な自然属性ではなく、テスト、課題、制度的期待によって定義されてきたとわかる。最終回では現実の能力判断と哲学的可能性を区別しなければならない。理論上の類似は制度上の同一性を自動的に意味しない。Turingの論文はAGIの道程よりも、「機械知能をどう論じるか」の方法論的転換として読むのが適切である。模倣ゲームは思考の本質に関する形而上学的論争を一時棚上げし、操作可能な行動基準に置き換える。この戦略は計算的知能の研究を開く一方、持続する問いを残す。テストは遂行能力を測れても、理解、主体性、社会的責任には自動的に答えられない。Turingは機械知能への多様な反論を先取りし、その論理的弱点を一つずつ示す。重要なのは「機械が考えると証明した」ことではなく、本質の定義から検証可能な問いへ議論を移した点である。今日もこの抑制を保つべきであり、モデルがテストに合格しても、支持されるのはその課題での能力判断に限られる。行動の遂行は証拠になりうるが、意識、理解、主体性の結論へ急拡大してはならない。ここでTuringが与えるのは知能を論じる際の概念的規律であり、技術の未来の予言ではない。操作主義は今日も鋭い問いを残す。テストが多様化すると、システムはある課題では優れていても別の課題では頑健でないことがある。「知能」を統一的な総称とし続けるか、異なる能力へ分けるべきか。この問いは、生成AIの能力主張の尺度と証拠基準に直結する。]
- Russell, Stuart. Human Compatible: Artificial Intelligence and the Problem of Control. Viking, 2019. (Read: Introduction; selected chapters on objectives, uncertainty, and the control problem).
[注:RussellはAI研究の内部から制御とアラインメントを問う。核心は特定の超知能災害の予言ではなく、機械の目標を単一の効用関数として事前に固定する設計前提への批判である。機械が人間の選好について不確実性を保つことを提案する。批判理論と併読する際には、その選好がどう形成され、誰が代表する権限を持ち、社会的対立を選好学習へ還元できるのかをさらに問うべきである。AIは高度に異質な総称である。Russellは先進AIのリスクを、目標設定と不確実性の管理として定式化する。機械が人間の与えた目標を完全に確定した効用関数と扱えば、能力が高まるほど設定の誤りが拡大しかねない。この枠組みは社会科学との対話に値する。人間の選好自体が統一的、透明、静的ではないからである。アラインメントは制御工学だけでなく、価値対立、民主的な権限付与、制度的代表にも関わる。「人間の価値」は一つにまとまった対象としてシステムへ直接入力できない。集団間には現実の対立があり、価値も状況によって変わる。アラインメントを選好学習だけで扱うと、民主政治の論争、権利、取引に付せない原則が、最適化変数へ圧縮されかねない。技術的制御の問題は必然的に政治哲学へつながる。現実の制度では、Russellの問いも複数主体間の対立を扱う必要がある。病院、企業、国家、個人に仕えるシステムは、誰の選好を基準とするのか。異なる価値を常に一つの目的関数へ圧縮できるわけではない。したがって研究には、手続的正義、権利による制限、民主的意思決定との接続が必要である。そうでなければ「人間と両立する」という語が、存在しない統一的人類意志を前提にしてしまう。]
- Haraway, Donna. "A Cyborg Manifesto" In Simians, Cyborgs, and Women, 149–181. Routledge, 1991.
[注:Harawayのサイボーグは、人間と機械の融合の単なる予言ではなく、境界を解体する政治的主体の理論である。「人間」が純粋な生物的身体と統一的アイデンティティを政治の基礎にしなければならないかを問い直し、マルチチュードに均質性を前提としない連合の想像を与える。「A Cyborg Manifesto」は、資源、労働、生態、組織史がAGI未来主義をどう制約するかを、組織と制度の次元へ導く。サイボーグ政治は、純粋で統一的かつ自然化されたアイデンティティを連合の基礎にすることを拒む。人間と機械、自然と文化の境界は技術的実践によって絶えず書き換えられ、政治的主体も異質な関係から成る。AIとの関連は融合の予言ではなく、本質主義的な人間性を仮定せずに責任と連帯を再編することにある。同時に、これは反本質主義的な連合政治である。純粋な女性、純粋な人間、自然的アイデンティティの代わりに、技術、身体、権力の交錯における部分的接続を強調する。AI時代の意義は「誰もが半分機械になる」ことではなく、政治的主体がもともと技術と社会関係によって構成されている点にある。責任と連帯は、技術以前の純粋な人類の回復には基づけない。サイボーグは、技術を外在的道具とみなす想像も崩す。身体、医療、伝達、労働はすでに技術装置によって再構成され、人間は技術の外で独立して存在する主体ではなかった。AI統治の「人類の保護」は、抽象的で純粋な人間性に訴えず、どの能力、関係、生活条件を守るかを具体化しなければならない。Harawayの意義は、この混成性を技術が招くアイデンティティの危機ではなく、政治的資源へ変える点にある。]
- Hayles, N. Katherine. Bacteria to AI: Human Futures with Our Nonhuman Symbionts. University of Chicago Press, 2025. (Read: Chapters 1–2, 6–7, and 9–10).
[注:Haylesはこの2025年の著作で統合的認知枠組みを提案し、物質性、身体性、関係性を重視しつつ、認知と意味生産を人間から生物や一部の人工システムへ広げる。How We Became Posthumanより現代AIの議論に適しているのは、GPT時代の機械認知の主張に直接向き合いながら、言語モデルを人間の心と同一視しないためである。価値は、強い人間中心主義と「機械はすでに主体になった」という主張の間に、第三の理論的道筋を与える点にある。ポストヒューマン批判も物質的政治からの逃避になってはならない。Bacteria to AI: Human Futures with Our Nonhuman Symbiontsが示す、リスク統治、解放の想像力、民主的制度の間で技術の未来が争われる問題は、テクスト内の概念論争を超える。Haylesは認知を、人間、非人間の生命、技術体系の関係網へ分散させ、知能を孤立した主体の内的能力だけとは捉えなくする。この2025年の議論は、生態とAIをつなぐのに適している。技術体系は生物的共生に埋め込まれると同時に、認知の境界についての人間の理解を変える。ポストヒューマンの未来は、まず関係の再編であり、単一種の能力向上ではない。Haylesはさらに、認知を尺度をまたぐ過程と捉える。細菌から技術体系まで、環境を感知し応答する仕方は異なりうる。この認知多元主義は、人間理性を唯一の尺度にすることに反対するが、あらゆる情報処理を人間の知能と同一視するものでもない。重要な貢献は、異なる認知形態を区別する言葉を与え、ポストヒューマンの議論を人間例外主義と機械の擬人化の間の揺れから解放することにある。]
- Bostrom, Nick. Superintelligence: Paths, Dangers, Strategies. Oxford University Press, 2014. (Read: Chapter 8, "Is the Default Outcome Doom?", pp. 115–142).
[注:Bostromは加速主義哲学とは大きく異なる存在的リスクの枠組みを示す。破滅の想定をあらかじめ受け入れる必要はなく、重要なのは論証構造の分析である。高度に自律的で目標指向的な行為者を仮定すると、なぜアラインメントが中心問題になるのか。政治経済学との比較によって、AIリスクを所有権、労働、制度から離れた純技術的問題へ抽象化できるかを問える。最終回では現実の能力判断と哲学的可能性を区別する必要がある。Superintelligence: Paths, Dangers, Strategiesの尺度では、AIの能力、産業の誇大宣伝、哲学的知能概念、政治経済の現実の区別は、まず具体的な関係として現れる。技術の未来は純粋な予測対象ではない。産業のロードマップ、リスクの語り、加速主義は投資、規制、公共の想像力に影響するため、「未来に何が起きるか」と「誰が未来を規定できるか」を併せて論じなければならない。Superintelligence: Paths, Dangers, Strategiesを通じて、この区別はより厳密な問いになる。実際に変わる仕組みは何で、単なる媒体の転換は何か、そして帰結を誰が負うのか。Bostromは極端な能力増大と目標の維持を前提に、存在的リスクと制御問題を導く。価値は未来のリスク推論の構造を明確に示す点にあり、限界は行為者、目標、技術経路の仮定に強く依存する点にある。現在の政治経済資料と並べることで、長期的シナリオ分析と、すでに観察できるインフラ権力を区別できる。]
- Brassier, Ray. "Prometheanism and Its Critics" In #Accelerate: The Accelerationist Reader, edited by Robin Mackay and Armen Avanessian, 467–488. Urbanomic, 2014.
[注:Brassierの価値はAIの開発路線の提示ではなく、技術、理性、政治的解放の関係を問題化する点にある。自然性、本真性、有機的生命を進歩の尺度として自動的に採用することを、なぜ拒むのかが読解の焦点となる。ここから反技術と反資本主義を区別する必要が生じる。技術自体を再編できるなら、技術の拒否は本当に解放をもたらすのか。同時に、プロメテウス主義の危険も検証しなければならない。普遍理性の政治が、再び差異と生態的条件への支配になりうるのか。「Prometheanism and Its Critics」、BostromのSuperintelligence: Paths, Dangers, Strategies、Landの「Teleoplexy: Notes on Acceleration」は、自動化、制御、アラインメント、ポストヒューマン主体の異なる問題を、異なる知的伝統に置く。AIは異質な総称である。Brassierの論文は、これらの問題を過度に抽象的な時代診断から解き放つ。技術の未来は予測だけの対象ではない。この論文の貢献は、理論を反証可能に保つ点にある。Brassierは技術的改変を理性の自己修正と結びつけ、既存の自然や伝統を超える実践をすべて支配衝動とみなすことに反対する。技術悲観主義に挑戦できるが、誰が世界を変える能力を持ち、どの制度が技術の方向を決めるかにも答える必要がある。政治経済学を離れると、プロメテウス主義は不平等に配分された現実の技術能力を、普遍的な「人類」の能力へ抽象化しやすい。]
- Land, Nick. "Teleoplexy: Notes on Acceleration" In #Accelerate: The Accelerationist Reader, edited by Robin Mackay and Armen Avanessian, 511–520. Urbanomic, 2014.
[注:Landは本回で最も重要な対抗的極点である。資本、技術、自己加速を、人間の意志を超える動的過程として描く。読む価値は極端な結論の受容ではなく、資本が自動的に加速する体系と想像されるとき、人間の政治的主体にどんな位置が残るかを観察することにある。Brassierと併読すると、理性、資本、未来をめぐる左右の加速主義の相違が見える。「Teleoplexy: Notes on Acceleration」を軸に、資源、労働、生態、組織史がAGI未来主義をどう制約するかを、関連する過程へ分解できる。技術の未来は予測だけの対象ではない。Landの論文から見ると、これらの制約は制度的な権限付与や資源配分から切り離せない。技術的形態は組織権力と結びついて初めて持続的な社会的帰結を生む。Landは資本と技術の加速を、人間の政治的統御を超える自己組織化として描き、加速主義を反人間主義の極へ進める。対抗的なテクストとして最も有益なのは、資本の動力を自然化する思考の論理を露わにする点にある。加速を止められないとみなせば、政治には適応か退出しか残らない。批判は、制度、労働、集団行動がなお技術の軌道を変えられると再び示す必要がある。Landは「不可避性」の修辞を検証する極端な事例でもある。資本の加速が政治的選択を超える過程とされると、制度と集団行動はあらかじめ無効と宣告される。この思考の批判は、単に減速を訴えることではない。技術的速度、資本間競争、社会的目標を区別し、その方向が必然的に一致しないと示すことである。]
- Negarestani, Reza. Intelligence and Spirit. Urbanomic, 2018. (Read: Part I, especially Chapter 1, pp. 1–52).
[注:Negarestaniは知能を、「人間の脳を機械で複製できるか」から、概念、推論、自己修正、規範性の問題へ移す。これは技術のロードマップではなく思弁哲学である。批判理論に極めて難しい問いを突きつける。知能を人間の自然的属性として定義すべきでないなら、「人工」の知能と「自然」の知能の政治的差異をどう理解すべきか。Intelligence and Spirit、Landの「Teleoplexy: Notes on Acceleration」、HuiのRecursivity and Contingencyは、リスク統治、解放の想像力、民主的制度の間で技術の未来がどう争われるかを、巨視的判断から仕組みの分析へ移す。ポストヒューマン批判も物質的政治からの逃避になってはならない。歴史的類比は仕組みの次元でのみ有益である。アルゴリズム、プラットフォーム、モデルが登場しただけで時代の交替を宣言すれば、理論は連続と断絶を区別する力を失う。Intelligence and Spiritの貢献は、大きな判断を識別できる制度的仕組みへ戻す点にある。技術の未来は予測だけの対象ではない。その争奪のあり方は、Intelligence and Spirit以外の実証資料と競合する説明によっても検証されなければならない。Negarestaniは知能を、理由を示し、コミットメントを修正し、規範の空間へ参加する実践として理解し、単なる問題解決能力とはみなさない。すると知能は特定の生物的基質に独占されないが、モデル性能の指標とも簡単には同一視できない。機械が規範的推論の空間へ入るかは、責任、概念使用、自己修正に関わり、「人間に似ているか」よりも厳密な問いとなる。]
- Hui, Yuk. Recursivity and Contingency. Rowman & Littlefield, 2019. (Read: Introduction, pp. 1–28).
[注:Huiは、近代技術の道は一つしかないという前提に挑む。再帰性と偶然性によって、技術は直線的進歩だけでなく、異なる宇宙論、文化、制度に形づくられた歴史的実践となる。AGIの未来も、集中、強力化、自動化が進む一方向に限定する必要はなく、技術制度には異なる歴史的経路がありうる。最終回では現実の能力と哲学的可能性を区別しなければならない。Recursivity and Contingencyは、AIの能力、産業の誇大宣伝、哲学的知能概念、政治経済の現実の区別に、より明確な因果の層を与える。Recursivity and Contingencyのこの説明は、対応する制度的仕組みを実証資料から識別できる場合に成立する。再帰性は技術の自律性への単純な称賛ではなく、偶然性とともに技術史の複数の経路を開き直す。近代技術だけが世界を組織する唯一の可能性ではなく、異なる宇宙論や制度的伝統から異なる技術関係が生まれうる。AIの未来も、同じ規模拡大と集中化の道と前提する必要はない。宇宙技芸という方向は、技術の多様性が同じ機器に別の文化的ラベルを貼ることではなく、技術の目的、自然観、社会秩序の異なる組合せであると示す。世界的なAIインフラに対して、これは製品の現地化より深い問いを提起する。異なる社会は、技術発展の尺度、速度、規範的目標を決める能力を持つか。技術の多様性は、まず政治的能力の問題である。既定の世界的インフラ上で文化的様式を選ぶだけでなく、自らの技術制度を本当に決定できるかが問われる。]
- Braidotti, Rosi. The Posthuman. Polity, 2013. (Read: Chapter 1, "Post-Humanism: Life beyond the Self," pp. 13–54).
[注:Braidottiはポストヒューマン主体と従来の人文主義的主体を区別し、トランスヒューマニズムの能力増強とは異なるポストヒューマニズムを示す。技術の未来は「より強い人間」を目標にすべきなのか、それとも生命、生態、関係的主体を捉え直せるのか。AIは異質な総称である。The Posthumanは、自動化、制御、アラインメント、ポストヒューマン主体の分析を証拠の限界へ戻す。技術、制度、政治的帰結の間に説明できる接続がなければ、理論的類比は示唆にとどまる。The Posthumanはトランスヒューマニズムの能力増強と区別しなければならない。より強い人間の製造ではなく、主体と非人間的生命、技術、環境の関係の再考が中心であり、AI未来主義への重要な対抗軸となる。Braidottiのポストヒューマニズムは技術的増強主義と根本的に異なる。自律的で理性的なヨーロッパ男性主体を尺度とする人文主義の普遍性を批判し、主体は常に生命、技術、生態の関係に埋め込まれていると強調する。したがってポストヒューマン政治は、強い個人の追求ではなく、人間、非人間的生命、技術体系、環境の間の依存と責任の再配分を問う。肯定の政治も、批判が人文主義の排除の暴露にとどまってはならないと教える。主体の関係性は新たな倫理の出発点になりうる。人間、動物、生態、技術体系の相互依存を認め、それに基づいて責任を配分し直すのである。AIの未来は能力向上だけでなく、どの生命や環境を技術発展の価値計算に含めるかにも関わる。]
- Srnicek, Nick, and Alex Williams. Inventing the Future: Postcapitalism and a World Without Work. Verso, 2015. (Read: Introduction; chapters on full automation, post-work politics, and institutional strategy).
[注:SrnicekとWilliamsは、自動化を技術的宿命ではなく左翼の政治戦略に位置づける。完全自動化、労働時間の短縮、普遍的ベーシックインカムなどによって、技術能力を資本の利潤論理から方向転換しようとする。最終回ではLand、Bostromと三角の緊張関係をなす。自動化は資本の自己加速なのか、制御すべき技術リスクなのか、それとも政治組織によって必要労働の削減へ変えられる制度条件なのか。Inventing the Future: Postcapitalism and a World Without Workは、資源、労働、生態、組織史がAGI未来主義をどう制約するかを、修辞から制度分析へ移す。Inventing the Futureは、自動化をポストワーク政治の制度的想像に置く。必要労働の削減が解放的になるには、所得、公共サービス、労働時間、所有権の同時の変化が必要である。そうでなければ同じ能力が失業リスクと管理強化にも転じうる。技術の潜在力と政治的結果の間に自動的な通路はない。ポストワーク政治は、労働時間の削減と有意義な活動の消滅も区別する。一部のタスクは自動化できても、ケア、教育、文化、政治参加の価値を効率の論理へ単純に還元できない。規範的な目標は必要労働と自由時間の再配分であり、「仕事がないこと」自体を終点とすべきではない。真に急進的なのは、自動化と社会的権利を一体で考える点にある。労働時間の短縮、基本所得、公共サービス、労働者の交渉力はどれも欠かせず、そうでなければ生産性の上昇は利潤と失業圧力として資本に吸収されうる。技術進歩自体に解放の方向はない。制度によって初めて、生産性を時間のより大きな自律へ変えられる。]
- Benjamin, Ruha. Imagination: A Manifesto. W. W. Norton, 2024. (Read: Introduction and chapters on imagination as a site of struggle).
[注:Benjaminは想像力を、柔らかな文化的付属物ではなく政治的資源として扱う。Fisher、Kornbluh、AGI未来主義と並べるのに適している。技術企業が製品ロードマップによって大衆の未来像を形づくれるなら、政治的抵抗も「どの未来を想像する価値があるか」を決める能力を奪い返す必要がある。ポストヒューマン批判も物質的政治からの逃避になってはならない。リスク統治、解放の想像力、民主的制度の間で技術の未来が争われるあり方は、Imagination: A Manifesto以外の実証資料と競合する説明によっても検証されなければならない。Benjaminは想像力を、私的な心理活動から公共の政治的能力へ引き上げる。未来は技術企業が届けるのを待つ製品ではなく、語り、設計、制度、集団行動によって争われる空間である。最終回に置くことで、ポストヒューマンとAGIの議論を民主主義へ戻せる。誰が未来を提案でき、どの集団の経験が未来の設計に含まれ、どの制度が代案を実行可能にするのか。想像力の政治は、既存制度を否定する能力にも関わる。技術統治は未来を現在の製品やビジネスモデルの漸進的な改良に限定しがちだが、Benjaminは抑圧された経験、歴史、共同体の実践から未来を再構成できると強調する。読書会が最後に問うのはこの開放性である。技術の未来は予測の対象だけでなく、制度的創造の対象でもある。こうして想像力は制度的実践につながる。代わりの未来が組織、公共資源、持続可能なルールにならなければ、再びブランド化された展望へ退いてしまう。この政治は技術設計にも直接つながる。データセット、製品ロードマップ、政策シナリオは「未来のユーザー」と「正常な社会」の想像を含み、設計はすでに可能性を配分している。想像力を広げるとは、影響を受ける集団を問題定義と目標設定に参加させることであり、完成後に意見を求めるだけではない。]
“数智时代的批判理论”夏季读书会导言
在福特主义与工业资本主义的全盛期,即我们所熟悉的早期甚或“古典”的现代性阶段,机器大生产建立在严密的物理空间区隔(Enclosure)与时间的线性切分之上。资本通过将劳动力强行集中于标准化的厂房流水线,进行对无产阶级肉身与时间双重维度的物理规训。这种基于空间高密度聚集的剥削模式,在客观上为经典的阶级联合、工会运动以及集体政治意识的觉醒提供了不可或缺的组织学前提。劳动者在同一屋檐下的物理共在(Co-presence),使得剥削的可见性转化为反抗的凝聚力。然而,进入控制论(Cybernetics)主导的后福特主义阶段,剥削的拓扑结构发生了根本性的本体论异变。资本的积累逻辑不再单纯依赖于物理空间中绝对剩余价值的直接榨取,而是转向了对非物质劳动(Immaterial Labor)、认知盈余以及由数字化平台持续捕获的“行为剩余(Behavioral Surplus)”的持续抽取。
在当下的数智平台与算法资本主义(Algorithmic Capitalism)语境中,传统意义上的物理工厂已被彻底解构与重组。劳动力被API接口、众包平台、外包数据标注网络与零工分配算法打散,呈现出高度的分布式、黑箱化与去中心化特征。控制的媒介从可见的砖石高墙与监工的凝视,下沉并弥散至不可见的TCP/IP协议、深度包检测(DPI)以及隐秘运作的预测性机器学习模型之中。在此背景下,传统的“占领华尔街”式物理空间抗争,由于在工具-认识论层面存在着严重的结构性错位,在“代码即法律(Code is Law)”的底层网络架构面前已不免归于失效。如若剥削不再体现为直接的暴力强制,而是化身为无缝嵌入日常生活的界面交互与个性化推荐时,反抗的物理抓手是否也随之消解?
本期夏季读书会旨在悬置传统的人文主义道德批判或过于简单的技术抵制论,将理论勘探的视域直接下沉至协议(Protocol)、密码学基建、行星级计算堆栈(The Stack)与异质性主体的微观权力运作网络之中。“数智时代”并不是一个用来概括技术进步的方便标签,而是一个需要被历史化和理论化的问题场域。数字平台、机器学习、云计算、数据中心、自动化管理与生成式人工智能正在重新编排资本积累、国家治理、媒介中介、劳动组织、知识生产和政治主体之间的关系,但这些变化既不能被简单描述为对工业资本主义的替代,也不能被还原为旧制度换了一层技术界面。更值得追问的是,哪些制度关系因计算而改变了尺度、速度与可见性,哪些权力形式只是借助新的媒介获得了更细密的操作条件,哪些被命名为“断裂”的现象实际上仍延续着管理科学、广告业、殖民治理、基础设施和劳动分工的长历史。由此,数字化与智能化既构成新的研究对象,也迫使批判理论重新检验自身概念的适用边界。
第一条线索围绕控制形式与可计算主体的生成展开。从Wiener的反馈与通信,到Foucault对规训装置的分析,再到Deleuze的modulation、Galloway的protocol和Rouvroy的algorithmic governmentality,几种理论所描述的并不是同一种权力机制的连续升级。规训依赖空间分隔、时间安排、检查与规范化;协议通过技术标准组织互操作性;算法系统则可能以分类、评分、预测、排序与推荐的形式塑造行动环境。它们既可能彼此叠加,也可能在国家法权、企业治理和技术架构之间形成冲突。问题因此不在于“算法是否成为新的主权者”,而在于计算结果如何进入制度程序,何种组织把概率输出转化为资格、风险或惩罚,谁拥有解释、修改和申诉的权力,以及模型中的不确定性如何在组织环节被固化为现实后果。
第二条线索进入监控、提取、殖民性与基础设施化。数据并非自然存在、等待采集的资源,而是在传感、日志、标签、身份系统、平台接口、商业模型和治理需求中被持续生产。Zuboff的surveillance capitalism、Couldry与Mejias的data colonialism、Browne对种族化监控的历史分析,以及Mbembe关于殖民性、敌意与生命暴露的政治思考,构成彼此并不完全相容的解释框架。“殖民”“封建”“帝国”等概念只有在对应机制能够被说明时才具有分析意义:资源占有、法律不对称、知识权威、强制性依赖、空间分割和差异化风险并不是可以相互替代的修辞。同样,“云”也不能停留在界面隐喻中。数据中心、海底光缆、芯片、能源、矿产、土地、频谱、标准、物流和维护劳动共同构成计算的物质条件,平台权力也因此必须与基础设施、领土、租金和退出成本一起讨论。
第三条线索关注平台政治经济、劳动与新的依附关系。Varoufakis与Durand提出的technofeudalism命题在这里被置于平台资本主义、垄断、租金理论与基础设施研究之间加以检验。真正需要辨析的不是“资本主义是否已经死亡”这一修辞性问题,而是云端入口、网络效应、无形资产与平台治理是否形成了传统资本理论难以充分说明的依附结构,以及平台租金与封建地租究竟在哪些层面可比、在哪些层面不可比。平台劳动研究则把这些宏观问题重新拉回订单、评分、派单、微任务、数据标注、内容审核、创业主体与全球劳动链。Berlant Cruel Optimism所提供的情感-时间理论在此尤为重要:不稳定的“自由”“灵活”和“自我创业”为什么仍然能够成为主体难以放弃的依恋对象,制度变革又如何同时触及所有权、风险分配、生活时间和未来预期。
第四条线索转向媒介中介、公共性与组织。数字政治经常被描述为连接成本下降、信息传播加速和守门机制削弱,但公共性并不等于流量,组织也不等于即时聚集。Kelty的recursive public强调公众能否维护和修改自身存在的技术-制度条件;Warner与Fraser讨论传播循环、反公众与参与不平等;Kornbluh Immediacy, or The Style of Too Late Capitalism则从文化形式上批判对透明、即时、沉浸和去中介化的迷恋。政治行动往往依赖程序、代表、记忆、延迟、解释、责任链与可持续组织,而平台媒介却可能把“立刻可见、立刻回应、立刻流通”塑造成价值本身。由此,去中心化并不天然等于民主,中介也不天然等于压迫;更重要的问题是中介能否被问责、被修改并由参与者共同塑造。
第五条线索集中于一般智力、算力集中与技术未来。Marx在Grundrisse“机器片段”中讨论社会知识、固定资本、劳动时间与价值形式之间的张力;Virno进一步强调general intellect并不只凝固于机器,也存在于语言、判断、记忆、协作和共同能力之中。生成式AI使这一问题以新的形式出现:模型训练依赖公共文本、代码、图像、反馈、标注劳动和巨额物理算力,而这些社会条件又能够被重新封装为私人模型、接口和云服务。需要分析的因此不是“LLM是否就是一般智力”,而是社会共同生产的认知痕迹通过何种产权和基础设施机制被集中,访问权、训练权、部署权与收益权如何分配,以及开放权重、公共算力、互操作和合作治理分别能够改变哪一层关系。
最后关于AI、自动化与后人类未来的讨论,并不以预测AGI的时间表为目标。Turing、Russell、Bostrom、Brassier、Land、Negarestani、Haraway、Hayles与Braidotti对“智能”“人”“机器”“控制”“解放”和“未来”有高度不同的理解;Crawford、Bender与Hanna、Narayanan与Kapoor、Hao则持续把这些哲学问题拉回现实技术、组织与资源关系。哲学推演、工程判断、产业预测、政治经济学和经验社会研究拥有不同的证据标准,彼此不能代替。描述性判断、因果解释与规范性判断也应保持区分:揭示某种权力机制并不会自动推出唯一的政治方案,技术上可行的制度安排也不因此获得正当性。在本读书会中,我们尝试在概念史、经验机制与规范问题之间建立可检验的联系,使经典理论既不被奉为万能钥匙,也不因时代差异而失去重新解释现实的能力。
| 节次 | 核心问题 |
| 第一节:控制论、控制与数字治理的生成 | 控制论、规训、控制社会与协议如何构成数字治理的历史前提? |
| 第二节:算法治理术与生命政治 | 分类、预测、评分与排序如何制造可计算主体并获得制度效力? |
| 第三节:监控、提取与数据殖民性 | 数据提取、监控、种族化分类与殖民性之间如何建立严格而非修辞性的联系? |
| 第四节:行星级计算与基础设施空间 | 云、平台、数据中心与基础设施如何重组空间、权威、租金和退出成本? |
| 第五节:递归公众、中介与数字政治 | 公共性如何拥有并修改自身媒介条件?为什么中介与延迟可能是政治组织的条件? |
| 第六节:平台劳动、不稳定性与新阶级关系 | 平台劳动、算法管理和不稳定生活如何重组劳动关系与集体行动? |
| 第七节:一般智力与对算力集中化的抵抗 | 社会共同智力如何被固定资本化、基础设施化,又可能如何重新公共化? |
| 第八节:人工智能、自动化与后人类未来 | 如何区分 AI 的现实能力、产业炒作、风险论证与后人类未来想象? |
第一组:控制的拓扑学(第1–2周)
第一节:控制论、控制与数字治理的生成
本节从数字政治理论经常略过的前史进入:在“算法”成为日常词汇之前,二十世纪已经形成了以communication、feedback、information和control重新描述机器、组织与社会的知识语言。Wiener的控制论、冷战时期的command-and-control、工业管理的信息处理需求、分布式通信网络以及软件协议共同构成数字治理的历史条件,但这些条件并不组成一条通往当代算法治理的线性必然链条。Foucault的disciplinary power依赖空间分隔、时间安排、检查和规范化,Deleuze的modulation指向连续而可变的控制形式,Galloway的protocol则处理分布式网络如何在允许连接的同时规定可操作性。三者的差异首先是对象、尺度与机制的差异。
围绕这一谱系,控制不能被简化为一个“中心是否存在”的问题。标准、协议、接口、身份、权限、硬件与法律分别在不同层次上形成约束;执行可以高度分布,而标准制定、资源所有权和修改权仍然可能集中。Baran的分布式通信首先是工程设计,May的密码无政府主义是一种政治宣言,Berners-Lee的Web提案起源于信息管理问题,DeNardis则显示技术标准始终处于跨国机构、企业和国家权力的交叉地带。同一种网络拓扑可以服务于军事韧性、开放协作、商业平台或监控体系,因此不能从“分布式”“开放”“加密”等架构名称直接推出其政治属性。
本节由此建立一个多层次的控制概念:技术系统既可能通过规则编码限定可操作性,也可能依赖市场集中、组织程序与国家法权获得实际效力。所谓“协议政治”并不是法律政治的替代物,而是标准、软件、制度和主权彼此嵌套的结果。这个区分为下一节的算法治理提供基础,因为只有先明确规则究竟发生在架构、协议、组织还是法律层,才可能进一步分析评分、预测与推荐系统如何把行为转化为可计算对象。
必读文献:
- Wiener, Norbert. The Human Use of Human Beings: Cybernetics and Society. Houghton Mifflin, 1950. (Read: Chapters 1–2, pp. 15–47).
【按:Wiener用communication、feedback与control重新描述人与机器以及社会组织的关系。这里并不是要把控制论直接当作算法治理理论,而是考察一种“以信息-反馈管理系统”的知识语言如何形成。它为后面的协议与算法提供历史背景,也提醒我们:今天所谓“数据驱动治理”的某些基本想象并非从互联网开始,而有更长的控制论谱系。更重要的是先明确原文本的对象和技术层,再寻找结构同构之处;一旦法律环境、组织主体或物质条件不同,就必须把类比降格为启发而不是证据。这一框架用于当代数字对象时,关键在于辨明相似性究竟发生在何种层次;一旦法律环境、组织主体或物质条件发生变化,概念的解释范围也随之改变。Wiener The Human Use of Human Beings: Cybernetics and Society进一步把“控制论、规训、控制社会与协议治理之间的历史差异”落实到制度关系之中。数字治理的关键不在于中心是否消失,而在于命令、反馈、标准与维护权如何在不同组织之间重新组合。执行可以高度分布,标准制定和资源所有权却仍可能集中,因此网络拓扑与政治去中心化必须分开判断。Wiener的意义还在于他并未把反馈仅仅视为机器内部的技术环节,而是把通信、熵与组织联系起来,由此形成一种能够跨越工程与社会理论的语言。恰恰因为这种语言具有高度可迁移性,它在后来管理学、军事系统和计算文化中获得了远超原始工程语境的影响。今天重新阅读控制论,更适合考察反馈何时成为管理原则、何时又被神话为社会自我调节的普遍模型。】
- Foucault, Michel. Discipline and Punish: The Birth of the Prison. Translated by Alan Sheridan. Vintage Books, 1977. Originally published 1975. (Read: Part III, Chapter 3, "Panopticism," pp. 195–228).
【按:本章构成比较性的理论起点。Foucault讨论的并不是“监控”本身,而是纪律如何通过空间分隔、时间安排、检查、记录和可见性的制度化来生产可治理的身体。全景敞视的关键不是某个持续盯着人的监视者,而是让被观察者在原则上始终处于“可能被看见”的状态,从而将外部强制转化为主体的自我调整。将这一框架迁移到数字平台时必须保持距离:数据系统并不简单复制建筑空间中的身体定位;它更可能通过日志、权限、评分、推荐与身份系统,把行动转化为可以被计算和再调用的数据对象。由此可以进一步追问:当监视从“谁在看我”转向“什么系统能够记录并处理我的行动”时,全景敞视理论需要怎样被改写?平台数据化与规训装置之间的关系只能通过具体机制比较,不能由“可见性”这一共同词汇直接推出。数字平台与全景敞视之间真正可比较的部分,不是“看见”这一表象,而是可见性如何被制度化并进入主体形成。平台记录往往同时服务于商业预测、风险控制和组织管理,其数据可见性也并非对所有行动者对称开放。由此,数字监控更适合被理解为规训技术与数据基础设施之间的重新组合,而不是Bentham式建筑模型的简单电子化。更进一步说,数字环境改变了纪律权力的时间结构。传统检查往往以明确时点发生,而平台记录具有连续积累和事后重算的可能;一个过去看似无关的行为可以在新的模型或政策中获得新的意义。这种可回溯、可再计算的可见性,使数字治理既继承规训的记录逻辑,又形成不同于监狱建筑的时间政治。】
- Deleuze, Gilles. "Postscript on the Societies of Control" October 59 (1992): 3–7. Originally published 1990.
【按:Deleuze的“控制社会”并不是对互联网的经验性预测,而是一个关于权力形式变化的高度压缩模型。核心是从封闭的disciplinary mold转向连续的modulation;"dividual"则意味着个体可以以多个数据片段、信用、密码和身份形式进入不同系统。这里应把它当作问题生成器,而不是把dividual直接翻译成“数据画像”。若将这一文本用于今天的数字治理,最需要补上的不是更多技术名词,而是实施链条:规则在哪里被编码,谁部署,谁维护,何种法律和市场关系让规则有效。“分布式拓扑与政治去中心化并非同一回事”在Deleuze "Postscript on the Societies of Control"这里因此不只是概念问题。数字治理的关键不在于中心是否消失,而在于命令、反馈、标准与维护权如何在不同组织之间重新组合。执行可以高度分布,标准制定和资源所有权却仍可能集中,因此网络拓扑与政治去中心化必须分开判断。Deleuze的mold/modulation区分尤其适合描述权力形式从阶段性认证转向连续更新的变化:账号状态、信用额度、价格、权限与推荐都可以随数据流实时调整。但“控制社会”不是对一切数字现象的总称。调制之所以具有分析力,正在于它指向持续变动的准入条件与身份分割,而国家法律、组织规章和物理强制仍可能与之并存。“控制社会”最值得保留的并不是一个历史阶段标签,而是对权力时间性的描述:控制通过密码、额度、资格和不断更新的状态持续调整行动空间。它与规训并不互相排斥,学校、监狱、企业等封闭制度仍然存在,只是越来越多地与连续评估和跨场景数据流结合。】
- Galloway, Alexander R.. Protocol: How Control Exists after Decentralization. MIT Press, 2004. (Read: Introduction; Chapter 1, "Physical Media"; and Chapter 2, "Form," pp. 1–92).
【按:Galloway的核心贡献是把控制问题从“中心-边缘”的简单模型中移开。他关心的是协议怎样在去中心化网络中建立规则关系:TCP/IP、DNS等系统可以允许大量节点彼此通信,同时通过标准化的地址、格式、权限和转换方式规定什么样的行动能够发生。这里尤其需要区分“协议是控制”与“协议就是法律”之间的差异:Galloway讨论的是技术系统如何具有政治形式,而不是说国家法律已经被网络协议取代。在理论关系上,它承接Foucault的纪律问题,同时为之后的算法治理、区块链和云平台提供概念过渡。我们要进一步问的是:当平台把注册、身份、接口、排序和访问权限都写进技术结构时,控制是否因此变得更少依赖明确的命令?从这一文本延伸到平台、云或AI时,需要把历史连续性与机制差异同时保留下来;前者说明问题为何重现,后者则限定概念不能越过的边界。协议政治由此呈现出一个看似悖论的结构:通信越依赖共同标准,网络越可能在执行层面分布,却在标准制定、命名体系或关键资源上形成新的集中。Galloway的价值正是在于把“去中心化”从规范性赞词还原为可分析的技术-政治关系,使拓扑结构、治理权限与所有权不再被混为一谈。协议还把政治冲突转移到版本、兼容性和标准演化之中。谁有资格提出修改、什么变化会导致分叉、旧节点是否仍能接入,都是技术共同体中的治理问题。因而protocol并非静止规则,而是一套不断维护的制度过程;控制也不只发生在执行时,更发生在标准被制定和更新的时刻。】
推荐文献:
- Lessig, Lawrence. Code and Other Laws of Cyberspace. Basic Books, 1999. (Read: Chapter 7, "What Things Regulate," pp. 85–99).
【按:Lessig的"code"论述最值得注意的地方恰恰在于它并没有把技术系统与法律制度截然分开。他指出行为受到法律、社会规范、市场和架构四种规制力量共同影响;网络中的代码因此可以通过改变行动条件来实施规制,但代码的有效性仍嵌入法律、经济和社会关系。阅读时应避免把"Code is Law"变成一句口号,而应分析:一个平台究竟如何通过密码、接口、身份验证、默认设置、支付规则和内容排序,把某种行为从“违法”进一步转变成“根本无法操作”?这一问题构成理解“架构性权力”的基础。技术结构的政治效力总要经过组织和制度。协议能够规定互操作性,硬件决定可实现范围,市场集中塑造退出成本,国家法权则决定某些规则何时具有强制力;这些层次共同构成数字控制的实际形态。Lessig的四种规制力量——law、norms、market与architecture——使数字治理不必被压缩为“代码决定一切”。架构可以让某些行为在技术上不可能,也可以提高行动成本,但代码本身的设计与执行仍受企业商业模式、法律责任和社会规范影响。平台治理最值得研究的正是四者如何叠合:默认设置、价格机制、社区规则和国家监管往往在同一界面中共同发挥作用。Lessig的框架也提示,平台权力最稳定的形态往往不是单一规制方式,而是多重力量互相加固。例如商业激励推动某种界面设计,界面又改变用户规范,随后法律以平台既成技术条件为治理对象。数字架构的政治性正产生于这种循环,而不是代码本身拥有超越社会的独立主权。】
- Beniger, James R.. The Control Revolution: Technological and Economic Origins of the Information Society. Harvard University Press, 1986. (Read: Introduction and Chapter 1).
【按:Beniger把现代信息社会放回工业化带来的“控制危机”:生产、运输、分配和消费速度不断提高,旧有组织方式越来越难以处理这种复杂性,于是信息处理与管理技术不断发展。它能够避免一种“互联网突然发明控制”的故事,并把平台、物流、算法调度放进更长的管理史。讨论重点是控制究竟服务于什么样的经济和组织需求。技术结构的政治效力总要经过组织和制度。这一长时段视角尤其适合解释平台物流、实时定价和算法调度:它们并非凭空出现的“智能”,而是对速度、规模和复杂性持续增长的管理回应。Beniger因此把信息技术史重新接回资本主义组织史;所谓数字化转型,也必须同时考察生产与流通的加速如何制造新的协调压力,以及谁有能力承担这种协调。从组织史看,“控制危机”还揭示了信息处理与规模扩张之间的共生关系。企业只有在能够追踪库存、运输、消费者和劳动流程时,才可能维持更大范围的协调;新的信息技术既回应复杂性,也使进一步扩张成为可能。数字平台的增长同样可放入这一循环,而不必被描述为纯粹技术突破。因此,信息处理能力本身也可以被视为组织扩张的生产条件:它既解决既有复杂性,又为更大规模的集中协调打开空间。这种组织-信息的双向强化,也使Beniger成为连接工业管理史与平台治理史的重要中介。它因此把管理技术与资本主义规模问题放在同一历史框架中。信息革命于是被还原为组织能力与经济结构共同演化的结果。这一视角也能解释为何“实时性”会成为现代管理持续追求的能力。】
- Baran, Paul. "On Distributed Communications Networks" IEEE Transactions on Communications Systems 12, no. 1 (1964): 1–9.
【按:Baran的论文应作为工程史文本阅读。它提出的关键问题是通信系统如何在高风险环境下提高冗余与生存能力,而不是建立一种政治意义上的“去中心化”。正因为如此,它特别适合检验后来数字政治中的类比:同一种分布式拓扑为什么可以被解释为军事韧性、网络自由、平台弹性乃至监控基础设施?技术结构本身并不携带单一政治方向。“分布式拓扑与政治去中心化并非同一回事”通过Baran "On Distributed Communications Networks"进入了组织与制度层面。技术结构的政治效力总要经过组织和制度。 在Baran "On Distributed Communications Networks"的尺度上,“分布式拓扑与政治去中心化并非同一回事”需要同时处理连续性与变形:旧制度可能借新技术获得新的操作方式,新技术也可能在既有制度中被重新吸收。Baran的工程论证也提醒我们,分布式网络首先是一种可靠性方案,而不是政治制度。节点冗余能够提高通信系统在破坏中的生存能力,却不能回答谁拥有网络、谁制定协议以及谁能够切断接入。后来把"distributed"直接翻译成自由、自治或民主,实质上把工程拓扑转换成了未经证明的规范判断。工程上的冗余还揭示了“韧性”与“自治”之间的差别。一个网络可以因为多路径设计而更难被摧毁,却仍然由少数机构掌握设备、频谱和接入许可。Baran的价值正在于把这一差别保留在技术史内部,使后来关于分布式网络的政治想象必须回答所有权与治理问题,而不能只凭拓扑结构推断权力关系。】
- DeNardis, Laura. The Global War for Internet Governance. Yale University Press, 2014. (Read: Chapter 1, pp. 1–24).
【按:DeNardis直接把互联网的技术治理问题放到全球政治中。IP地址、域名、标准、路由与安全并不是纯粹“技术问题”,但也不能简单被还原为国家主权。它可以与Galloway对读,使我们看到协议、企业、国际组织和国家机构如何共同生产网络秩序。DeNardis The Global War for Internet Governance所揭示的“技术设计、维护劳动与治理合法性之间的连接”并不止于文本内部的概念争论。技术结构的政治效力总要经过组织和制度。DeNardis The Global War for Internet Governance使“技术设计、维护劳动与治理合法性之间的连接”能够被限定在特定对象与因果链条之中;超出这一范围的类比,需要新的证据,而不能依靠术语本身完成。DeNardis把互联网治理推进到一个更精确的层面:域名、地址分配、路由、安全证书和技术标准既具有工程属性,也构成全球权力关系。网络秩序并不是在国家与互联网之间二选一,而由政府、标准组织、运营商、平台企业和技术社群共同维系。基础设施治理因此常常发生在公众不易察觉,却能深刻改变可接入性与表达条件的技术层。互联网治理的复杂性也说明,技术层并不是政治的前阶段。标准组织如何达成共识、运营商怎样交换流量、证书机构如何建立信任,这些看似后台的安排本身就是制度。DeNardis由此把网络主权拆成多个可识别的治理点,使“国家控制互联网”或“互联网超越国家”这两种总括命题都显得过于粗糙。】
- Turner, Fred. From Counterculture to Cyberculture: Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianism. University of Chicago Press, 2006. (Read: Chapter 1, pp. 11–39).
【按:Turner追踪美国反主流文化、Whole Earth网络与个人计算文化之间的历史转译。适合检验一种常见的Silicon Valley起源叙事:60年代的共同体、个人自治、技术媒介和网络想象并不是自然地走向平台资本主义,而是在不同组织、商业和文化力量中不断重组。在Turner From Counterculture to Cyberculture: Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianism的分析尺度上,“控制论、规训、控制社会与协议治理之间的历史差异”首先表现为一组具体关系。更重要的是,它使反馈、信息与适应成为组织思维的一部分,而后来的规训、协议和平台治理又分别以不同方式重新组织这些关系。Turner的历史叙述显示,反文化共同体与后来的数字资本主义之间并非简单的背叛关系。个人自治、去科层化、网络协作和创业文化在特定组织环境中被重新组合,逐渐形成一种兼具共同体想象与市场逻辑的技术文化。理解这一转化,有助于解释为什么开放、参与和创新这些词能够同时服务于反体制政治与企业管理。这一历史也能解释硅谷文化中一个持久悖论:反科层话语经常与高度集中的企业组织并存。个人自由在用户界面和创业神话中被强调,后台基础设施、资本控制和劳动管理却可能更加集中。反文化遗产因此不是平台资本主义的简单对立面,而是其文化合法性来源之一。】
- Kittler, Friedrich. Optical Media. Translated by Anthony Enns. Polity, 2010. Originally published 2002. (Read: "Theoretical Presuppositions," pp. 29–46).
【按:Kittler强调媒介的物理条件与技术存储/传输能力对文化形式的塑造。放在协议政治之前,可以帮助我们避免把代码理解成纯抽象符号:协议必须运行在硬件、频谱、存储与计算能力之上。其媒介决定论需要被批判性阅读,但正是这种极端立场能够迫使数字媒介研究认真面对数字权力的物质底座。Kittler Optical Media与Turner From Counterculture to Cyberculture: Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianism和Barbrook "The Californian Ideology"把“标准、协议、软件、硬件与制度权力的分层关系”置于不同的知识传统中。Kittler Optical Media让“标准、协议、软件、硬件与制度权力的分层关系”摆脱了过于抽象的时代判断。Kittler Optical Media对“标准、协议、软件、硬件与制度权力的分层关系”的贡献,正在于让理论保持可反驳性。媒介物质性的强调也改变了软件政治的分析方向。所谓“云端”“实时”或“无缝体验”都依赖具体的存储、传输、显示与计算装置;这些装置决定哪些信息可以被记录、以何种速度流通、在何处发生失真。Kittler的强硬媒介论虽然不能取代制度分析,却能防止批判停留在界面与话语层面。】
- Barbrook, Richard, and Andy Cameron. "The Californian Ideology" Science as Culture 6, no. 1 (1996): 44–72.
【按:文章批判数字技术文化把自由主义个人主义、反官僚主义与技术进步结合起来的意识形态形式。它与Turner构成一组历史-意识形态对读:前者讨论文化想象如何被重新组织,后者则提供历史谱系。这里不宜把"Californian Ideology"当作对Silicon Valley的万能解释,而应考察其揭示了哪些真实张力。围绕Barbrook "The Californian Ideology"展开时,“分布式拓扑与政治去中心化并非同一回事”可以被重新拆分为若干相互关联的环节。 从Barbrook "The Californian Ideology"来看,“分布式拓扑与政治去中心化并非同一回事”不能脱离制度授权与资源分配单独成立;技术形式只有与组织权力结合,才会产生稳定的社会后果。文章最值得保留的不是一个固定的“硅谷意识形态”标签,而是它揭示的政治混合物:反科层、个人解放与市场自由可以在技术乌托邦中彼此缠绕。由此,开放网络或个人计算机并不天然对应某种左翼或右翼政治;其政治方向取决于产权结构、劳动关系以及公共投入被如何叙述和分配。"Californian Ideology"的批判还涉及公共投资如何在技术神话中被抹去。互联网、计算机与科研体系长期依赖国家和大学资源,商业成功却常被重新叙述为个人创业与市场自发性的胜利。这种记忆政治帮助塑造了硅谷对监管、税收和公共责任的特殊态度。这种意识形态之所以长期有效,还因为它能够把相互冲突的价值重新包装成一致叙事:个人自治与风险资本、反官僚主义与企业扩张、开放文化与知识产权可以在“创新”名义下共存。批判的任务不是寻找纯粹起源,而是分析这些矛盾如何被技术进步的时间叙事暂时缝合。】
- May, Timothy C.. "The Crypto Anarchist Manifesto" In High Noon on the Electronic Frontier, edited by Peter Ludlow, 41–44. MIT Press, 1996. Originally written in 1988 and circulated in the early cypherpunk milieu; read at the September 1992 founding meeting. (Read: entire manifesto, pp. 41–44).
【按:这篇文本首先是一份历史语境材料,而不是足以独立承担规范判断的理论权威。May把强加密、匿名通信、数字现金和电子契约想象成削弱国家监管、税收和信息控制能力的技术条件,同时也暴露出密码朋克文化中明显的自由意志主义和技术决定论倾向。它在此处的意义,在于由此可以看到“去中心化”“隐私”“匿名”“代码自治”这些后来被不同阵营反复使用的词汇,最初是在怎样的政治想象中被连接起来的。这里尤其需要把它与Galloway、Lessig对照:如果技术架构确实具有规制能力,那么加密为什么既可以被理解为反抗国家监控的工具,也可以成为新的私人权力基础设施?May "The Crypto Anarchist Manifesto"与Barbrook "The Californian Ideology"和Berners-Lee "Information Management: A Proposal"共同使“技术设计、维护劳动与治理合法性之间的连接”从宏观判断下沉到机制层。May "The Crypto Anarchist Manifesto"对“技术设计、维护劳动与治理合法性之间的连接”的贡献,在于把宏观判断压回可识别的制度机制。May的文本也说明,密码学的政治意义从一开始就不是单一的。隐私技术能够限制国家或企业的可见性,却也可能弱化税收、劳动保护与公共监管的实施条件。围绕加密技术的争论因此不能停留在“国家-个人”的二元对立,而应把平台资本、金融中介、基础设施运营者和公共责任同时纳入分析。】
- Berners-Lee, Tim. "Information Management: A Proposal" CERN, 1989. (Read: entire proposal).
【按:这份提案不是成熟的Web乌托邦宣言,而是解决CERN信息管理问题的工程提案。它作为原始设计文本很重要,因为它由此可以看到超文本、链接、网络共享与开放信息的初衷。将它与后来围墙花园、平台经济和云服务比较,可以讨论设计愿景如何在商业化、标准化和资本积累过程中发生转向。“控制论、规训、控制社会与协议治理之间的历史差异”从Berners-Lee "Information Management: A Proposal"出发,可以获得更清楚的因果层次。工程设计本身不携带固定政治方向。开放、加密、分布式或模块化都可能在不同制度中产生相反后果;判断其政治含义,仍要回到标准制定权、修改权、资源所有权与故障处置权。从这一原始提案回望平台化Web,还可以看到“链接”与“控制入口”之间的历史分叉。开放超文本所强调的是异质信息之间可扩展的关联,而平台化则越来越依赖账号体系、推荐逻辑和专有接口来组织可见性。二者的差异提醒我们,网络技术的政治效果往往产生于后来形成的商业和制度结构,而不由最初的工程方案预先决定。Berners-Lee的方案还保留了一个后来愈发重要的制度问题:开放协议能够保证互联,却不能保证信息生态不会被少数入口重新集中。Web的基础标准仍然开放,但搜索、社交、身份和云服务可以在其上形成新的瓶颈。开放底层与集中上层的并存,是理解当代互联网政治不可忽略的结构。这使开放标准与平台集中之间的关系成为理解Web历史的一条重要线索:底层互联并不会自动阻止上层入口重新形成垄断。】
- Edwards, Paul N.. The Closed World: Computers and the Politics of Discourse in Cold War America. MIT Press, 1996. (Read: Introduction; selected chapters on command-control systems and Cold War computing).
【按:Edwards将冷战时期的计算、军事指挥控制系统与一种更广泛的“封闭世界”政治想象放在同一历史框架中考察。它的价值不在于把所有数字技术都归因于军方,而在于说明计算架构、模拟、指挥链和政治叙事曾经如何彼此塑造。对于本节而言,这提供了Wiener的控制论与后来网络治理之间不可缺少的制度史中介,也提醒我们不要把“控制”只当作软件层面的抽象形式。工程设计本身不携带固定政治方向。Edwards所说的closed world把冷战计算机置于军事指挥、模拟与封闭系统想象中。计算并非只是一种工具,也参与塑造何谓敌人、环境、决策和可控世界的政治认识论。由此,数字治理的历史不能仅从民用互联网追溯;军事组织对实时信息、指挥控制和系统集成的需求,同样深刻参与了计算基础设施的形成。closed world的封闭性也体现在知识生产方式上:复杂现实被转化为可模拟、可计算和可指挥的系统,决策者由此获得一种“从中心看全局”的想象。今天的数据仪表盘和实时监控延续了部分这种认识论冲动,但其组织结构与冷战军事系统并不相同。比较的重点应是系统可视化如何塑造决策者对可控性的信念。从这一角度看,仪表盘、模拟和实时态势感知并非单纯提高决策效率,而会塑造何谓“完整信息”和“理性控制”的标准。系统越强调可视化全局,越可能把无法进入模型的地方知识、模糊性和政治争议处理为噪声。计算治理的认识论边界因此与组织权力直接相连。这种系统视角也会影响责任配置:当失败被归因于“信息不足”或“模型不完整”,组织往往进一步追求数据扩张,而较少反思目标设定本身。closed world因此也是一种不断要求更多可计算性的治理逻辑。】
- Pickering, Andrew. The Cybernetic Brain: Sketches of Another Future. University of Chicago Press, 2010. (Read: Introduction; selected chapters on British cybernetics).
【按:Pickering通过英国控制论传统中的具体人物与实验,强调cybernetics不只有一种通向集中管理、预测和优化的历史道路。他用performative、adaptive的视角展示另一种与未知环境持续耦合的机器观。放在Wiener、Beniger之后,这本书的重要作用是给“控制论必然导向控制社会”的线性谱系踩刹车:同一组反馈、适应和系统概念,在不同制度与实践中可能形成很不一样的政治含义。借助Pickering The Cybernetic Brain: Sketches of Another Future,“分布式拓扑与政治去中心化并非同一回事”能够从修辞性命题转化为制度分析。工程设计本身不携带固定政治方向。 在Pickering The Cybernetic Brain: Sketches of Another Future这里,“分布式拓扑与政治去中心化并非同一回事”并不导向单一结论,而是要求比较不同组织场景中的权力配置,由此判断概念的解释范围与失效条件。Pickering通过英国控制论的异端传统展示,控制论并不必然导向集中控制。Ashby、Pask等人的实验更关注适应、涌现和与复杂环境共舞,由此形成与command-and-control不同的技术想象。这一差异能够防止把“控制论”视为单一政治谱系,也说明同一套反馈概念可以导向不同的工程实践与社会哲学。Pickering的"performative"控制论还提供另一种技术政治:与其预设环境可以被完整表征,不如让系统通过反馈不断与未知环境适应。这种姿态并不天然民主,却能对“预测得越准确就治理得越好”的数据主义形成重要反例。复杂性并不总要通过更全面的模型解决,也可以通过开放的适应关系处理。】
- Chun, Wendy Hui Kyong. Programmed Visions: Software and Memory. MIT Press, 2011. (Read: Introduction; Chapter 1, "On Software, or the Persistence of Visual Knowledge").
【按:Chun追问software为什么会被经验成一种既不可见又仿佛能够直接命令机器的东西,并把软件的可执行性、界面、记忆与控制话语放进媒介史中。它能修正把protocol或code当作纯逻辑规则的倾向:代码的权力效果依赖编译、硬件、界面、制度与用户实践,“可编程”也不是无摩擦的主权命令。工程设计本身不携带固定政治方向。Chun Programmed Visions: Software and Memory最终使“技术设计、维护劳动与治理合法性之间的连接”成为一个可以被经验研究继续拆解的问题,而不是一个预先完成的时代诊断。Chun尤其揭示了软件的双重性:它既被呈现为透明、可执行的命令,又必须依靠大量不可见的抽象层与物质过程才能运行。正是在这种“可见的控制感”与“不可见的技术中介”之间,软件取得了独特的治理效力。对数字政治而言,分析代码不能只问它写了什么,还要问它如何被编译、部署、更新并嵌入组织。软件的“可执行性”也容易制造一种直接性的幻觉,仿佛代码写下之后就能无摩擦地实现意图。实际上,执行始终经过操作系统、硬件、库、网络与用户环境,错误、兼容和更新都是系统常态。Chun由此使software从抽象命令重新变成具有历史和物质条件的媒介。软件因此应被理解为一种中介制度:它把复杂硬件转换成可操作界面,又通过抽象层隐藏大量执行条件。用户所经验到的“点击即发生”来自持续维护的技术链,而不是命令与结果之间不存在距离。正是在这种被隐藏的中介性中,软件获得了强大的日常治理效果。】
第二节:算法治理术与生命政治
本节从“规则如何规定行动”转向“计算如何组织行动的可能性”。Rouvroy提出的algorithmic governmentality并不以重建主体的意义世界为前提,而是通过数据收集、相关性分析和概率预测,在行动发生之前形成分类、风险与机会结构。Cheney-Lippold的algorithmic identity进一步说明,主体可以被拆分成持续更新的数据类别;Bucher则拒绝把“算法”想象为单一而神秘的黑箱,强调代码、训练数据、平台目标、工程实践、用户行为与制度环境共同构成算法权力。由此,可计算主体并不是一个被机器完整“认识”的人,而是在不同制度场景中被重新编码、分组和排序的行动者。
算法的政治效力也不能仅由模型性能解释。一个概率输出只有在贷款审批、福利资格、边境审查、招聘筛选、内容分发或保险定价等组织程序中被赋予行动后果,才成为治理的一部分。因此,预测、解释、判断和正当化必须区分开来:统计上的相关性并不等于因果说明,模型给出的风险分数也不能自动承担法律与伦理上的理由。Amoore对不确定性与概率判断的分析、Eubanks对福利自动化的经验研究、Pasquale对黑箱与问责的讨论,都指向同一问题——技术输出如何在机构内部被转译为具有现实约束力的决定。
本节同时把“算法偏见”从笼统的价值判断还原为制度链条。训练数据、标签体系、目标变量、评价指标、阈值设定、部署环境、人工复核和申诉渠道都可能成为不平等生成或放大的环节。分类并非纯粹描述世界,它会在被组织采用之后参与塑造机会结构与社会身份。算法治理术因此既涉及知识问题,也涉及行政程序、组织责任与权利保障;可计算性的扩展并没有取消政治判断,而是在许多场景中改变了判断被隐藏、分配和追责的方式。
必读文献:
- Rouvroy, Antoinette. "The End(s) of Critique: Data Behaviourism versus Due Process" In Privacy, Due Process and the Computational Turn, edited by Mireille Hildebrandt and Katja de Vries, 143–167. Routledge, 2013.
【按:本章是本节的概念核心。Rouvroy所说的algorithmic governmentality强调一种不必首先理解主体意义、动机和意图,而通过数据收集、相关性分析和预测来进行行动前干预的治理。关键在于区分correlation/causation、prediction/justification、profiling/judgment之间的差异:算法并不需要“真正知道你是谁”,它只需要让制度在概率上相信某些分类足以支撑行动。这样才能理解算法治理与传统法律程序之间真正的张力。更细地说,这一章与due process的冲突不在于“机器没有人性”,而在于数据行为主义倾向于把主体从可说明理由、可回应指控的法律-政治人格转换为概率性行为轮廓。相关性分析可以在不建立因果解释的情况下支持预防性干预,这使critique本身面临困难:如果系统不再要求你是谁、为什么如此,只需要依据过去数据调整你将遇到的环境,那么传统以揭露意识形态或纠正错误表征为中心的批判是否仍然足够?因此本章应与Yeung、Eubanks和Amoore对读,分别从规制、福利行政与安全判断检验algorithmic governmentality是否在不同制度中同样成立。Rouvroy的data behaviourism之所以构成算法治理术的核心,在于它尝试绕开传统主体的意向、叙述与规范判断,直接从海量数据相关性中发现行为模式。治理因而可以在主体尚未表达之前,以概率方式预先塑造其机会空间。真正的法理问题随之出现:相关性可以支撑预测,却不能自动提供理由;程序正义要求的不只是准确结果,还包括可以争辩的规范依据。】
- Cheney-Lippold, John. "A New Algorithmic Identity: Soft Biopolitics and the Modulation of Control" Theory, Culture & Society 28, no. 6 (2011): 164–181.
【按:Cheney-Lippold提出algorithmic identity,用来解释数字系统如何通过行为数据生产动态类别。它适合与Deleuze对读:身份不再只是由主体自我声明或人口统计学固定决定,也可以成为系统根据行为不断重新计算的概率对象。其限度同样明确:算法分类仍然以社会既有类别为材料,并非脱离历史的纯数学分类。algorithmic identity与传统人口分类的连续性,应落实在类别来源、制度用途与更新机制上,而不能停留在术语对应。可计算主体不是被模型完整认识的个人,而是在不同制度场景中被拆分为评分、概率、类别与行为轨迹的对象。分类一旦进入福利、信贷、招聘或内容分发,就会反过来塑造机会结构和主体行动。algorithmic identity把身份从稳定属性转化为不断更新的概率归类。系统不必确认“真正的你”是谁,只需根据行为轨迹把个体暂时放入某些可操作类别,并据此决定广告、价格、风险或可见性。Cheney-Lippold因此揭示了身份政治的新层面:主体可能并不知道自己被划入何种类别,更难直接参与类别的定义与修订。这一动态分类还有一个重要后果:身份不再只在登记时被确认,而可以在每次点击、移动和消费后被重新推断。主体面对的并非一个固定档案,而是一系列不断变动、彼此可能矛盾的概率身份。申诉因此格外困难,因为受影响者甚至未必知道是哪一种推断在何时触发了差别待遇。算法身份的政治困难由此不仅是“被误判”,还包括主体很难知道自己正被何种类别描述,以及这些类别如何跨场景流动。】
- Bucher, Taina. If...Then: Algorithmic Power and Politics. Oxford University Press, 2018. (Read: Chapter 2, "The Multiplicity of Algorithms," pp. 19–40; Chapter 3, "Neither Black nor Box," pp. 41–65).
【按:Bucher的重要贡献是拒绝把算法权力想象成某个神秘、统一、不可见的“黑箱”。算法权力分布在代码、数据、平台目标、用户行为、工程实践和机构制度之间。它有助于把Rouvroy的宏观治理概念落实到具体系统:我们面对的不是“算法”这个单一行动者,而是一组不同的软件、制度和实践。由此,所谓“算法透明”也需要被重新定义:透明代码并不自动意味着透明治理。预测系统经常只需构造足够可操作的proxy,而不需要建立完整身份理论;这恰恰是algorithmic identity与传统身份政治之间既相关又不可等同之处。Bucher If...Then: Algorithmic Power and Politics由此为“量化、数据生产与算法系统之间的制度链条”提供了一种有限而有效的解释尺度:它能够指出机制,但并不替代对组织、法律与物质条件的进一步分析。Bucher拒绝把算法想象成单一的黑箱主体,而强调算法由代码、数据、商业目标、用户实践与组织制度共同构成。所谓algorithmic power因此是一种关系性权力:它既存在于排序逻辑,也存在于工程团队的选择、平台指标和用户对系统的适应之中。透明性若只被理解为公开源代码,就无法覆盖这一复杂的社会技术装配。Bucher对algorithmic imaginaries的关注还说明,算法权力不仅存在于系统实际做什么,也存在于人们相信系统能做什么。用户为了迎合想象中的推荐规则调整表达,机构也可能因“算法更客观”的信念扩大自动化部署。算法的社会效力因此部分产生于围绕它形成的预期与不确定性。】
- Amoore, Louise. Cloud Ethics: Algorithms and the Attributes of Ourselves and Others. Duke University Press, 2020. (Read: Chapter 1, pp. 1–24).
【按:Amoore关注概率判断如何在边境、安全和风险治理中取得行动效力。重要问题不是算法能否“准确预测”,而是概率输出何时被机构转化成了现实的决策,以及这种转化如何改变责任与不确定性的政治。透明也不是单一解决方案。本节把可解释性放进更大的due process与institutional accountability中理解。若把Amoore Cloud Ethics: Algorithms and the Attributes of Ourselves and Others置于本节的整体脉络中,“透明、解释、申诉与责任如何在自动化决策中重新分配”呈现出更具体的结构。可计算主体不是被模型完整认识的个人,而是在不同制度场景中被拆分为评分、概率、类别与行为轨迹的对象。分类一旦进入福利、信贷、招聘或内容分发,就会反过来塑造机会结构和主体行动。Amoore把算法伦理的焦点从“模型是否正确”转向概率判断如何被付诸行动。边境、安全与风险系统往往在不确定条件下运行,关键不是消除不确定性,而是机构如何选择阈值、赋予某些属性更高权重,并让局部相关性获得现实后果。由此,责任问题不能被转移给模型:不确定性的制度化分配本身就是政治决定。Amoore由此把伦理问题推进到“属性化”本身:系统从局部痕迹推断一个人或事件的潜在性质,并据此提前分配注意、怀疑与资源。概率不是中性的等待状态,而可能成为行动理由。风险治理的政治性正在于谁被要求承受不确定性的代价,以及哪些错误被制度认为可以接受。】
推荐文献:
- Gillespie, Tarleton. "The Relevance of Algorithms" In Media Technologies, 167–194. MIT Press, 2014.
【按:Gillespie讨论“相关性”本身如何成为平台最重要的公共决策之一。搜索、推荐、排序并非单纯计算信息,而是在决定什么值得被看见、什么值得被认为重要。它为algorithmic governmentality增加了媒介与公共文化维度。理论上的相似并不自动意味着制度上的同一。预测准确、统计相关和行政正当性属于不同层次;一个风险分数只有在阈值、人工复核、申诉和执行程序中获得后果,才真正成为治理。“相关性”因此具有规范性而非纯技术性质。平台所称的relevance通常把点击概率、停留时间、商业目标与内容价值压缩进同一排序过程;算法并不直接裁决真理,却能通过分配注意力改变公共议题的出现条件。Gillespie由此把算法研究从模型内部推向平台制度:真正需要解释的是相关性标准如何被设定、调整并获得正当性。推荐系统的公共性还体现在它改变了“共同世界”的可见结构。不同用户接受高度个性化的排序后,公共议题不再拥有完全相同的入口,平台同时又保留调整推荐目标和内容边界的权力。由此,对算法相关性的研究必须连接媒介制度与商业模型,而不只是测量排序是否准确。平台排序于是既是信息处理,也是注意力配置;当相关性成为商业目标的函数时,媒介权力就进入算法设计的核心。Gillespie进一步提醒,算法排序的规范性常以“用户需求”或“相关性”名义出现,从而把价值判断技术化。什么值得推荐、什么构成垃圾、哪些内容应当被降权,都需要预设某种公共秩序。算法研究若忽略这些分类背后的制度目标,就会把平台最重要的政治判断误写成纯粹工程优化。】
- Eubanks, Virginia. Automating Inequality. St. Martin's Press, 2018. (Read: Introduction, pp. 1–14; Chapter 1, pp. 17–36).
【按:本书把抽象的算法治理落到福利、贫困与行政系统。核心不是证明算法“有偏见”,而是观察社会资源分配如何通过技术系统被重新组织,以及历史不平等如何进入风险分类。它能够使分析从“平台体验”转向国家行政与社会政策。Eubanks Automating Inequality所提供的并不是对“概率输出如何被组织转译为具有现实后果的判断”的终局命名,而是一种可被竞争性解释检验的理论切口,这种开放性本身构成其研究价值。Eubanks Automating Inequality还表明,自动化治理首先落在最缺乏退出能力的人群身上。技术系统将资源紧张、福利审查和风险管理重新包装为效率问题时,贫困本身容易被转换为可疑性与可计算风险。Eubanks的案例表明,自动化福利系统并非简单地把人工判断换成机器判断,而是在资源紧缩、行政改革和贫困治理的既有框架中重新组织筛选。技术可以加快审核,却也可能把历史上针对穷人的怀疑与惩戒固定为数据规则。算法公平因此无法脱离福利国家的制度目标:一个准确执行苛刻政策的系统,仍可能制造深刻的不正义。Eubanks的经验材料还揭示技术改革常见的政治不对称:行政机构以降低成本和防止欺诈为目标部署系统,而被审核者承担数据错误、复杂流程和自动拒绝的时间成本。技术效率若只从机构一侧计算,就会把弱势群体承担的额外劳动排除在绩效之外。自动化治理的评价尺度本身因此需要政治化。这些案例还显示,贫困治理中的自动化并不只产生个别错误,而可能改变国家与福利申请者之间的关系。面对面解释和个案裁量被数据库验证与风险评分替代后,申请者需要承担更多证明自身资格的责任。程序数字化由此可能把行政负担从机构内部转移给资源最少的人群。】
- Pasquale, Frank. The Black Box Society. Harvard University Press, 2015. (Read: Introduction; Chapter 2, "The Digital Poorhouse," pp. 35–58).
【按:Pasquale的重点在于自动化决策中的信息不对称和问责问题。由此可以用它检验“开源代码”是不是充分答案:一个系统即使开放代码,也可能因为数据、模型复杂度、商业机密或组织权限而继续不可问责。“量化、数据生产与算法系统之间的制度链条”通过Pasquale The Black Box Society进入了组织与制度层面。 在Pasquale The Black Box Society的尺度上,“量化、数据生产与算法系统之间的制度链条”需要同时处理连续性与变形:旧制度可能借新技术获得新的操作方式,新技术也可能在既有制度中被重新吸收。Pasquale The Black Box Society还提示,“量化、数据生产与算法系统之间的制度链条”的分析必须保留对象差异与制度边界,不能让概念的修辞强度替代经验说明。Pasquale The Black Box Society把不可见性从技术黑箱扩展到制度黑箱:商业机密、数据控制、专业壁垒和组织权限共同决定谁能解释系统。因而,可解释性只有与审计、申诉和责任归属结合,才可能转化为真正的问责。Pasquale所说的black box同时包含技术复杂性、商业秘密和制度性信息不对称。金融评分、搜索排序或平台决策的不可见,并不都能通过“解释模型”解决;受影响者往往还缺乏数据访问、申诉渠道和谈判能力。问责因此必须从信息披露扩展到组织责任,明确谁有权作出决定、谁必须说明理由以及谁能够纠正错误。】
- Harcourt, Bernard E.. Exposed: Desire and Disobedience in the Digital Age. Harvard University Press, 2015. (Read: Chapter 1, "The Expository Society," pp. 1–28).
【按:Harcourt将暴露、欲望和监控放在一起讨论。这里不应把主体展示行为简单解释成“自愿被监控”,而要分析平台设计、社会规范、可见性的欲望和商业数据提取如何相互强化。透明也不是单一解决方案。Harcourt Exposed: Desire and Disobedience in the Digital Age所揭示的“透明、解释、申诉与责任如何在自动化决策中重新分配”并不止于文本内部的概念争论。Harcourt Exposed: Desire and Disobedience in the Digital Age使“透明、解释、申诉与责任如何在自动化决策中重新分配”能够被限定在特定对象与因果链条之中;超出这一范围的类比,需要新的证据,而不能依靠术语本身完成。Harcourt所描述的“暴露”还揭示了监控研究中的一个重要转向:主体并非始终作为被动对象存在,而是在社交、消费和自我展示中主动生产可见性。问题不在于这种主动性是否取消了强制,而在于欲望如何被商业架构捕获,并在不对称的数据访问权中转化为预测和治理资源。Harcourt的论述使“暴露”不再只是隐私丧失,而成为一种社会关系。主体通过分享、搜索、定位和消费不断生产自身,而平台将这些表达重新转化为可计算资源。欲望与监控因此不是相反力量:正因为数字环境满足了被看见、连接和表达的需要,数据提取才可能以参与和便利的形式深入日常生活。暴露社会的悖论正在于,主体获得了表达与连接的自由,同时把越来越多的生活痕迹交给不可对等观察的平台。自由与监控并非简单此消彼长,而可能在同一互动架构中共同增长。】
- Beer, David. The Data Gaze: Capitalism, Power and Perception. Sage, 2018. (Read: Introduction and Chapters 1–2).
【按:Beer从data gaze的概念讨论数据化如何改变“看”和“被看”的关系。它适合连接算法分类与社会感知:当生活越来越被数据化之后,不仅机构在“看”我们,人们也越来越通过指标来理解自己。在Beer The Data Gaze: Capitalism, Power and Perception的分析尺度上,“分类、预测、评分与排序如何构成可计算主体”首先表现为一组具体关系。所谓算法偏见不能停留在价值判断。训练数据、标签体系、目标变量、评价指标、部署环境与既有制度不平等分别可能制造不同问题,只有逐层追踪才能判断偏差究竟在哪里生成。data gaze并不只是机构对人的凝视,也是一套让主体通过指标重新认识自身的感知机制。步数、信用分、热度、排名和绩效数据把生活转译成可比较的量值,并反过来塑造行动目标。Beer因此使算法治理与量化主体化联系起来:数据既描述现实,也参与规定何种现实值得追求。指标化自我并不完全来自强制。人们会主动使用排行榜、健康数据和社交反馈组织生活,但这种自愿使用发生在平台预设的可测量维度之内。主体性因此既包含自我量化的欲望,也受到指标架构的引导。Beer的分析使权力不再只是外部凝视,而进入日常自我评价的技术形式。这种自我量化进一步说明,数据权力常通过主体的主动参与实现,而不是单纯依靠外部监视或强制采集。指标的权力由此既来自测量,也来自人们围绕指标重新安排自身行动。也正是在这种自我调整中,数据凝视进入主体形成。】
- Yeung, Karen. "Algorithmic Regulation: A Critical Interrogation" Regulation & Governance 12, no. 4 (2018): 505–523.
【按:文章提供一个制度与法治角度的校正。algorithmic regulation并不只是技术效率提高,而涉及规则如何被机器执行、监测与动态调整。它有助于在Rouvroy与Lessig之间建立更清晰的理论桥梁。Yeung "Algorithmic Regulation: A Critical Interrogation"与Beer The Data Gaze: Capitalism, Power and Perception和Seaver "Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems"把“概率输出如何被组织转译为具有现实后果的判断”置于不同的知识传统中。Yeung "Algorithmic Regulation: A Critical Interrogation"让“概率输出如何被组织转译为具有现实后果的判断”摆脱了过于抽象的时代判断。所谓算法偏见不能停留在价值判断。Yeung "Algorithmic Regulation: A Critical Interrogation"对“概率输出如何被组织转译为具有现实后果的判断”的贡献,正在于让理论保持可反驳性。Yeung的制度视角进一步说明,算法规制的关键不只是自动化程度,而是反馈回路如何改变规则执行。持续监测能够使治理从事后处罚转向实时调整,但动态性也可能压缩程序性审查与稳定预期。因而,效率、适应性与法治所要求的可预见性、可解释性之间存在真实的制度张力。】
- Seaver, Nick. "Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems" Big Data & Society 4, no. 2 (2017). (Read: entire article).
【按:Seaver反对把算法只当作可以从代码中直接读出的技术对象,主张把算法系统视为由工程师的分类、组织文化、用户模型、数据实践与维护工作共同构成的文化对象。对本节而言,这一点非常关键:如果算法治理术涉及预测和分类,那么研究者不能只问模型数学上如何运行,还要问“相关性”“用户”“异常”“成功”这些类别在组织内部是怎样被定义出来的。围绕Seaver "Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems"展开时,“量化、数据生产与算法系统之间的制度链条”可以被重新拆分为若干相互关联的环节。所谓算法偏见不能停留在价值判断。 从Seaver "Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems"来看,“量化、数据生产与算法系统之间的制度链条”不能脱离制度授权与资源分配单独成立;技术形式只有与组织权力结合,才会产生稳定的社会后果。Seaver的文化人类学路径把算法从“独立运转的代码”重新放回组织生活。工程师如何理解用户、什么指标被视为成功、团队如何处理异常情况,这些文化实践都参与构成算法系统。算法民族志的意义由此不在于进入秘密代码仓库,而在于观察技术对象如何在日常工作、叙事和制度惯例中被持续制作出来。】
- Fourcade, Marion, and Kieran Healy. "Classification Situations: Life-Chances in the Neoliberal Era" Accounting, Organizations and Society 38, no. 8 (2013): 559–572. (Read: entire article).
【按:Fourcade与Healy把评分、排名和分类放回社会分层研究,强调现代分类系统并非中性描述,而会改变个体能够获得的机会、价格、资源与制度待遇。"classification situation" 的概念尤其适合与algorithmic identity对读:计算主体不仅是数据上的表征,它还处于一组由分类结果塑造的现实机会结构中。Fourcade "Classification Situations: Life-Chances in the Neoliberal Era"与Seaver "Algorithms as Culture: Some Tactics for the Ethnography of Algorithmic Systems"和Espeland "A Sociology of Quantification"共同使“透明、解释、申诉与责任如何在自动化决策中重新分配”从宏观判断下沉到机制层。透明也不是单一解决方案。Fourcade "Classification Situations: Life-Chances in the Neoliberal Era"对“透明、解释、申诉与责任如何在自动化决策中重新分配”的贡献,在于把宏观判断压回可识别的制度机制。所谓算法偏见不能停留在价值判断。Fourcade "Classification Situations: Life-Chances in the Neoliberal Era"把“透明、解释、申诉与责任如何在自动化决策中重新分配”重新置于历史条件之中。相似的数字表象并不必然来自相同机制,而概念的有效性也因此依赖对差异的保存。Fourcade与Healy的classification situations把分类与life-chances直接联系起来。信用、保险、教育与消费场景中的分类不仅描述社会差异,还会决定利率、机会、待遇和可见性,由此产生反馈效应。数字化强化的并非“分类”这一行为本身,而是分类能够以更细粒度、更高频率和更低可见性嵌入日常制度。】
- Espeland, Wendy Nelson, and Mitchell L. Stevens. "A Sociology of Quantification" European Journal of Sociology 49, no. 3 (2008): 401–436. (Read: entire article).
【按:Espeland与Stevens为量化研究提供方法论基线:数字、指标和排名既能够压缩复杂性,也会通过commensuration把原本不可直接比较的对象置于同一尺度。把这篇文章放进算法治理,不是为了说所有算法都只是“打分”,而是由此可以看到预测模型得以进入制度决策之前,往往先发生了对象可量化、指标可比较、结果可排序的社会工程。“分类、预测、评分与排序如何构成可计算主体”从Espeland "A Sociology of Quantification"出发,可以获得更清楚的因果层次。透明性并不是自动化治理的单一解法。commensuration一旦完成,政治争论往往会被重写成指标设计问题。大学排名、信用评分或风险模型之所以具有权力,并非数字本身具有强制力,而是组织开始依据这些数字配置机会与资源。量化社会学因此提供了一条重要路径:追踪分类标准进入制度之后如何改变行为,同时反过来稳定分类本身。Espeland与Stevens还强调reactivity:被排名和被测量者会调整行为,以适应评价体系。指标一旦成为目标,就可能改变原本要测量的对象。这一机制对于算法治理尤其关键,因为预测模型的部署会改变人的策略,进而改变未来数据。量化由此不是被动记录,而是一种能够重塑制度场景的社会过程。当指标成为行动目标之后,量化体系会改变被量化对象本身,这也是排名与评分能够产生制度反馈的根本原因。commensuration与reactivity合在一起,构成量化制度持续自我强化的基本机制。】
- O'Neil, Cathy. Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. Crown, 2016. (Read: Introduction; selected case studies on education, employment, credit, and policing).
【按:O'Neil用一组制度案例说明大规模评分模型如何在不透明、反馈闭环和高影响场景中放大不平等。它不是算法理论的终点,却非常适合作为概念压力测试:只有把模型放回招生、雇佣、信用或警务等制度流程,我们才看得见错误如何被分配、申诉渠道是否存在,以及量化判断如何获得事实上的强制力。透明性并不是自动化治理的单一解法。O'Neil所说的WMD强调模型的规模、不透明与反馈性如何共同放大伤害。其通俗化概念需要与更严格的算法研究配合,但它抓住了制度部署的核心:模型一旦进入教育、就业、司法等高风险领域,错误不再只是统计偏差,而会改变人的机会,并把结果重新写回未来数据。模型评估因此必须包括社会反馈,而不只是离线准确率。这一反馈性使WMD概念可以与社会科学中的performativity连接起来。模型产生的判断会改变教育资源、信贷机会或治安关注,随后这些结果又可能被当作下一轮训练的“现实数据”。如果不打断这种循环,历史不平等就会以模型表现的形式获得新的经验合法性。因此,评估模型不能只问预测是否准确,还要考察预测被部署以后是否会制造新的数据,并把既有差异不断自我证成。WMD框架还提出一个问责难题:当模型被视为客观工具时,机构容易把不受欢迎的决定解释为“数据结果”,从而模糊政策选择。哪些变量被选中、何种误差可接受、是否允许人工覆写都仍是制度决策。自动化并未消除判断,只是把一部分判断提前写入模型和流程。制度因此不能把模型责任理解为开发者一次性完成的技术责任。部署、更新、阈值调整和人工覆写都会持续改变系统后果,问责应覆盖整个生命周期。】
- Kitchin, Rob. The Data Revolution: Big Data, Open Data, Data Infrastructures and Their Consequences. Sage, 2014. (Read: Introduction; chapters on data infrastructures and data assemblages).
【按:Kitchin将big data从“更多数据”的宣传语还原为一套data assemblage:采集装置、标准、数据库、分析工具、组织目标与制度实践共同决定数据如何被生产。它能给本节补上Rouvroy理论中容易被抽象化的基础设施环节:所谓数据行为主义并不是面对自然存在的数据,而是面对经过选择、清洗、格式化和制度化的数据对象。借助Kitchin The Data Revolution: Big Data, Open Data, Data Infrastructures and Their Consequences,“量化、数据生产与算法系统之间的制度链条”能够从修辞性命题转化为制度分析。透明性并不是自动化治理的单一解法。Kitchin对data infrastructure的分析使“大数据”不再像自然资源那样被想象为自行存在。数据总是经过采集协议、数据库结构、清洗标准、元数据和机构目标才成为可用对象。所谓data-driven决策因此从一开始就包含选择与排除;理解模型输出,必须向前追溯数据如何被生产、分类与维护。Kitchin同时区分数据开放与数据权力。数据可以在技术上开放,却仍因格式、计算资源、专业能力和制度权限而只对少数主体真正可用。由此,开放数据政治不能停留在发布文件,还要处理谁有能力分析、关联并将数据转化为决策资源。数据基础设施本身就是不平等能力的分配机制。data infrastructure的治理还包括标准兼容和长期维护。数据能否被不同机构理解和复用,往往依赖元数据、格式与持续更新;这些低可见度工作决定了所谓“数据资产”能否成立。】
- Rieder, Bernhard. Engines of Order: A Mechanology of Algorithmic Techniques. Amsterdam University Press, 2020. (Read: Introduction; selected chapters on ranking, recommendation, and algorithmic techniques).
【按:Rieder的mechanology路径把注意力从“算法”这个过度宽泛的名词转向ranking、recommendation、classification等具体技术机制及其历史。它适合放在Bucher之后,用来训练一种更细粒度的批判:不同算法技术具有不同的输入、目标函数、评价方式和组织用途,不能被统一写成一个无所不在的“算法权力”。透明也不是单一解决方案。透明性并不是自动化治理的单一解法。Rieder的细粒度方法尤其适合抵抗“算法”一词的泛化。ranking、classification与recommendation拥有不同的问题设定、技术传统和评价指标,它们在组织中的权力效果也不相同。只有把模型类型与具体制度用途对应起来,算法批判才可能从宏大诊断进入可验证的机制分析。mechanology还要求把算法技术放回工具谱系,而不是只追踪最新模型。排序、匹配、聚类和推荐都有较长历史,也形成不同的工程常识。理解技术谱系有助于分辨哪些治理效果来自机器学习的新能力,哪些其实来自更早的信息检索、统计分类或数据库管理。这种技术分解也为治理提供更准确的对象:不同排序、匹配和分类机制应接受不同的透明度、审计与责任要求。Rieder所强调的algorithmic techniques还可以改变审计方式。与其要求一个抽象“算法”整体透明,不如分别分析输入数据、排序函数、反馈信号和组织目标。这样的分解能够让责任对应具体环节,也使批判不必在“黑箱不可知”与“公开代码即可解决”之间二选一。】
第二组:提取、殖民性与基础设施(第3–4周)
第三节:监控、提取与数据殖民性
本节把算法治理进一步放入数据生产、资源提取与历史不平等之中。Zuboff的surveillance capitalism强调行为数据如何被转化为预测性资产,Couldry与Mejias的data colonialism则把数据关系放入更长的殖民占有史中,Browne从奴隶制、种族化识别与监控技术的历史连续性出发,揭示分类系统如何嵌入差异化的可见性与暴露。它们彼此并不等价:资本积累、殖民占有、种族化治理与生命政治拥有不同的历史机制。把这些概念并置的意义,正在于检验它们解释数据经济时各自能够说明什么,以及何时会因类比过度而遮蔽差异。
数据提取也不是一个纯粹发生在“用户同意”之后的交换过程。平台依赖身份系统、默认设置、追踪基础设施、广告市场、数据经纪与网络效应,把日常行动持续转换为可存储、关联和预测的记录。不同群体在被记录、被识别和被排除时承担的风险并不对称;生物识别、信用评分、广告定向和公共行政系统中的分类都会把既有社会差异带入技术过程。Noble、Benjamin、Magnet、Dencik等文献分别从信息检索、种族技术、身份识别与数据正义切入,使“数据偏见”不再只是模型误差,而成为社会分类、资源分配与权力关系的组成部分。
Mbembe在此并不被用于制造“账号封禁即数字死亡”之类的简单类比,而是提供理解殖民性、敌意、边界与差异化暴露的政治背景。数据殖民性的分析只有在强制依赖、占有关系、知识权威、法律不对称和退出成本能够被具体说明时才成立。由此,本节关注的并不是为数字社会寻找最有震撼力的历史名称,而是厘清数据化如何把观察、提取、分类和治理连接起来,以及这些机制怎样在不同制度场景中继承或重组更早的权力结构。
必读文献:
- Zuboff, Shoshana. The Age of Surveillance Capitalism. PublicAffairs, 2019. (Read: Chapter 3, "The Discovery of Behavioral Surplus," pp. 63–96).
【按:Zuboff的"behavioral surplus"概念把数据治理与资本积累联系起来。重点阅读她如何区分用户行为、数据提取、预测产品和市场化过程。因此必须检验:这一模型是否适用于整个数字经济,还是主要解释Google/广告平台式的数据提取机制?阅读Chapter 3时必须把behavioral surplus的论证链完整拆开:用户为了获得服务而留下行为痕迹,企业把超出即时服务改进所需的数据转化为额外输入,再通过计算过程生产面向未来行为的预测产品。真正需要检验的是这一链条在广告型平台之外能否推广,以及surveillance capitalism是否把国家、金融、物流和非广告平台之间的差异压得过平。因此不宜把Zuboff当作数字经济总理论,而把她作为一种强模型:它在哪些产业解释力最大,在哪些场景必须由platform capitalism、data colonialism或行政自动化补充?迁移成立必须说明强制、所有权、空间控制、知识等级或资源流动等具体机制,同时承认经典殖民统治与平台数据关系之间的重要差异。Zuboff的behavioral surplus概念把用户活动、数据提取和预测市场纳入同一资本积累过程。其理论野心远超在线广告,因此也引发“监控资本主义是否被过度一般化”的争论。最具生产力的读法不是把整个数字经济都装入这一概念,而是具体区分哪些业务真正依赖行为预测,哪些平台收益更多来自基础设施租金、交易佣金或垄断定价。】
- Couldry, Nick, and Ulises A. Mejias. The Costs of Connection. Stanford University Press, 2019. (Read: Chapter 2, "Cloud Empire," pp. 37–68; Chapter 3, "The Coloniality of Data Relations"; Chapter 5, pp. 167–208).
【按:Couldry与Mejias用data colonialism描述社会生活持续被转换为可提取数据的过程。"Cloud Empire"尤其适合本节,因为它把云、平台、数据提取和社会关系连在一起。这里最值得讨论的是:数据殖民是否能够成为理解云基础设施的历史框架,以及这种框架在哪些方面可能与经典殖民主义类比过度。作者把data relations放进更长的殖民资源占有史,强调社会生活本身被持续转译为可以商业利用的数据流,而"cloud empire"则指向支持这种关系的基础设施与企业权力。这里不能只抓住"colonialism"这个强词;必须细读他们如何从appropriation of social life、normalization of data extraction与新的social order建立类比。与Browne对读时还要追问:如果殖民性从来与种族化、领土占有和暴力治理纠缠,仅从数据提取出发的普遍化殖民模型会不会忽视历史差异?监控并非只有一个中心主体。国家安全、广告产业、雇主管理、金融风控和社交平台可能共享数据与技术,却具有不同法律授权和利益结构;研究时必须避免用一个surveillance capitalism概念抹平所有制度差异。Couldry与Mejias把数据关系描述为新型殖民式占有,意在强调日常生活本身被持续转化为可提取资源。这个概念的力量在于连接资本积累与主体关系,而风险则在于可能抹平历史殖民主义的暴力、领土与种族差异。比较的有效性因此取决于是否能指出占有机制、资源流向和依赖结构,而非只借用“殖民”一词的道德强度。】
- Browne, Simone. Dark Matters: On the Surveillance of Blackness. Duke University Press, 2015. (Read: Introduction; Chapters 1–2).
【按:Browne把surveillance studies的历史纵深推进到奴隶制、种族化识别与对黑人身体的持续管理,反对把数字监控写成互联网时代突然出现的全新问题。她提出的重点不是“所有监控都一样”,而是现代识别、边界管理和可见性制度如何继承并重组更早的种族化治理技术。放在Zuboff与数据殖民主义之间,由此需要检验:以商业数据提取为中心的解释是否足以说明监控权力的历史差异。“被看见”与“被数据化”也不是同一回事。许多治理系统的力量恰恰来自不可见的后端分类和推断,因此批判不能只围绕隐私暴露,还要研究分类机会、风险分配和知识权威。“殖民类比的分析收益及其不可越过的历史边界”在Browne Dark Matters: On the Surveillance of Blackness这里因此不只是概念问题。数据提取的政治性来自不对称关系:一部分主体掌握采集、关联、推断与再利用的能力,另一部分主体主要以被记录、被分类和被预测的形式出现。隐私只是其中一层,产权、知识权威与退出条件同样重要。Browne还迫使监控史重新处理种族与身体的历史连续性。生物识别、边境检查和数据库分类看似依赖新的技术形式,但“谁需要被识别、谁被视为风险、谁必须证明身份”从来不是技术中性的安排。数字监控的历史因此不能只从互联网企业开始,而要追溯可见性与种族化治理之间更长的制度谱系。Browne提出的dark sousveillance同时保留了抵抗维度:被监控者并非只有被动接受,也会发展反识别、反记录和重新编码可见性的实践。监控技术的历史由此不仅是权力不断扩张的历史,也是被分类主体持续扰乱分类制度的历史,这一点对数字身份与生物识别研究尤为重要。】
- Mbembe, Achille. Necropolitics. Translated by Steven Corcoran. Duke University Press, 2019. Originally published in French as Politiques de l'inimitié in 2016. (Read: Introduction; "Necropolitics"; selected chapters on enmity, borders, and exposure).
【按:Mbembe的necropolitics讨论的是主权、战争、殖民性以及生命被差异化暴露于死亡和伤害的政治条件。把它用于数智化问题时,绝不能把账号封禁、降权或平台排除直接命名为“数字死亡”;这种类比既削弱了概念的历史强度,也会遮蔽真实的暴力关系。更严格的迁移方式,是借它追问数据化边境、自动风险判断、军事化监控与资源分配如何参与制造不同程度的暴露、可伤害性和行动自由。进一步的问题是谁决定数据标准、服务器、模型和研究议程,以及被研究社群是否拥有退出、拒绝、修正和共同治理的能力。若把Mbembe Necropolitics置于本节的整体脉络中,“数据正义、知识权威与差异化暴露如何进入同一研究框架”呈现出更具体的结构。数据提取的政治性来自不对称关系:一部分主体掌握采集、关联、推断与再利用的能力,另一部分主体主要以被记录、被分类和被预测的形式出现。隐私只是其中一层,产权、知识权威与退出条件同样重要。Mbembe的necropolitics讨论的是主权如何通过暴露于死亡、战争化空间和差异化生存条件来组织生命,其历史背景远超数字治理。将这一理论放入数据殖民性讨论,不应把封号、降权等平台惩罚夸大为“数字死亡”,而应观察数字基础设施何时参与边境、警务、战争或资源分配,并与现实的生命政治和死亡政治发生连接。因此,Mbembe在此处更适合发挥尺度校正作用。数据化治理只有在与边境封锁、军事打击、警务风险或基本资源的差异化分配结合时,才可能进入necropolitical问题域。把概念保留在这种高强度的生命政治语境中,反而能够防止数字批判借用“死亡”修辞后失去历史重量。】
推荐文献:
- Benjamin, Ruha. Race After Technology: Abolitionist Tools for the New Jim Code. Polity, 2019. (Read: Introduction and Chapter 1).
【按:将算法与种族化权力联系起来。它可以补充Noble的信息检索视角,并提醒我们一般智力并不是无差别的“人类共同资源”:谁的知识、语言和身体经验更容易成为训练数据,本身就是政治问题。理论上的相似并不自动意味着制度上的同一。因此,Benjamin Race After Technology: Abolitionist Tools for the New Jim Code对“数据提取、监控、种族化分类与殖民性之间的历史关系”的说明只有在相应制度机制能够被经验材料识别时才成立。"New Jim Code"的分析把技术中立性本身变成批判对象:不平等并不一定以显性的歧视规则出现,它也可以通过看似客观的效率、优化和风险管理延续。Benjamin因此要求把算法系统放入已有的住房、教育、治安和就业结构中考察;技术的社会后果取决于它进入了怎样的制度历史。abolitionist tools的方向也不同于单纯的fairness修补。Benjamin关心的不是让既有系统对不同群体更平均地运作,而是某些系统是否本来就服务于不平等的分类与惩罚。由此,技术治理既包括“如何改进模型”,也包括更根本的制度选择:哪些判断根本不应交给自动化系统。Benjamin的abolitionist取向尤其适合把技术改革与制度改革区分开来。一个预测系统即使实现群体间统计公平,如果其用途本身是扩大惩罚、排斥或过度警务,公平指标也无法回答更根本的正当性问题。技术伦理因此需要保留“是否应当存在这个系统”这一制度性否定。】
- Noble, Safiya Umoja. Algorithms of Oppression. NYU Press, 2018. (Read: Introduction, pp. 1–14).
【按:Noble说明算法排序不是在空白空间中运行,而是会放大既有社会偏见。它与Crawford的数据/劳动视角不同,更关注信息环境中的分类和表示政治。监控并非只有一个中心主体。Noble Algorithms of Oppression所提供的并不是对“商业数据化与国家治理并存而非互相替代的机制”的终局命名,而是一种可被竞争性解释检验的理论切口,这种开放性本身构成其研究价值。Noble Algorithms of Oppression的关键并不只是证明搜索结果存在偏见,而是显示商业排序、广告逻辑与社会表征如何共同塑造“相关性”。搜索系统由此成为文化分类与市场激励交叠的场所,种族化表征并非单纯的数据误差。Noble以搜索引擎为例说明,商业排序能够把既有的种族与性别偏见重新包装为“最相关”的信息结果。搜索并非公共知识的中性镜面,其分类方式受广告市场、内容生态和平台设计影响。representation因此不只是文化层面的问题;当搜索结果成为理解群体与社会事实的入口时,排序权同时具有认识论权力。Noble的案例还说明信息基础设施具有累积效应。搜索结果不仅回应既有网络内容,也会影响后续点击、内容生产和公共认知,进而强化某些关联。平台若以历史行为作为相关性信号,社会偏见就可能通过反馈循环被技术化。算法排序因此同时处理信息与生产未来的信息环境。搜索结果的政治还涉及知识权威。平台以算法方式组织海量信息后,“最前面出现什么”很容易被用户理解为世界本身的自然排序。Noble揭示的正是这种界面自然化:商业和社会偏见经过技术排序后获得了客观性的外观,因而更难被识别为制度选择。】
- Lyon, David. Surveillance Studies: An Overview. Polity, 2007. (Read: Introduction; chapters on surveillance as social sorting).
【按:Lyon为surveillance studies提供一套比“被看见”更宽的概念框架,尤其强调surveillance作为信息收集、分类与social sorting的制度过程。它适合作为本节的基础读物,因为平台、国家与商业数据库往往并不需要持续观看个体,而是通过身份、风险、消费与资格类别决定不同人得到何种机会。“被看见”与“被数据化”也不是同一回事。“殖民类比的分析收益及其不可越过的历史边界”通过Lyon Surveillance Studies: An Overview进入了组织与制度层面。 在Lyon Surveillance Studies: An Overview的尺度上,“殖民类比的分析收益及其不可越过的历史边界”需要同时处理连续性与变形:旧制度可能借新技术获得新的操作方式,新技术也可能在既有制度中被重新吸收。Lyon Surveillance Studies: An Overview提供的价值在于把监控理解为社会组织关系,而不是单一设备。识别、分类、可见性与机构目的共同决定监控的形式,这也为商业平台、国家行政与安全体系之间的比较建立了基础。Lyon把surveillance理解为社会分类与组织实践,而非单纯的“窥视”。现代监控依赖身份记录、数据库和风险分类,把人口区分为不同待遇与流动权限。数字技术扩展了这一能力,却延续了监控与行政组织长期结合的特征。把监控研究放入这一社会学传统,有助于避免只围绕摄像头、平台或隐私展开技术中心主义叙事。】
- Andrejevic, Mark. Automated Media. Routledge, 2020. (Read: Introduction; selected chapters on automation, surveillance, and decision systems).
【按:Andrejevic讨论自动化媒介如何把感知、分析和反应越来越交给数据密集型系统。其关键不是宣布“人被机器取代”,而是分析机构如何把监测规模扩大到超出人工处理范围,并以自动化相关性和预测来管理信息环境。与Zuboff对读时,可以更细致地区分资本积累、媒介自动化与治理能力三者之间的关系。Andrejevic Automated Media所揭示的“数据正义、知识权威与差异化暴露如何进入同一研究框架”并不止于文本内部的概念争论。“数据正义、知识权威与差异化暴露如何进入同一研究框架”因而仍需接受Andrejevic Automated Media之外的经验材料和竞争性解释的共同检验。automated media的核心问题由此不是“机器是否自主”,而是感知-分析-反应链条被压缩之后,哪些判断不再经过公开解释。大规模自动化可以显著扩大监测与响应能力,也可能让决策依据更加远离受影响者。Andrejevic所强调的正是这种规模政治:自动化改变了机构能够看见多少、处理多快以及在何种条件下行动。自动化媒介也改变了政治行动的时间尺度。系统能够在人工理解之前先行标记、阻断或推荐,机构因此可能越来越依赖机器先做反应、事后再解释。速度成为权力资源:能够实时处理海量信号的组织,比只能事后申诉的个体拥有结构性优势。自动化的政治必须同时分析这种时间不对称。速度优势一旦与机构强制力结合,就会产生新的程序问题:自动标记可以瞬间发生,解释和纠错却需要漫长时间。申诉成本的不对称因此成为自动化治理的重要评价指标。】
- Turow, Joseph. The Daily You: How the New Advertising Industry Is Defining Your Identity and Your Worth. Yale University Press, 2011. (Read: Introduction; selected chapters on behavioral targeting and audience classification).
【按:Turow从广告产业的组织变化出发,追踪受众如何被细分、估值并持续定向。它为surveillance capitalism提供重要的产业史背景:数据提取不是在真空中发明出来的,而与广告市场、受众测量、客户关系管理和差异化定价的既有实践连接。这里尤其应注意“身份”在这里首先是商业分类,而不必等同于主体自我认同。在Turow The Daily You: How the New Advertising Industry Is Defining Your Identity and Your Worth的分析尺度上,“数据提取、监控、种族化分类与殖民性之间的历史关系”首先表现为一组具体关系。广告产业的这一历史也说明,数据价值并非天然存在于个体行为之中,而是通过市场分类和可交易的受众想象被制造出来。“高价值用户”“易流失客户”或“高风险人群”都是组织为特定目标构造的类别。由此,个体化营销与社会分层不是两个独立过程,而可能在同一数据基础设施上彼此强化。Turow的产业史使“个性化”显露出另一面:市场越能细分用户,就越能实行差异化待遇。个性化推荐并不只是提高匹配效率,也可能把不同消费者分配到不同价格、优惠和信息环境中。由此,商业分类与机会分层之间的边界变得模糊,广告技术开始具有更广泛的社会分类效应。广告技术由此不仅分配商品信息,也分配被企业认为值得投入的注意和优惠,商业估值开始直接参与社会差异化。】
- van Dijck, José. "Datafication, Dataism and Dataveillance: Big Data between Scientific Paradigm and Ideology" Surveillance & Society 12, no. 2 (2014): 197–208. (Read: entire article).
【按:van Dijck区分datafication、dataism与dataveillance,有助于把“万物皆可数据化”既看作技术实践,也看作带有认识论承诺的文化意识形态。文章的价值在于指出:数据并非自动拥有真理地位,平台和研究机构对大规模行为数据的信赖需要被解释。它可与Rouvroy的相关性治理相互照亮,但在本节更侧重数据提取获得合法性的知识文化条件。van Dijck "Datafication, Dataism and Dataveillance: Big Data between Scientific Paradigm and Ideology"与Turow The Daily You: How the New Advertising Industry Is Defining Your Identity and Your Worth和D'Ignazio Data Feminism把“商业数据化与国家治理并存而非互相替代的机制”置于不同的知识传统中。监控并非只有一个中心主体。van Dijck "Datafication, Dataism and Dataveillance: Big Data between Scientific Paradigm and Ideology"让“商业数据化与国家治理并存而非互相替代的机制”摆脱了过于抽象的时代判断。dataism之所以重要,在于它把数据化从技术操作提升为一种认识论信念:只要记录足够全面、计算足够复杂,社会行为似乎就能够被直接把握。van Dijck对这一信念的批判为算法治理提供了必要的认识论层面——数据的规模不能替代概念选择,相关性也不能自行回答因果、意义与正当性问题。】
- D'Ignazio, Catherine, and Lauren F. Klein. Data Feminism. MIT Press, 2020. (Read: Introduction; Chapters 1–3).
【按:D'Ignazio与Klein把数据实践放进feminist intersectional analysis,强调power、context、embodiment与被遗漏经验。它不是一本泛泛的“AI公平”手册,而是要求研究者反思数据集是谁定义的、谁被记录、谁被遗漏、谁承担分类后果。放在Browne、Benjamin与Noble旁边,能够把批判从“系统有没有偏差”推进到数据生产本身的政治。“被看见”与“被数据化”也不是同一回事。围绕D'Ignazio Data Feminism展开时,“殖民类比的分析收益及其不可越过的历史边界”可以被重新拆分为若干相互关联的环节。 从D'Ignazio Data Feminism来看,“殖民类比的分析收益及其不可越过的历史边界”不能脱离制度授权与资源分配单独成立;技术形式只有与组织权力结合,才会产生稳定的社会后果。Data Feminism把权力分析直接嵌入数据实践,从提问、分类、可视化到解释都要求追踪谁被代表、谁被遗漏以及谁承担后果。它并不把“更多女性或少数群体数据”视为充分答案,而强调挑战既有权力结构与二元分类。数据正义由此不仅是修正偏差,更涉及重新设计知识生产的目标和参与关系。Data Feminism还特别强调context与pluralism。数据集脱离采集情境后容易把具体社会关系抽象成自然事实,而多元知识位置能够暴露这种抽象的代价。由此,数据实践的民主化不只是开放下载权限,还包括记录数据来源、保留不确定性并让受影响群体参与问题定义。】
- Birhane, Abeba. "Algorithmic Colonization of Africa" SCRIPTed 17, no. 2 (2020). (Read: entire article).
【按:Birhane用algorithmic colonization描述数字技术以普遍主义、发展主义和效率话语进入非洲社会时可能复制的不对称权力:技术、资本、数据和标准往往由外部主体掌握,而受影响社群被当作部署场景或数据来源。它与Couldry/Mejias的data colonialism形成直接对话,同时也提醒我们,“殖民”概念必须通过具体的所有权、知识权威和制度依赖来证明,而不能只作为修辞。Birhane "Algorithmic Colonization of Africa"与D'Ignazio Data Feminism和Ricaurte "Data Epistemologies, the Coloniality of Power, and Resistance"共同使“数据正义、知识权威与差异化暴露如何进入同一研究框架”从宏观判断下沉到机制层。Birhane "Algorithmic Colonization of Africa"对“数据正义、知识权威与差异化暴露如何进入同一研究框架”的贡献,在于把宏观判断压回可识别的制度机制。Birhane "Algorithmic Colonization of Africa"把“数据正义、知识权威与差异化暴露如何进入同一研究框架”重新置于历史条件之中。相似的数字表象并不必然来自相同机制,而概念的有效性也因此依赖对差异的保存。Birhane的论证把“殖民”从比喻重新压回制度关系:模型、云服务和数据标准由谁设计,知识需求由谁定义,技术收益又流向何处,构成判断殖民性的具体依据。这样处理后,algorithmic colonization不再指向一切跨国技术流动,而是指向持续的不对称依赖及其对本地知识、自治和公共能力的侵蚀。】
- Ricaurte, Paola. "Data Epistemologies, the Coloniality of Power, and Resistance" Television & New Media 20, no. 4 (2019). (Read: entire article).
【按:Ricaurte把数据化与coloniality of power联系起来,重点不只在资源提取,还在知识生产:什么可以被认识、怎样分类、谁有权把数据结果定义为社会现实。它能使“数据殖民主义”避免只停留在经济比喻,转而讨论epistemic authority。对这一讨论而言,这也是把拉丁美洲去殖民理论引入数据研究的重要节点。“数据提取、监控、种族化分类与殖民性之间的历史关系”从Ricaurte "Data Epistemologies, the Coloniality of Power, and Resistance"出发,可以获得更清楚的因果层次。Ricaurte进一步把数据问题放到知识生产的殖民性之中:并非所有世界都以同样方式被数据化,也并非所有知识形式都能无损地进入数据库和模型。数据基础设施在选择分类、变量与可测量对象时已经包含认识论政治。抵抗因此不仅涉及隐私或产权,也涉及谁有权规定什么能够被认作知识。这种认识论批判还涉及数据主权。若本地社群只能提供原始数据,而分类标准、模型设计和解释权集中在外部机构,知识生产就会复制中心-外围结构。去殖民化因而不能只增加“代表性数据”,还要重新分配研究议程、基础设施与解释结果的权力。数据去殖民化因而包含基础设施问题:如果计算、存储和模型能力始终依赖外部主体,知识自主仍然十分有限。这也使数据基础设施本身成为知识政治和去殖民实践的争夺对象。知识权与基础设施权在此无法截然分离。】
- Magnet, Shoshana Amielle. When Biometrics Fail: Gender, Race, and the Technology of Identity. Duke University Press, 2011. (Read: Introduction; selected chapters on fingerprinting, facial recognition, and biometric failure).
【按:Magnet通过生物识别技术说明所谓“客观身份”如何依赖数据库、传感器、阈值和既有社会分类。系统失败并非只是工程误差;失败对不同身体分布不均,并可能把种族、性别与国籍差异重新编码为识别问题。在这一位置,它为宏观的surveillance/data coloniality提供一个高度具体的技术对象。监控并非只有一个中心主体。生物识别技术尤其清楚地显示分类与身体之间的循环:系统首先依赖既有的人口类别训练和验证,随后又以“技术识别”的形式把这些类别反馈到边境、安全或身份制度中。失败率并非单纯的工程瑕疵;当识别系统具有强制后果时,误差会沿着既有社会差异被不均等地分配。Magnet特别说明“失败”并非系统之外的偶然事件。生物识别在被宣称为客观、自动和普遍时,其误识别会被制度视为个体异常,而不是分类体系的问题。技术失败由此具有政治分配:某些身体被迫承担更多验证、怀疑和证明自己的成本。识别准确率必须与这种不均等后果一起评估。生物识别的政治评价必须把错误率与错误后果结合起来,因为同一统计误差在不同身份制度中会产生完全不同的社会成本。因此,技术精度与制度正义从来不能被当作两个互不相关的评价维度。当生物识别被用于身份证、边境或福利系统时,失败不只是使用体验问题,而可能影响移动、领取资源和证明合法身份的能力。由此,技术性能评价必须与制度后果相连:同样的误识率,在娱乐应用与强制身份系统中具有完全不同的政治重量。】
- Mejias, Ulises A., and Nick Couldry. Data Grab: The New Colonialism of Big Tech and How to Fight Back. University of Chicago Press, 2024. (Read: Introduction; chapters on data territories, colonialism, and resistance).
【按:这本书延续并更新作者对data colonialism的论证,把日常生活被持续数据化与大型科技公司的基础设施扩张放在殖民关系的长历史中。与2019年The Costs of Connection相比,它更明确地面向当代Big Tech和抵抗策略,因此适合作为同一本节内的后续论争材料,而不是把两本书拆到不同本节重复承担同一功能。阅读重点应放在殖民类比的分析收益及其历史边界。“被看见”与“被数据化”也不是同一回事。借助Mejias Data Grab: The New Colonialism of Big Tech and How to Fight Back,“殖民类比的分析收益及其不可越过的历史边界”能够从修辞性命题转化为制度分析。Data Grab延续data colonialism框架,但更集中于Big Tech如何把连接、便利和基础设施转化为持续的数据占有。其政策意义在于把问题从个人隐私选择扩展到集体层面的数据关系:当退出大型平台意味着失去工作、社交或公共服务机会时,“同意”很难承担全部合法性基础。抵抗因此必须涉及制度性限制与公共替代。Mejias与Couldry进一步强调,data grab并不依赖一次性的强制占有,而是通过连接成为生活前提后持续发生。平台越不可退出,数据提取越容易被包装成服务交换的自然条件。真正的制度问题于是转向能否建立数据最小化、集体权利和公共基础设施,使社会参与不再以无限数据让渡为代价。】
- Taylor, Linnet. "What Is Data Justice? The Case for Connecting Digital Rights and Freedoms Globally" Big Data & Society 4, no. 2 (2017). (Read: entire article).
【按:Taylor提出data justice,是为了把隐私、数据保护与更广泛的社会正义问题连接起来。文章尤其有助于从批判转向规范性提问:如果数据化造成可见性、代表性、资源分配和行动能力的不平等,我们评价制度时不能只看个体consent。它因此可以作为本节结尾,把surveillance与coloniality的诊断转化为权利、能力和全球不平等的研究议程。data justice因此把数字权利从个人隐私扩展到结构性分配问题。谁能够访问数据、谁能够拒绝被数据化、谁承受自动化错误、谁能从数据经济中获益,这些问题无法用单一的“同意”原则解决。Taylor的框架使数据治理与社会正义相接,也为跨国比较保留了制度差异。Taylor把data justice放到全球数字权利框架中,能够避免把欧美隐私法经验普遍化。不同国家的数据基础设施与社会保障、身份制度和发展政策结合方式不同,数字权利也因此具有不同的紧迫性。跨国正义需要同时处理个人自由与资源分配,而不能只复制单一法律范式。data justice的分析单位因此不只是个人数据主体,而是由国家、平台、社区和基础设施共同构成的数据关系。data justice还能够连接几个通常被分开的政策领域:隐私保护关心数据控制,反歧视关心分类后果,发展正义关心基础设施与能力分布。Taylor的框架使这些问题进入同一规范空间,也提醒数字权利不能只以具有充分选择能力的个人消费者为默认主体。】
第四节:行星级计算与基础设施空间
本节把“云”重新放回土地、能源、芯片、数据中心、海底光缆、卫星、频谱、城市和法律安排。Bratton The Stack把行星级计算描述为多层架构,Easterling Extrastatecraft分析基础设施空间如何通过标准、物流与空间格式产生准治理效应,Star的基础设施研究则提醒我们,基础设施通常在顺利运行时退入背景,在故障、冲突或排除中才重新显形。平台权力因此不仅存在于界面、算法和商业模式中,也存在于连接条件、物质资源和维护体系之中。
基础设施化同时改变了权威与领土的关系。Sassen关于territory、authority和rights的重新组装,Graham与Marvin关于差异化城市网络的分析,以及Plantin等人对platform studies与infrastructure studies的连接,都说明“平台”与“基础设施”之间的界限正在变得不稳定。某些平台通过API、云服务、身份体系和支付接口成为其他组织无法轻易绕过的底层条件;而传统基础设施也越来越采用平台式的数据化、服务化和分层访问逻辑。这里的政治问题不是国家是否被平台简单取代,而是不同权威如何在跨尺度网络中重新分配。
Varoufakis与Durand提出的technofeudalism命题在这一基础设施背景下才具有可检验性。云资本、无形资产、平台入口、租金和退出障碍是否构成一种不同于传统资本-劳动关系的依附结构,需要与平台资本主义、垄断资本主义和租金理论逐项比较。Morozov等反方的重要性正在于迫使这一概念说明自己的增量解释力。若“封建主义”只是把垄断、租金与市场权力重新命名,它的理论必要性就十分有限;若平台控制的基础设施条件确实改变了市场进入、所有权与依附方式,则需要进一步说明这种改变发生在哪些制度层次。
必读文献:
- Bratton, Benjamin H. The Stack: On Software and Sovereignty. 10th anniversary ed., with a new preface by the author. MIT Press, 2026. Originally published 2016. (Read: new preface; "The Nomos of the Cloud"; "Platform and Stack, Model and Machine"; "Cloud Layer").
【按:Bratton的stack不是简单的“云计算国家论”。"Nomos of the Cloud"讨论行星计算对传统政治地理的挑战;"Platform and Stack"解释计算架构如何成为政治模型;"Cloud Layer"则把数据中心、平台、基础设施与国家治理放到同一架构中。与其把Google或Amazon直接命名为国家,而应精读sovereignty如何在不同层级和不同组织之间分布。采用MIT Press 2026十周年新版;阅读时既要回看原书对planetary-scale computation的架构化描述,也要利用新版序言检验这一框架在十年后的有效性与盲点。Stack的六层模型之所以有启发,不是因为现实恰好整齐分成Earth、Cloud、City、Address、Interface、User六层,而是因为它允许我们追踪跨层关系:能源与矿产如何支撑云层,地址系统如何让对象进入可计算秩序,接口怎样把制度规则表现成用户可操作的选择,"User"又如何可能是人、组织、传感器或机器代理。它是一种model of models,而非经验地图。2026新版最适合用来反问:十年来主权、云平台、AI代理与地缘政治的变化,哪些强化了stack的解释力,哪些要求修改其层级想象?Bratton的Stack提供了一种跨尺度架构语言:Earth、Cloud、City、Address、Interface与User不是简单的技术层,而是计算如何重新编排领土、身份和主权的分析切片。这一模型的优势是能够把数据中心、城市和用户界面放到同一行星系统中,风险则是架构抽象可能遮蔽地区差异。它更适合作为比较框架,而不是现实世界的完整地图。】
- Easterling, Keller. Extrastatecraft: The Power of Infrastructure Space. Verso, 2014. (Read: Chapter 2, "Zone: The Infrastructure Commons," pp. 25–70).
【按:Easterling关注的不是单纯“特区是法外之地”,而是infrastructure space如何通过空间产品、协议、标准、物流和制度安排产生超出传统国家规划的治理效应。把它用于数字平台时,需要保持类比的限度:Amazon的云平台和自由贸易区不是同一种空间,但二者都提示我们规则可能通过基础设施和标准而非传统法律文本来发挥作用。把平台称为国家、帝国或领地必须谨慎。自由贸易区与云平台的可比性只存在于基础设施化规制这一层面,二者的法律地位、空间结构和经济关系仍须分别处理。infrastructure space的政治性恰恰体现在它以技术规范、空间模板和运营标准组织行为,而不总以明确的法律命令出现。自由贸易区、物流网络与数据中心并非“国家之外”的真空,它们往往由国家政策、私人资本和跨国标准共同生产。Easterling由此提供了一种分析准治理效力的空间语言。更重要的是,infrastructure space经常通过复制性发挥力量。某种zone、园区或物流模型一旦被包装为标准化空间产品,就能够跨地区扩散,并携带一套预设的治理逻辑。数字平台的API、云架构和服务条款也具有类似的可复制性,但其空间和法律机制不同。比较应停留在规则随基础设施扩散这一层面。这种空间政治说明,基础设施的权力往往通过“默认环境”发挥作用:规则被嵌入空间与系统后,比显性的命令更难被辨识。其核心并非“法外”,而是规则如何以空间、标准和运营形式取得持续效力。】
- Star, Susan Leigh. "The Ethnography of Infrastructure" American Behavioral Scientist 43, no. 3 (1999): 377–391.
【按:Star提供本节不可替代的方法论基础。基础设施之所以难以看到,是因为它通常成为工作和生活的背景条件;只有故障、冲突或被排除时,基础设施才突然显形。对于云计算而言,这意味着不能只研究界面和软件,还必须追问电力、网络、机房、维护劳动和标准。若租金、网络效应和垄断可被既有资本主义理论充分解释,那么“封建”只是修辞;若存在新的依附、入口控制和非市场配置,则要说明这些机制为何需要新的社会形态概念。“租金、利润、入口控制与退出成本之间的政治经济差异”在Star "The Ethnography of Infrastructure"这里因此不只是概念问题。Star的方法论意义还在于把维护、修理与例外经验置于基础设施研究中心。对大多数使用者而言,基础设施因稳定运行而退到背景;对维护者、被排除者或故障中的使用者而言,它却始终可见。数字基础设施的社会差异因此常常在断网、停服、兼容失败或权限拒绝时最清楚地显现。基础设施的关系性还意味着,同一系统对不同群体并不以同样方式成为“背景”。稳定宽带对城市中产可能不可见,对偏远地区、残障用户或维护工人却始终是显性的限制条件。Star的民族志方法因此天然包含差异视角:基础设施是谁的基础设施,必须具体回答。维护视角还改变了创新叙事:一个系统是否可靠,往往取决于长期、重复而低可见度的修理劳动,而非最初设计本身。对数字系统而言,故障日志与维护现场往往比宣传材料更能显示真实的权力关系。】
- Plantin, Jean-Christophe, Carl Lagoze, Paul N. Edwards, and Christian Sandvig. "Infrastructure Studies Meet Platform Studies in the Age of Google and Facebook" New Media & Society 20, no. 1 (2018): 293–310.
【按:这篇文章几乎可以作为本节的方法论宣言:platforms与infrastructures不能再被完全分开,因为平台逐渐取得基础设施特征,而传统基础设施也被平台逻辑重新组织。它正好连接Star与Bratton,也让“云”不再只是一个技术层,而成为社会组织形式。行星尺度概念容易让具体地点消失。宏观模型只有落实到具体节点才具有解释力:矿产、芯片、数据中心、光缆、城市、劳动力或法律辖区,并检验不同地区承担的成本是否对称。若把Plantin "Infrastructure Studies Meet Platform Studies in the Age of Google and Facebook"置于本节的整体脉络中,“行星尺度模型与具体地点、劳动、维护之间的尺度张力”呈现出更具体的结构。Plantin等人提出的平台基础设施化与基础设施平台化,准确捕捉了Google、Facebook等数字平台从可选服务转向社会依赖节点的过程。平台取得基础设施性质后,其稳定性、普遍接入和公共责任问题会更加突出;传统基础设施引入平台逻辑后,则可能被数据化、商业化和接口化。两种变化共同模糊了公共设施与私人平台之间的边界。这种双向变化还会重写公共责任。基础设施通常被期待具备稳定、普遍和长期可用性,而平台企业可以快速改变服务、价格与规则。当私人平台成为教育、工作或公共沟通不可绕开的节点时,企业灵活治理与社会基础设施责任之间的矛盾便会凸显。】
推荐文献:
- Sassen, Saskia. Territory, Authority, Rights: From Medieval to Global Assemblages. Princeton University Press, 2006. (Read: Chapter 7, "Digital Networks, Global Cities," pp. 323–377).
【按:Sassen的重要性在于她没有把全球化理解为国家主权简单消失,而是讨论territory、authority、rights如何在跨国网络中重新组装。它为本节提供一个比“平台国家”更精确的理论框架:不同的权威形式可以同时存在,并在特定空间节点交叠。数字基础设施因此不是天然的“超国家空间”,而是新旧治理尺度重新组合的场所。Sassen所说的global assemblages使“国家退场”的叙事失去解释力。全球数字网络需要法律承认、城市节点、金融制度与跨境规则才能运作,它们既可能削弱某些国家能力,也可能重组并强化另一些能力。权威由此不是从国家简单转移给平台,而是在多重尺度之间重新装配。Sassen的分析还提示“去领土化”与“再领土化”往往同时发生。跨境数据流看似摆脱地方,但服务器选址、税收、司法管辖和人才集聚又在特定城市与国家重新固定。全球网络不是空间消失,而是不同尺度的空间关系被重新排序。因此,数字全球化并未取消territory,而是让领土、权威和权利以新的尺度组合进入同一基础设施网络。数字网络的跨国性因而始终伴随着新的节点集中和司法辖区竞争。全球性始终通过具体地点被生产出来。全球网络中的权威重组还意味着企业与国家之间并非零和。国家可以通过税制、数据本地化和安全政策塑造云基础设施,平台也可以利用跨境架构改变国家监管成本。Sassen的assemblage视角有助于分析这种共同生产关系,而不是用“谁取代谁”的主权叙事简化现实。】
- Graham, Stephen, and Simon Marvin. Splintering Urbanism. Routledge, 2001. (Read: Chapter 1, pp. 1–36).
【按:从城市基础设施的差异化供给理解网络如何制造空间分化。与Easterling对读,可以分析基础设施并不必然产生公共性,而可能通过价格、优先级、连接方式和空间选择形成“服务等级”。把平台称为国家、帝国或领地必须谨慎。splintering urbanism揭示了基础设施并不天然具有普遍主义方向。网络化服务可以通过价格、技术标准和空间投资形成分层连接,让高价值区域获得高速、冗余和稳定服务,而边缘地区承担中断与不足。将这一视角用于数字基础设施,可以进一步分析“连接”本身如何成为新的社会差异。网络分裂并不仅发生在有没有接入之间,也发生在接入质量、延迟、带宽、可靠性和服务可负担性之间。数字基础设施越成为工作、教育和公共服务前提,这些技术差异越会转化为社会机会差异。基础设施政治因此必须超越“普及率”这一单一指标。连接质量的分层最终会沉淀为教育、就业和公共参与机会的分层,基础设施公平也因此具有明确的分配政治。基础设施研究因此必须把服务质量和空间差异纳入,而不能只统计是否接入网络。基础设施差异最终会沉淀为社会机会差异。高质量连接、冗余保障与维护资源如何分布,往往比名义接入率更能揭示数字城市中的阶层和地域分化。网络公平因此必须进入基础设施设计和投资决策。Splintering Urbanism还提示,基础设施市场化会把普遍服务逻辑改写成盈利能力逻辑。高价值用户和地区更容易获得优先投资,而低收益空间长期处于维护不足。数字网络若完全依赖相同的商业选择机制,也可能把城市已有的阶层和地域差异进一步固化。】
- Mattern, Shannon. Code and Clay, Data and Dirt: Five Thousand Years of Urban Media. University of Minnesota Press, 2017. (Read: Introduction; selected chapters on urban information infrastructures).
【按:Mattern把城市媒介史拉长到文字、地图、建筑、物流、数据库和传感系统,拒绝把smart city当作突然降临的数字革命。她的关键方法是把code与clay、data与dirt并置:信息基础设施总要依附物质载体、空间制度与维护劳动。放在本节,它能给Bratton的宏观stack一个更历史化、更城市尺度的校正。“租金、利润、入口控制与退出成本之间的政治经济差异”通过Mattern Code and Clay, Data and Dirt: Five Thousand Years of Urban Media进入了组织与制度层面。Mattern把城市理解为长期由文字、声音、砖石、地图、传感器与数据库共同构成的媒介环境,从而抵抗“智慧城市”把数字化描述为突然到来的革命。城市计算必须落在既有的道路、管网、行政档案和社会记忆之上。数字基础设施因此不是覆盖旧城市的一层无形网络,而是与历史物质层不断叠加和摩擦。Mattern的长时段城市媒介史还反对“数据城市”将旧基础设施视为过时背景。传感器依附于道路,数据库继承档案分类,城市界面又受建筑和行政边界制约。数字化从来不是白纸上的重新设计,而是不断与既有物质、制度和历史层叠。这样的视角可以显著降低智慧城市话语中的技术断裂感。Mattern还使“媒介”概念从屏幕扩展到城市本身。街道标识、档案系统、建筑材料和传感器共同参与信息组织;所谓smart city只是这一漫长媒介化历史的新层,而不是城市第一次变得“可计算”。】
- Hu, Tung-Hui. A Prehistory of the Cloud. MIT Press, 2015. (Read: Introduction; selected chapters on cloud infrastructure, sovereignty, and labor).
【按:Hu通过“云”的前史拆解把网络想象成无地点、无摩擦空间的语言。他把云计算与更早的网络、军事、主机、数据中心和劳动关系联系起来,显示cloud是一种历史形成的基础设施与政治想象,而不是自然的技术层。与Bratton对读,可以比较架构模型与媒介考古两种方法如何处理同一对象。行星尺度概念容易让具体地点消失。Hu A Prehistory of the Cloud所揭示的“行星尺度模型与具体地点、劳动、维护之间的尺度张力”并不止于文本内部的概念争论。“行星尺度模型与具体地点、劳动、维护之间的尺度张力”因而仍需接受Hu A Prehistory of the Cloud之外的经验材料和竞争性解释的共同检验。Hu A Prehistory of the Cloud使“行星尺度模型与具体地点、劳动、维护之间的尺度张力”能够被限定在特定对象与因果链条之中;超出这一范围的类比,需要新的证据,而不能依靠术语本身完成。Hu追溯cloud这一隐喻的前史,揭示云计算如何借用无地点、无重量和无限扩展的想象遮蔽其军事、网络与数据中心基础。云的“虚拟性”是一种界面效果,而不是物质事实。数据中心的集中、用户终端的分散以及网络路径的层级化共同说明,云既制造空间抽象,也重新组织非常具体的地理依赖。cloud隐喻同时具有政治心理效应:它让用户更容易把存储和计算理解为一种自然环境,而不是由企业建设并收费的基础设施。服务因此看似像空气一样随处可得,后台却依赖少数超大规模运营者。Hu的历史化工作把这种自然化重新暴露为技术产业的文化产物。】
- Parks, Lisa. Cultures in Orbit: Satellites and the Televisual. Duke University Press, 2005. (Read: Introduction; selected chapters on satellite infrastructures and global visibility).
【按:Parks把媒介基础设施向轨道空间延伸,研究卫星如何把地面、信号、图像与地缘政治连接起来。它对“行星级计算”的意义在于提醒我们,planetary并不是修辞上的“全球”,而是由轨道、频谱、地面站、国家许可和资本投入构成的垂直空间。云服务同样需要在这些具体物质层中定位。在Parks Cultures in Orbit: Satellites and the Televisual的分析尺度上,“云、平台与基础设施如何连接土地、能源、标准和主权”首先表现为一组具体关系。technofeudalism是否成立,必须与垄断、租金和平台资本主义进行比较。若“封建”只是重新命名市场权力,它的解释增量有限;若基础设施控制确实改变了依附、进入与收益结构,则需要说明具体发生在哪些制度层面。卫星媒介把基础设施的垂直维度带入数字政治。轨道位置、地面站、频谱分配和遥感能力共同构成传播条件,使“全球覆盖”始终依赖高度具体的地缘政治安排。Parks因此使行星尺度不再只是宏观修辞,而成为可追踪的技术对象与资源竞争。从卫星电视到卫星互联网,轨道基础设施都显示通信主权具有垂直空间。国家边界无法完全解释信号覆盖,但发射能力、频谱协调和地面接收仍受制度控制。行星计算的“全球性”由此是由一系列稀缺位置、技术许可和资本投入具体制造出来的。卫星网络由此表明,“全球覆盖”不是没有边界,而是边界从地表延伸到轨道、频谱和地面站的制度安排。】
- Lehdonvirta, Vili. Cloud Empires: How Digital Platforms Are Overtaking the State and How We Can Regain Control. MIT Press, 2022. (Read: Introduction; selected chapters on platform governance and institutional power).
【按:Lehdonvirta把大型平台当作拥有规则制定、执行与争端处理能力的制度组织来比较,并追问它们为何能够在跨境数字市场中承担部分传统治理功能。标题中的 "overtaking the state" 应当作为需要检验的命题而非事实描述;其价值正在于让我们精确区分国家主权、私人治理、市场基础设施和用户依赖,而不是笼统宣称“平台已经成为国家”。Lehdonvirta Cloud Empires: How Digital Platforms Are Overtaking the State and How We Can Regain Control与Parks Cultures in Orbit: Satellites and the Televisual和Varoufakis Technofeudalism: What Killed Capitalism把“平台基础设施化与基础设施平台化的双向过程”置于不同的知识传统中。把平台称为国家、帝国或领地必须谨慎。Lehdonvirta Cloud Empires: How Digital Platforms Are Overtaking the State and How We Can Regain Control让“平台基础设施化与基础设施平台化的双向过程”摆脱了过于抽象的时代判断。technofeudalism是否成立,必须与垄断、租金和平台资本主义进行比较。若“封建”只是重新命名市场权力,它的解释增量有限;若基础设施控制确实改变了依附、进入与收益结构,则需要说明具体发生在哪些制度层面。Lehdonvirta把大型平台描述为类似“帝国”的治理结构,重点并非修辞上的国家类比,而是平台拥有规则制定、纠纷处理、身份管理和制裁能力。与传统国家相比,平台权威来自合同和基础设施依赖,缺乏领土主权却可能拥有跨境规模。比较两者最有价值之处,是明确不同类型治理能力的来源与限制。】
- Varoufakis, Yanis. Technofeudalism: What Killed Capitalism. Bodley Head, 2023. (Read: Chapter 3, "Cloud Capital," pp. 58–91; Chapter 4, "The Rise of the Cloudalists and the Demise of Profit," pp. 92–115; Chapter 7, "Escape from Technofeudalism," pp. 175–214).
【按:Varoufakis的核心命题是cloud capital改变了市场竞争与利润之间的传统关系,并通过平台入口、数据与云基础设施形成新的依附结构。这一文本真正有价值的问题不是接受"technofeudalism"这个词,而是检验它能否解释平台租金、网络效应、垄断和用户依赖。如果这些现象仍然能够通过垄断资本和平台资本主义解释,那么“封建主义”概念增加了什么?反之,它是否揭示了一种仅靠传统资本理论不容易说明的权力关系?精读时尤其要分清cloud capital与普通数字资本。Varoufakis的论证不是“科技公司很大所以封建”,而是平台积累了能够塑造行为、组织交易并从生态系统参与者处抽取cloud rent的基础设施,由此让市场本身越来越在私人平台内部被编排。Chapter 3与Chapter 4应连续读:前者定义cloud capital,后者才提出cloudalists与profit/rent关系的转变;Chapter 7则显示他并非只做时代诊断,还必须面对退出与政治重组问题。讨论要拿Amazon marketplace、app stores或云服务等不同案例检验这种结构是否一致。Varoufakis的cloud capital命题试图说明平台不再只是资本主义市场中的企业,而是通过云基础设施和用户行为塑造交易环境,并从依赖关系中攫取cloud rent。争论的关键不在“封建”一词是否醒目,而在利润、租金、市场和平台权力的关系是否真的发生了结构变化。若传统垄断理论已经能够解释这些收益,"technofeudalism"的额外解释力就必须被明确证明。】
- Durand, Cédric. How Silicon Valley Unleashed Techno-Feudalism: The Making of the Digital Economy. Translated by David Broder. Verso, 2024. Originally published as Technoféodalisme: Critique de l'économie numérique in 2020. (Read: Chapter 2, "On Digital Domination"; Chapter 3, "The Rentier Class of the Intangible"; Chapter 4, "The "Techno-Feudal" Hypothesis").
【按:Durand的版本结构非常适合与Varoufakis对读:Chapter 2讨论digital domination,Chapter 3转向无形资产与rentier class,Chapter 4才系统提出technofeudal hypothesis。由此可以把“技术封建主义”拆成三个层次:支配机制、积累形式和社会形态假说。这里尤其应比较他与Marxist rent theory、monopoly capitalism和platform capitalism的差异,而不是把“租金”直接等同于“封建地租”。Durand与Varoufakis的相似词汇掩盖了不同路径:他更强调数字支配、无形资产、知识产权与rentier logic如何重组积累,从而把technofeudalism放在资本主义内在发展与潜在形态转变的边界上。正因为Chapter 2–4呈递进结构,不能直接跳到“封建假说”;先要判断domination和intangible rent是否已经被证明,再讨论这些机制是否足以支持社会形态层面的命名。把两本书并排阅读的目标不是选冠军,而是比较其“封建性”的判准到底是什么。Durand How Silicon Valley Unleashed Techno-Feudalism: The Making of the Digital Economy与Varoufakis Technofeudalism: What Killed Capitalism和Thorhauge The Economic Lives of Platforms: Rethinking the Political Economy of Digital Markets共同使“行星尺度模型与具体地点、劳动、维护之间的尺度张力”从宏观判断下沉到机制层。行星尺度概念容易让具体地点消失。Durand的技术封建主义论证比单纯的时代命名更适合拆成三个问题:数字支配如何形成,无形资产如何产生租金,以及这些变化是否足以构成新的社会形态。这样处理后,“封建主义”就成为待检验的政治经济假说,而不是预先接受的结论。平台依赖、知识产权与垄断租金之间的关系必须分别分析。】
- Thorhauge, Anne Mette, Andreas Lindegaard Gregersen, Eva Iris Otto, Jacob Ørmen, and Morten Axel Pedersen, eds. The Economic Lives of Platforms: Rethinking the Political Economy of Digital Markets. Bristol University Press, 2024. (Read: "Rethinking the Political Economy of Digital Markets," pp. 1–16; "Modes of Connectivity," pp. 17–36).
【按:这部论文集的作用是给technofeudalism争论提供更细粒度的比较材料。它不从一个总括性社会形态概念出发,而考察平台怎样通过connectivity、market design、数据与中介关系组织经济活动。放在Varoufakis、Durand之后,可迫使这一讨论把“云地主”“租金”之类的大词拆成可研究的制度机制:平台到底控制了什么入口,如何定价,如何组织供需,何处真正存在退出成本。平台的经济生命不能只用“市场中介”概括。连接方式、定价机制、数据积累与规则治理共同决定平台如何创造并分配价值;不同平台类型之间也存在显著差异。论文集的意义正在于把政治经济学从单一商业模式推进到比较分析,避免以少数巨型科技公司代替整个平台经济。不同平台的connectivity modes还决定谁能够进入交易、以何种身份参与以及何种关系会被数据化。平台经济学若只研究价格和网络效应,容易忽略这些连接模式本身也是制度安排。把连接视为被设计的关系,有助于分析平台如何把市场结构写入技术接口。平台政治经济学也因此必须研究interface如何把经济关系制度化,而不能只在财务报表或市场份额中寻找平台权力。连接关系本身就是平台经济的制度设计,而不只是交易发生之前的技术前提。这正是“平台”区别于普通市场中介的重要方面。平台如何定义参与资格、交易单位和可见信息,实际上已经在技术接口中预先组织了市场关系。平台权力也由此进入市场形成之前。】
- Mezzadra, Sandro, Niccolò Cuppini, Mattia Frapporti, and Maurilio Pirone, eds. Capitalism in the Platform Age: Emerging Assemblages of Labour and Welfare in Urban Spaces. Springer, 2024. (Read: "Introduction. The Platform Age," pp. 3–14; "Operations of Platforms," pp. 15–31; "The Politics of Platforms," pp. 81–95).
【按:这部开放获取论文集可以把本节从经典数字劳动理论推进到近年的平台劳动研究。它把平台视为同时组织劳动、城市空间、福利与社会合作的基础设施,并把valorization、algorithmic management、worker subjectivity和alternatives放在同一研究框架中。与Scholz、Woodcock对读,可以检验“平台合作主义”究竟是在改变所有权,还是更广泛地改变了劳动组织与社会再生产。把平台称为国家、帝国或领地必须谨慎。数据中心、城市、边境、供应链和维护现场能够显示宏观架构如何被实际执行,也能暴露不同地区承担的环境与劳动成本是否对称。Capitalism in the Platform Age把平台研究推进到城市、福利与劳动再生产的交叉地带。平台不是孤立的应用程序,而可能重新组织交通、配送、照护和公共服务,使劳动市场与城市空间共同平台化。由此,平台资本主义的分析不能只围绕企业估值或数据提取,还要追踪地方制度如何被改写以及社会再生产成本如何重新分配。城市维度尤其能够显示平台资本主义与地方国家并非彼此替代。配送、网约车、短租和照护平台必须与交通规则、住房政策、劳工法和福利制度发生摩擦。平台的扩张既利用地方制度差异,也促使地方政府重新定义监管对象。平台化对福利制度的影响尤其值得重视。工作越被拆成临时任务和多平台收入,传统以稳定雇佣为基础的保险、税收和劳动保护越难覆盖。平台资本主义因此不只是企业组织创新,也在逼迫福利国家重新定义劳动身份与社会权利。城市成为这一冲突最先显现的制度尺度。】
- Morozov, Evgeny. "Critique of Techno-Feudal Reason" New Left Review, nos. 133/134 (2022). (Read: entire essay).
【按:Morozov对techno-feudalism的批判非常适合成为本节内部的反方。他质疑把数字垄断、知识产权、平台租金和金融化轻率地解释为资本主义已经让位于封建主义,要求重新检验资本主义竞争、利润与国家政策仍在发挥的作用。它使这一讨论不会被两本technofeudalism著作带着走,而是把“技术封建主义”变成一个可证伪、可比较的理论假说。借助Morozov "Critique of Techno-Feudal Reason",“租金、利润、入口控制与退出成本之间的政治经济差异”能够从修辞性命题转化为制度分析。 在Morozov "Critique of Techno-Feudal Reason"这里,“租金、利润、入口控制与退出成本之间的政治经济差异”并不导向单一结论,而是要求比较不同组织场景中的权力配置,由此判断概念的解释范围与失效条件。Morozov对technofeudal reason的批判为本节提供必要的反方:把当代资本主义过快宣布为“封建主义”,可能遮蔽资本、竞争、金融和国家仍然持续运作的机制。概念新颖性不能代替历史分期的证据。真正有价值的争论应当比较不同理论对同一现象——平台租金、垄断、基础设施依赖——能否给出更精确的因果解释。Morozov的批判同时提醒,历史类比会影响政治策略。如果把平台权力理解为封建领主关系,解决方案可能偏向“恢复资本主义竞争”;如果问题仍是资本集中和私有基础设施,则反垄断、公共所有和劳动政治会呈现不同优先级。概念选择因此不是命名游戏,而会组织政策想象。】
- Mosco, Vincent. To the Cloud: Big Data in a Turbulent World. Paradigm Publishers, 2014. (Read: Introduction; chapters on cloud computing, big data, and political economy).
【按:Mosco从政治经济学与传播研究角度对cloud的产业结构、劳动、数据与意识形态进行去神秘化。它比后来的stack/technofeudalism论述更早,也因此适合作为历史节点:云计算从一开始就不只是技术外包,而与企业集中、数据中心、能源、劳动和国家政策纠缠。行星尺度概念容易让具体地点消失。与其追求概念覆盖面的最大化,不如让理论保持可反驳性。只有在经验材料能够支持其核心机制时,这一框架才应继续延伸。“行星尺度模型与具体地点、劳动、维护之间的尺度张力”因而仍需接受Mosco To the Cloud: Big Data in a Turbulent World之外的经验材料和竞争性解释的共同检验。Mosco对“云”的批判首先是一种去神秘化:所谓cloud并没有消除地点,而是通过语言把服务器、数据中心、电力和劳动从用户经验中隐藏起来。云计算的意识形态效果与其物质架构因此必须同时研究;越是呈现为轻盈、无处不在的服务,越需要追踪其固定的土地、能源和资本投入。云计算的集中化还具有经济尺度效应。大型数据中心通过资本、能源采购和技术运维获得成本优势,使“随取随用”的便利与后台高度集中同时增长。Mosco的去神秘化因此不仅关乎语言,也关乎产业结构:云越像无处不在的公共环境,其所有权问题越容易被遮蔽。Mosco还把cloud放回全球传播政治之中:数据中心和网络服务的集中使少数企业拥有跨行业基础设施地位,传统媒体、政府和普通企业都可能成为其客户。云服务由此形成一种“基础设施依赖”,其政治问题不只在数据隐私,也在服务中断、定价和技术标准的控制权。】
第三组:中介、组织与劳动(第5–6周)
第五节:递归公众、中介与数字政治
本节从控制与提取转向公共性的制度条件。Kelty提出的recursive public并不是“在线讨论的人群”,而是能够参与维护、修改并再生产自身技术与制度基础的公共形式。自由软件社群之所以具有理论意义,不仅因为代码开放,更因为许可证、协作规范、讨论程序、版本控制和基础设施共同构成了一种可递归修改的公共生活。Warner关于publics与counterpublics的分析则强调,公众不是预先存在的固定群体,而是在传播、回应与再传播中被持续召集出来;Fraser进一步把参与不平等与反公众的问题置于公共领域理论内部。
Kornbluh Immediacy, or The Style of Too Late Capitalism使这一讨论获得了媒介形式上的张力。即时、透明、沉浸和“无中介”经常被当作数字文化的天然价值,但政治组织恰恰依赖程序、代表、延迟、解释、记忆与责任链。Tufekci、Dean、Gerbaudo等关于网络运动的研究进一步显示,低成本动员和高速度传播并不自动转化为持续的组织能力。连接可以迅速扩大参与规模,同时也可能让决策、资源积累和长期协调变得脆弱。
因此,数字政治的关键并不是在“中心化”与“去中心化”之间做抽象选择,而是判断何种中介能够被参与者理解、问责和修改。一个公众是否拥有自己的服务器并不是唯一标准,更重要的是它能否影响规则制定、资源分配、可见性结构和争议解决程序。公共性只有在媒介条件与组织形式都能够被政治化时,才不至于退化为平台提供的临时注意力空间。递归公众由此成为连接媒介技术、制度设计与集体自治的理论节点。
必读文献:
- Kelty, Christopher M. Two Bits: The Cultural Significance of Free Software. Duke University Press, 2008. (Read: Chapter 1, "A Recursive Public," pp. 27–63).
【按:这是本节中心文本。Kelty的recursive public指一种不仅使用公共媒介,而且能够修改、维护和重新生产其基础设施与制度条件的公共形式。重点阅读软件许可证、代码、讨论、维护与身份如何共同形成公共性。将它与平台时代相比,由此产生的问题是:当参与者失去对代码、服务器、账户和规则的实际控制之后,“公共性”还剩下什么?反过来,公共性是否必须拥有某种技术基础设施的自治权才能持续存在?公共性不能由参与人数或传播速度直接衡量。一个群体可能非常可见,却没有保存档案、修改规则、拥有基础设施或决定退出条件的能力;这些制度性维度必须与表达权区分。Kelty Two Bits: The Cultural Significance of Free Software进一步把“公共性是否拥有维护和修改自身媒介条件的能力”落实到制度关系之中。公共性并不是信息流量的同义词,而需要持续的规则、记忆、程序与媒介条件。平台能够扩大可见性,却也可能把公众赖以存在的接口、数据和传播结构掌握在私人治理之下。Kelty的recursive public不是普通的在线讨论群体,而是能够反过来修改自身交流与协作条件的公共。自由软件社群之所以具有递归性,正在于成员不仅使用代码和许可证,也参与维护这些制度-技术基础。平台时代的尖锐问题由此出现:当公共只能在不可修改的专有平台上形成时,它的组织自治究竟受到了怎样的限制?递归性还意味着公共必须拥有某种“修复自身”的能力。当软件、许可证或沟通渠道出现问题时,成员能够通过共同规则进行修改,而不是只能等待外部平台决定。由此,技术自治与民主自治在Kelty那里并不完全重合,却在维护权这一点上发生关键连接。】
- Kornbluh, Anna. Immediacy, or The Style of Too Late Capitalism. Verso, 2024. (Read: "Circulation," pp. 24–43; "Writing," pp. 65–112; "Antitheory," pp. 148–190).
【按:Kornbluh这部并不是一本(直接)讨论技术的书,因此加入本节恰恰是为了扩大问题:平台时代的公共性为什么越来越追求“直接”“即时”“沉浸”和“无中介”?她把这种文化形式称为immediacy,并指出它倾向于降低对representation、distance和mediation的容忍。对于recursive public而言,这形成一个重要张力:公共性需要规则、程序、讨论和维护,而即时性文化却倾向于把这些中介感受为障碍。由此形成的问题是:当政治传播越来越追求即时反应时,组织所需要的延迟、解释和制度程序会发生什么?"Circulation""Writing""Antitheory"三部分需要连起来读:immediacy不是单一媒介现象,而是一种把流通、第一人称表达、沉浸、透明和反理论姿态联结起来的晚期资本主义风格。她的polemical强度也必须成为批判对象:中介减少是否总是政治退步?边缘主体争取直接发声时,representation的制度价值该如何讨论?把Kornbluh放在Warner/Fraser/Kelty之间,正是为了避免将mediation单向美化;我们要区分必要中介、可问责中介与垄断性中介。中介并不天然反民主。真正的问题是中介能否被看到、问责和重构。Kornbluh对immediacy的批判把文化形式与政治组织连接起来。直接性、沉浸、第一人称和即时流通被当成价值时,中介容易被理解为异化或拖延;然而公共政治恰恰依赖代表、程序、解释与组织这些不可消除的中介。数字传播的速度因此具有双重性:它扩大响应能力,也可能侵蚀形成持续判断所需的距离与时间。】
- Warner, Michael. "Publics and Counterpublics" Public Culture 14, no. 1 (2002): 49–90.
【按:Warner强调public并不是一个先于媒介而存在的固定群体,而是在传播、回应、再传播过程中被不断召集出来。这个观点对平台时代尤其重要:推荐系统、转发、群聊、订阅和算法时间线都可能改变公众如何形成。把Warner与Kelty放在一起,可以讨论技术条件究竟怎样参与公共性的生产。网络动员的低成本与组织能力不是同一指标。快速聚集可以放大议题,却未必积累信任、资源、角色分工和冲突处理机制;阅读时应持续区分mobilization、organization与institution。Warner "Publics and Counterpublics"由此为“public、counterpublic、networked public与recursive public的差别”提供了一种有限而有效的解释尺度:它能够指出机制,但并不替代对组织、法律与物质条件的进一步分析。Warner强调public不是预先存在的人群,而是在文本流通与陌生人之间的持续回应中被召集出来。公共性因此具有媒介依赖:传播节奏、地址方式和可获得性都会改变“谁构成公众”。算法时间线进一步介入这一过程,使公共形成不再只由参与者的传播行为决定,也受到平台排序与推荐逻辑的塑造。Warner所谓counterpublic还强调某些公众以与主流不同的表达方式和社会节奏组织自身。数字平台一方面降低反公共形成的传播成本,另一方面又以统一的内容格式和指标衡量所有表达。平台标准化与反公共差异之间的张力,是理解网络公共性不可忽略的一层。】
- Fraser, Nancy. "Rethinking the Public Sphere: A Contribution to the Critique of Actually Existing Democracy" Social Text, no. 25/26 (1990): 56–80.
【按:Fraser通过subaltern counterpublics、参与平等和身份差异修正经典公共领域理论。它有助于避免把“统一公共领域”当作唯一理想,而是思考多个公共、反公共和边缘公共如何共存。若把Fraser "Rethinking the Public Sphere: A Contribution to the Critique of Actually Existing Democracy"置于本节的整体脉络中,“表达、基础设施所有权与民主治理不能被混为一谈”呈现出更具体的结构。公共性并不是信息流量的同义词,而需要持续的规则、记忆、程序与媒介条件。平台能够扩大可见性,却也可能把公众赖以存在的接口、数据和传播结构掌握在私人治理之下。Fraser的counterpublics概念同时修正了“一个共同公共领域”的规范想象。社会不平等意味着进入公共讨论的资源、语言与可见性从来不均等,边缘群体往往需要先形成自己的反公共空间。数字网络扩大了这种分化的可能,也使不同公共之间的相遇更容易受到平台排序与商业逻辑塑形。Fraser对participatory parity的强调进一步指出,形式开放并不足以保证公共参与平等。社会资源、身份地位和表达习惯都会影响谁更容易被听见。平台可以降低发言门槛,却同时通过注意力竞争制造新的不平等。多重公众的存在因此必须与参与条件的实际差异一起考察。counterpublic的价值正在于把差异保留在公共性内部,而不是要求所有参与者先转换成同一种语言和表达规范。】
推荐文献:
- Gramsci, Antonio. Selections from the Prison Notebooks. Edited and translated by Quintin Hoare and Geoffrey Nowell Smith. International Publishers, 1971. Originally published in Italian selections from the prison notebooks, 1948–1951. (Read: "The Intellectuals," pp. 5–23).
【按:本篇的作用不是把程序员简单命名为“有机知识分子”,而是重新理解知识、生产和组织之间的联系。Gramsci讨论的是知识分子如何从特定社会力量中产生,并在文化、组织与霸权形成中承担作用。将其迁移到数字环境,应分别考察软件工程师、协议开发者、平台管理者、数据劳动者与研究者的知识位置,而不是假设“懂技术的人”天然代表某一政治阶级。公共性不能由参与人数或传播速度直接衡量。 借助Gramsci Selections from the Prison Notebooks,“公共性是否拥有维护和修改自身媒介条件的能力”可以从宏观判断转向机制分析;概念是否成立,最终仍取决于相应的制度关系能否在经验层面被识别。Gramsci的有机知识分子概念强调知识劳动与社会集团、组织实践之间的联系,而不是赋予“聪明人”特殊身份。将其用于数字社会,更适合分析工程师、研究者、技术社群与劳动组织如何参与形成共同的世界观和组织能力。技术专业性本身并不决定政治立场;关键在于知识如何被组织、服务何种社会力量并转化为集体实践。有机知识分子的分析还要求区分技术权威与政治代表。工程师掌握系统知识,并不因此自动拥有替其他群体定义公共利益的资格;相反,专业知识只有进入组织、协商和共同实践,才能成为集体政治能力的一部分。这一区分对AI专家治理尤其重要。Gramsci的框架还强调知识分子与组织之间的双向关系:知识分子帮助组织社会经验,组织又为知识活动提供持续的社会位置。数字公共性若只依赖短周期意见领袖,很难形成这种长期功能。工程社群、劳动组织和公共研究机构如何积累知识并传递组织记忆,是“新有机知识分子”问题的真正核心。】
- Habermas, Jürgen. The Structural Transformation of the Public Sphere. Translated by Thomas Burger. MIT Press, 1991. Originally published 1962. (Read: Part V, pp. 141–180).
【按:作为公共领域理论的历史基线。重点不是把平台等同于Habermas所说的大众传媒,而是理解公共领域需要怎样的制度、媒介和社会条件,以及商业化如何改变这些条件。中介并不天然反民主。真正的问题是中介能否被看到、问责和重构。Habermas的历史分析仍然提供一个重要基准:公共性依赖特定的制度空间、媒介形式和社会资格,而不是仅仅依赖信息流通。平台时代的问题因此不能化约为“人人都能发言”;更关键的是公共讨论如何被商业模式、所有权、注意力分配和社会不平等组织起来,以及何种制度能够维持批判性交往。公共领域的商业化问题在数字环境中获得新的基础设施形式。传统大众传媒主要通过所有权和广告影响公共讨论,平台则进一步控制身份、排序、传播和反馈机制。公共领域理论因此需要从内容生产延伸到分发架构,讨论私有技术系统如何规定公共交往的条件。这一变化使“谁控制传播基础设施”成为公共领域理论无法回避的问题,媒介所有权与技术架构必须被同时讨论。公共性的物质条件由此从报刊、咖啡馆和广播进一步延伸到服务器、接口与推荐系统。媒介架构由此成为公共领域制度的一部分。因此,重新讨论公共领域不意味着复原旧媒介秩序,而是分析新的传播基础设施如何影响平等参与、议题形成与公共判断。Habermas的规范框架当然不能直接覆盖网络时代的碎片化公共,但它仍能提供衡量公共交往制度的若干尺度:参与条件是否平等、议题是否能脱离私人利益控制、理由能否公开接受争辩。将这些尺度与平台基础设施结合,公共领域理论仍具有批判效力,而不必复原其历史社会模型。】
- Benkler, Yochai. The Wealth of Networks. Yale University Press, 2006. (Read: Chapter 3, "Peer Production and Sharing," pp. 59–90).
【按:Benkler提供对等生产的制度与经济学分析。与Kelty一起阅读,可以讨论commons-based peer production是否提供了区别于企业和市场的组织形式,以及这些公共生产成果如何被平台重新吸收。网络动员的低成本与组织能力不是同一指标。“public、counterpublic、networked public与recursive public的差别”通过Benkler The Wealth of Networks进入了组织与制度层面。peer production的政治潜力来自生产协作与排他性产权之间关系的改变,而不是来自网络本身。开放许可证、模块化协作和共享资源能够降低参与门槛,却仍需要治理规则、维护劳动和可持续的物质基础。Benkler的分析由此可与平台化形成鲜明对照:开放协作并不等于平台免费劳动。Benkler的peer production还要求区分“非市场协作”与“免费向企业贡献”。前者通常存在共享规则、共同资源和参与者能够影响生产过程的制度,后者则可能由平台单方面控制基础设施和收益。两种形式都使用网络协作,却具有完全不同的产权与治理结构。这一区分对理解开源模型尤其关键。这一点也为开放模型提供清晰区分:一个项目即使允许免费下载权重,如果训练、治理和部署仍由单一公司决定,它与commons-based peer production之间仍存在重要制度距离。】
- Coleman, Gabriella. Coding Freedom. Princeton University Press, 2012. (Read: Chapter 3, pp. 61–84).
【按:Coleman的民族志使自由软件和黑客文化从抽象的“技术自由”进入具体社群。重点分析技术能力、伦理、政治身份和组织文化之间的关系。Coleman Coding Freedom所揭示的“表达、基础设施所有权与民主治理不能被混为一谈”并不止于文本内部的概念争论。“表达、基础设施所有权与民主治理不能被混为一谈”因而仍需接受Coleman Coding Freedom之外的经验材料和竞争性解释的共同检验。Coleman Coding Freedom使“表达、基础设施所有权与民主治理不能被混为一谈”能够被限定在特定对象与因果链条之中;超出这一范围的类比,需要新的证据,而不能依靠术语本身完成。Coleman Coding Freedom把自由软件伦理还原为具体社群实践:技术能力、幽默、法律意识、黑客伦理与组织边界彼此交织。公共性因此不仅来自开放许可证,也来自参与者如何形成对代码、规则和共同身份的持续承诺。Coleman的民族志尤其重要,因为它展示自由软件政治并非从许可证文本自动推导出来。幽默、技艺、声望、法律意识与日常协作共同塑造黑客社群的伦理。技术自治因而是一种被实践和组织出来的文化形式,而非只要代码开放就自然出现的政治属性。Coleman的研究还说明,技术共同体的政治性往往以法律细节和日常规范出现,而不是宏大宣言。对版权、匿名、言论自由和协作义务的具体理解塑造了社群行为。自由软件文化因此是一种制度化伦理,其稳定性来自长期实践,而不是成员共享某种单一意识形态。】
- Tufekci, Zeynep. Twitter and Tear Gas. Yale University Press, 2017. (Read: Introduction and Chapters 1–2).
【按:Tufekci关注网络媒介如何降低动员成本,却不一定提高长期组织能力。它适合与Kornbluh对读:即时性有利于快速传播,但政治组织需要记忆、信任、资源积累和程序。公共性不能由参与人数或传播速度直接衡量。在Tufekci Twitter and Tear Gas的分析尺度上,“公共性是否拥有维护和修改自身媒介条件的能力”首先表现为一组具体关系。因此,Tufekci Twitter and Tear Gas对“公共性是否拥有维护和修改自身媒介条件的能力”的说明只有在相应制度机制能够被经验材料识别时才成立。 “公共性是否拥有维护和修改自身媒介条件的能力”经由Tufekci Twitter and Tear Gas获得了一个更严格的问题形式:哪些机制真正发生变化,哪些只是媒介转换,以及这些变化由谁承受其后果。Tufekci Twitter and Tear Gas还提示,“公共性是否拥有维护和修改自身媒介条件的能力”的分析必须保留对象差异与制度边界,不能让概念的修辞强度替代经验说明。Tufekci Twitter and Tear Gas尤其显示,网络化运动可以在极短时间内获得传播规模,却未必同步形成谈判、资源管理和内部冲突处理能力。所谓“能力匮乏”并非对网络动员的否定,而是对组织基础与传播速度之间不对称的描述。Tufekci通过网络化抗争说明,数字媒介能够大幅降低动员成本,让运动迅速获得规模与可见性,却未必同步生成谈判、决策和冲突处理能力。传统组织缓慢积累的经验有时正是在快速动员中缺失的部分。由此,网络运动的“脆弱性”并非传播失败,而可能来自组织能力与传播能力之间的不对称增长。】
- Dean, Jodi. Crowds and Party. Verso, 2016. (Read: Chapter 3, "The Distributive Party," pp. 119–156).
【按:Dean的意义在于质疑“连接越多,政治组织越强”的假设。她讨论网络传播如何形成大量个体化的聚集,却不必然形成持续组织。与Kelty对读,能够把“递归公众”与“网络群众”区分开。Dean Crowds and Party与Tufekci Twitter and Tear Gas和Gerbaudo Tweets and the Streets: Social Media and Contemporary Activism把“即时性、传播速度与组织所需的中介和延迟之间的冲突”置于不同的知识传统中。中介并不天然反民主。真正的问题是中介能否被看到、问责和重构。Dean Crowds and Party让“即时性、传播速度与组织所需的中介和延迟之间的冲突”摆脱了过于抽象的时代判断。Dean Crowds and Party对“即时性、传播速度与组织所需的中介和延迟之间的冲突”的贡献,正在于让理论保持可反驳性。Dean Crowds and Party所强调的并不是对群众动员的简单否定,而是群众的瞬时聚集与能够承担决策、纪律和持续行动的组织形式之间存在结构差异。数字媒介可以放大前者,却未必自动生产后者。Dean所强调的组织问题使“网络连接”失去天然民主的光环。快速传播能够制造共同情绪和短期聚集,却未必形成承担决策、资源配置与长期责任的组织。crowd与party之间的张力由此成为理解数字行动主义的关键:政治能力不仅需要表达,也需要持久的组织形式。】
- Gerbaudo, Paolo. Tweets and the Streets: Social Media and Contemporary Activism. Pluto Press, 2012. (Read: Chapter 1, pp. 1–20).
【按:Gerbaudo对leaderless network的讨论能够分析社交媒体如何重新分配政治可见性和领导权。它不是证明网络运动“没有民主”,而是提示平台界面可能产生新的中心。网络动员的低成本与组织能力不是同一指标。围绕Gerbaudo Tweets and the Streets: Social Media and Contemporary Activism展开时,“public、counterpublic、networked public与recursive public的差别”可以被重新拆分为若干相互关联的环节。Gerbaudo对“无领导网络”的修正尤其重要。社交媒体并没有让领导消失,而是通过账号可见性、情绪编排和行动召集产生新的soft leadership。网络结构可以在形式上分布式,同时在注意力与议程设置上高度集中。数字行动主义因此需要分析隐性的策展、协调与代表机制,而不能只依据组织自我描述判断其去中心化程度。Gerbaudo所谓choreography of assembly显示,线上传播与线下集会之间存在具体的组织劳动。谁选择地点、时间和口号,谁维持信息流,都会塑造所谓“自发”运动。数字媒介并没有取消组织,而是使部分组织劳动以策展、转发与情绪动员的形式变得更不显眼。soft leadership也可以帮助理解平台时代的新型代表问题。高可见账号既可能有效协调行动,又可能在没有正式授权的情况下取得解释运动的权力。网络政治因此需要新的问责形式,使传播优势不自动转化为不可质疑的代表权。领导与横向参与之间不是二选一,而是组织设计问题。】
- Milan, Stefania. Social Movements and Their Technologies. Palgrave Macmillan, 2013. (Read: Chapter 2, pp. 31–54).
【按:讨论社会运动如何自行建设通信和媒介基础设施。它把recursive public的问题扩展到政治行动者,说明“拥有媒介”与“使用媒介”之间可能有巨大差别。Milan Social Movements and Their Technologies与Gerbaudo Tweets and the Streets: Social Media and Contemporary Activism和Juris Networking Futures共同使“表达、基础设施所有权与民主治理不能被混为一谈”从宏观判断下沉到机制层。历史类比只有在机制层面才具有价值。若只是因为今天出现了算法、平台或模型便宣告时代更替,理论便失去区分连续性与断裂的能力。Milan Social Movements and Their Technologies对“表达、基础设施所有权与民主治理不能被混为一谈”的贡献,在于把宏观判断压回可识别的制度机制。“表达、基础设施所有权与民主治理不能被混为一谈”因而仍需接受Milan Social Movements and Their Technologies之外的经验材料和竞争性解释的共同检验。社会运动技术的关键不只是选择哪一种媒介,而是能否建立对传播条件的相对自主。自主服务器、加密通信、独立媒体和技术志愿者都可能形成运动自身的基础设施能力。Milan由此把媒介使用推进为媒介治理问题:政治组织对技术条件的控制程度会直接影响行动的持续性。】
- Juris, Jeffrey S. Networking Futures. Duke University Press, 2008. (Read: Chapter 1, pp. 1–26).
【按:以全球化反抗运动为案例,观察“网络式组织”如何在真实行动中产生。它可以避免把分布式网络组织理解成纯粹的技术结构。公共性不能由参与人数或传播速度直接衡量。“公共性是否拥有维护和修改自身媒介条件的能力”从Juris Networking Futures出发,可以获得更清楚的因果层次。因此,Juris Networking Futures对“公共性是否拥有维护和修改自身媒介条件的能力”的说明只有在相应制度机制能够被经验材料识别时才成立。 围绕Juris Networking Futures展开时,“公共性是否拥有维护和修改自身媒介条件的能力”最值得追踪的是机制之间的连接,而不是寻找一个能够覆盖全部现象的统一术语。Juris Networking Futures的民族志进一步表明,网络式组织并不是技术拓扑向政治形式的直接投射。会议、信任、资源、行动传统和跨组织协商共同维持网络,所谓horizontalism也始终伴随实际的协调与非正式领导。Juris对反全球化运动的民族志显示,networking不仅是技术结构,也是一种组织规范和政治文化。横向协商、自治与多样性通过会议、邮件列表和行动网络被不断实践,同时也伴随协调成本与非正式权力。网络组织的民主性因此是一种需要维护的制度成果,而不是节点连接方式自动产生的属性。Juris同时展示横向主义的实际困难:非正式网络并不会自动消除权力,技术熟练度、资源、语言和旅行能力仍会产生中心。网络运动的组织分析因此需要同时看正式结构与非正式层级。所谓horizontalism更适合作为持续实践的规范,而不是已经实现的组织事实。】
- Papacharissi, Zizi. Affective Publics: Sentiment, Technology, and Politics. Oxford University Press, 2015. (Read: Introduction; selected chapters on affective publics and networked storytelling).
【按:Papacharissi把网络公共性中的affect、情绪表达与叙事结构放到中心位置,说明publics并不只靠理性论证和信息交换形成,也会围绕情感节奏、标签、可见性与故事流动被暂时组织起来。它适合接在Warner与Fraser之后,补足公共性理论中过度制度化或规范化的倾向,同时又可以与Berlant区分:这里关注的是公众形成,后者在下一节处理不稳定生活的情感依附。中介并不天然反民主。真正的问题是中介能否被看到、问责和重构。Papacharissi的affective publics把情感从公共理性的对立面中解放出来。标签、图像和叙事的流通能够把分散个体组织成共享感受的临时公众,情绪强度本身成为政治连接。问题在于这种连接能否转化为持续组织,以及平台对可见性和互动的优化如何选择性放大某些情绪。affect并不必然导向浅薄政治。情感可以维持记忆、身份和共同关切,也可能被短周期的注意力机制不断消耗。Papacharissi的框架因此适合区分情感公共的形成与制度化:前者解释连接如何发生,后者仍需要组织理论补充。情感公共能否持续,最终仍取决于组织、记忆与制度承载;情绪强度本身不能替代长期政治能力。情感流通还具有明显的平台依赖。某些情绪更容易通过短文本、视觉冲击和互动指标扩散,平台排序又可能进一步放大高参与度内容。情感公共不是纯粹自发的心理集合,而是在媒介形式与算法可见性中被塑形。理解数字政治情绪必须把这种基础设施条件纳入。】
- boyd, danah. "Social Network Sites as Networked Publics: Affordances, Dynamics, and Implications" In A Networked Self: Identity, Community, and Culture on Social Network Sites, edited by Zizi Papacharissi, 39–58. Routledge, 2011. (Read: entire chapter).
【按:boyd用networked publics讨论persistence、replicability、scalability、searchability等平台媒介条件如何改变公共互动。它的重要性不是宣称Facebook或Twitter自动生成新的民主公共领域,而是提供一套中层概念,说明媒介affordances如何改变发言的时间性、可见性、受众边界与再传播。与Warner的循环传播概念对读尤其有效。网络动员的低成本与组织能力不是同一指标。借助boyd "Social Network Sites as Networked Publics: Affordances, Dynamics, and Implications",“public、counterpublic、networked public与recursive public的差别”能够从修辞性命题转化为制度分析。boyd的networked publics框架强调affordances对公共交往的塑形:persistence、replicability、scalability与searchability改变了表达的时间、范围与可检索性。社交网络由此不是传统公共领域的简单线上版本。可见性长期保存、内容可复制和受众边界模糊,使身份管理、语境崩塌与公共参与同时发生。networked publics还改变了受众想象。发言者面对的往往不是明确群体,而是潜在、不可见且可被算法扩大的受众,这会改变自我呈现和风险判断。context collapse并非个人不会管理隐私,而是多个社会场景在同一平台界面被技术性地压缩。网络公共的社会张力由此进入界面设计。】
- Gillespie, Tarleton. Custodians of the Internet: Platforms, Content Moderation, and the Hidden Decisions That Shape Social Media. Yale University Press, 2018. (Read: Chapter 2, "The Myth of the Neutral Platform," pp. 24–44; Chapter 5, "The Human Labor of Moderation," pp. 111–140).
【按:书中一方面分析平台如何通过政策与技术规则进行内容治理,另一方面强调moderation依赖大量被隐藏的人类劳动。它因此能够修正“算法自动治理”的想象:平台治理常常是算法、规则、人工审查与商业风险管理共同完成的。平台内容治理由此具有准公共制度的性质,却缺乏公共机构通常承担的民主授权与程序约束。规则制定、自动检测和人工审核共同决定可见性边界,而商业公司同时掌握规则解释与基础设施。Gillespie的分析使“平台责任”不能只停留在删或不删内容,而必须处理治理合法性本身。content moderation还暴露私人平台的制度悖论:它们既宣称只是中立中介,又必须不断制定类似公共秩序的规则。审核规模迫使平台同时依赖自动系统与全球人工劳动,规则解释因而成为持续的组织过程。治理争议不能被压缩为“言论自由对审查”的单一对立。因此,moderation既是内容政策,也是劳动制度和基础设施治理问题,三者共同决定平台公共性的实际边界。平台因此既是私人公司,也是事实上的交往基础设施,这种身份张力构成治理争议的根源。moderation的规模还制造了“规则-案例”之间的持续摩擦。统一政策必须面对语言、文化和政治情境差异,自动检测又难以完整理解语境。平台因此不断在全球一致性与地方判断之间调整。内容治理研究的重点不应只是单个争议决定,而是这一庞大制度如何生产可执行的分类和例外。审核规则的政治性也体现在例外如何处理。越是试图把复杂语境压缩成统一政策,越需要大量边界判断;这些边界判断正是平台事实上行使公共权力的地方。】
第六节:平台劳动、不稳定性与新阶级关系
本节把平台政治经济落实到劳动过程与生活时间。平台通过派单、评分、动态定价、任务拆分、身份验证和自动停权重新组织雇佣关系,使“老板”的权力可以表现为界面、指标与模型,而不必总是由一个可见的管理者直接实施。Scholz、Woodcock、Rosenblat与Aloisi/De Stefano从合作主义、工人斗争、司机经验和劳动法等不同角度说明,算法管理并没有取消劳动关系,反而可能通过信息不对称和责任分散使劳动控制更难识别。
Terranova的free labor、Casilli关于微任务的研究、Gray与Suri对ghost work的描述,则把数字劳动扩展到用户参与、内容审核、数据标注与训练支持。所谓“自动化”往往依赖大量被空间上分散、在叙事上隐去的人类劳动;AI系统的运行也因此不能只从模型和算力理解。平台化进一步把工作风险转移给劳动者个人,使等待、接单、维持评分、承担设备成本和随时可用成为劳动过程的一部分。阶级关系由此不仅体现为工资水平,也体现为信息权、退出条件、数据所有权和时间支配。
Berlant Cruel Optimism为这一经验结构提供了情感-时间维度。不稳定劳动之所以能够持续,并不只是因为外部强制,也因为“灵活”“自由”“创业”“自我实现”等愿景可能成为主体维持生活叙事的依恋对象。平台合作社、工会、集体议价和公共基础设施所面对的因此不只是所有权问题,还包括风险如何分配、生活时间如何组织以及劳动者是否能够形成可持续的共同政治主体。制度替代只有进入这些层次,才可能超越“换一个平台所有者”的有限想象。
必读文献:
- Berlant, Lauren. Cruel Optimism. Duke University Press, 2011. (Read: Chapter 1, "Cruel Optimism," pp. 23–50; Chapter 6, "After the Good Life, an Impasse," pp. 191–222; Chapter 7, "On the Desire for the Political," pp. 223–264).
【按:Berlant不提供平台资本主义的直接理论,却提供一个理解平台劳动为何持续被“希望”黏住的强大概念。Cruel optimism指主体依恋于一个能够阻碍其繁荣的对象;Chapter 6把这种关系放进precarious present、工作和生活时间中;Chapter 7则进一步讨论政治欲望与行动的困难。将其放到平台劳动之后,可以提出一个技术方案经常忽略的问题:劳动者为什么会继续依赖“灵活、自主、创业式”的工作形式,即使这些形式不断损耗安全感和未来规划?制度替代不仅要改变平台所有权,也要处理这种被希望、生活节奏和不稳定性共同维系的依附。Chapter 1的概念定义必须与Chapter 6的impasse和Chapter 7的political desire连读。Berlant关心的不只是人为什么舍不得某个坏对象,而是当"good life"基础设施失灵后,主体如何在长期不稳定的present中improvisationally维持生活;政治欲望又为什么不能自动凝结为集体方案。这对平台劳动特别重要,因为flexibility既可能是平台话语,也可能在照护、迁移、教育或失业条件下成为劳动者真实需要。只有同时承认依恋的物质合理性与其结构性代价,cruel optimism才不会被简化成“工人被洗脑”。Berlant的cruel optimism提供的不是平台劳动经济学,而是一种理解依附关系的情感政治。主体可能依恋于“灵活工作”“自主创业”或“好生活”承诺,即使这些对象持续削弱其获得稳定生活的条件。precarious present因此不仅由工资和合同构成,也由时间感、希望与日常适应维系。劳动制度改革若忽略这种依附,很难解释不稳定形式为何具有持续吸引力。】
- Terranova, Tiziana. "Free Labor: Producing Culture for the Digital Economy" Social Text 18, no. 2 (2000): 33–58.
【按:Terranova的free labor讨论的是数字文化中自愿参与、社会互动与资本价值化之间的关系。它提醒我们不要把数字劳动仅仅理解为“付工资的在线工作”:用户创作、社交与数据生产本身也可能成为资本提取的条件。算法管理并不意味着管理者消失。规则设计、客服、运营、风险团队和资本所有者仍然存在,只是部分命令通过界面、评分和自动停权执行;责任追踪因此应穿透“是系统决定的”这一组织话语。“免费”参与、受薪雇佣与微任务劳动必须保持分析区分,否则数字劳动概念会失去辨识不同强制程度的能力。算法管理并没有取消管理关系,只是把部分命令转化为派单、评分、动态定价和自动停权。界面的非人格化容易分散责任,却并不意味着劳动者面对的控制因此减弱。free labor概念最有力量之处,在于它指出自愿参与与资本价值化可以同时发生。用户并不因为享受创作、社交或游戏就不再生产价值;反过来,把所有在线活动都命名为劳动也会抹平工资关系与强制程度的差异。Terranova因此提供的是一条分析连续谱,而不是把生活整体劳动化的简单公式。Terranova还揭示资本价值化与文化自主之间并非零和关系。数字文化可以真实地创造社群、审美和知识,同时这些活动又被平台转化为流量和数据。批判的任务不是否定参与者的自主意义,而是说明价值如何在参与者不完全控制的基础设施中被捕获。free labor的分析力度恰恰来自这种双重性:参与可以是真实的自主实践,同时也可以成为平台积累的无偿条件。】
- Scholz, Trebor. Uberworked and Underpaid: How Workers Are Disrupting the Digital Economy. Polity Press, 2016. (Read: Chapter 4, "Crowd Fleecing"; Conclusion, especially the sections on "Tactical Refusal, Defection, and Withdrawal from Data Labor" and platform cooperativism).
【按:本书的核心不是“开发一个工人版Uber”,而是重新提出谁承担平台经济的风险、谁控制数据、谁拥有平台。Chapter 4讨论crowd fleecing;结论则把refusal、defection、withdrawal与platform cooperativism联系起来。这样读能够避免把合作主义简化成一项商业创业,而把它放回劳动所有权、集体议价和基础设施治理。关于自动化最需要避免的是substitution fetish:机器替代某个任务,不等于一个职业或全部劳动消失。platform cooperativism的制度核心并非“复制现有平台后换一个所有者”,而是重新安排控制权、剩余分配和数据治理。合作社如果仍依赖外部云服务、资本融资或封闭算法,其自治仍然可能受到基础设施约束。Scholz的问题意识因此最终指向所有权之外的治理能力与技术主权。Scholz的platform cooperativism还要求面对规模问题。合作所有权能够改变收益与决策权,但网络效应、融资、软件维护和用户获取仍可能迫使合作平台依赖现有巨头。替代方案的可行性因此取决于能否建立共享技术、公共采购和跨合作社联盟等更广泛的制度生态。合作主义的难点由此从企业所有权进一步延伸到技术栈、融资和市场接入,平台自治需要更完整的制度生态。工人所有权只有与基础设施能力结合,才可能形成真正区别于传统平台的组织形式。合作制的技术条件不能被视为中性的外部资源。】
- Woodcock, Jamie. The Fight against Platform Capitalism: An Inquiry into the Global Struggles of the Gig Economy. University of Westminster Press, 2021. (Read: Chapter 3, pp. 45–67).
【按:Woodcock把平台劳动中的集体行动重新放回工人实践:停工、断联、组织、工会和跨平台协作。它与Scholz的制度替代路线形成张力:即使存在合作平台,劳动者仍然需要建立实际的组织能力和议价力量。合作主义也不是技术万能药。若把Woodcock The Fight against Platform Capitalism: An Inquiry into the Global Struggles of the Gig Economy置于本节的整体脉络中,“合作社、工会、劳动法与公地治理分别改变哪一层权力”呈现出更具体的结构。算法管理并没有取消管理关系,只是把部分命令转化为派单、评分、动态定价和自动停权。界面的非人格化容易分散责任,却并不意味着劳动者面对的控制因此减弱。Woodcock把平台劳动冲突放回工人组织实践,而不是仅仅讨论替代性商业模型。罢工、集体下线、工会和跨平台网络显示,即使劳动关系被任务化和个体化,集体行动并未消失。算法管理反而可能创造共同的不满对象。平台劳动政治的核心因此仍是组织能力:如何把分散经验转化为可持续的集体力量。平台工人的组织还必须面对劳动过程的技术中介。平台能够通过个性化派单和动态定价削弱共同经验,但同一界面也可能让工人发现相似的不公并发展新的组织工具。技术既是管理媒介,也可能成为集体行动的对象和资源。这种组织实践说明,平台劳动者并非算法的被动对象;他们也会围绕算法规则形成知识、策略与集体反制。】
推荐文献:
- Srnicek, Nick. Platform Capitalism. Polity, 2017. (Read: Chapter 2, pp. 35–63).
【按:把平台理解为数据提取与市场中介的经济组织形式。它为本节的“算力集中”提供资本主义规模效应与网络效应的前史,但不能直接把platform capitalism等同于AI capitalism。理论上的相似并不自动意味着制度上的同一。Srnicek把平台视为在数据提取、网络效应与市场中介中形成的特定商业模型,这一框架为理解平台集中化提供了经济基础。平台之间并非同质:广告、云、工业和精益平台依靠不同的收入结构与资产配置。区分这些类型,是分析平台劳动和算力集中时避免概念泛化的前提。这一类型学同时能够解释为什么“平台”不能被直接当作一种统一阶级关系:不同平台对固定资产、劳动合同和数据的依赖程度并不相同,其集中化路径也随之不同。平台资本主义框架的另一个优势,是把数据置于商业竞争而非抽象“信息社会”中。企业之所以积累数据,是因为数据能够改善服务、锁定用户、训练模型并形成进入壁垒。算力集中与数据集中因此需要与具体盈利模式相连,而不能仅以技术规模解释。由此,平台集中既来自网络效应,也来自数据、资本和基础设施之间的正反馈,任何替代政策都需要明确作用层级。平台类型学由此也是政策分析的前提,不同集中机制需要不同的规制和公共替代方案。经济类型的区分先于任何统一的“平台治理”方案。Srnicek的类型学也提示平台扩张存在不同资本强度。轻资产平台依赖外包和风险转移,云平台则需要巨额固定资本;两者都叫platform,却在劳动组织和竞争壁垒上完全不同。把平台资本主义与AI基础设施连接时,尤其需要区分这种资本结构,否则“平台”会掩盖算力产业的重资产特征。】
- Huws, Ursula. Labor in the Global Digital Economy: The Cybertariat Comes of Age. Monthly Review Press, 2014. (Read: Chapter 1, pp. 11–38).
【按:从全球分工与商品化角度理解数字劳动。Huws提醒我们,数字劳动并不只是软件工程师劳动,而是服务、外包、平台和知识劳动共同组成的新型全球劳动结构。算法管理并不意味着管理者消失。Huws Labor in the Global Digital Economy: The Cybertariat Comes of Age所提供的并不是对“不稳定生活的情感依附与劳动制度之间的关系”的终局命名,而是一种可被竞争性解释检验的理论切口,这种开放性本身构成其研究价值。Huws把数字劳动置于全球商品化与劳动分工之中,反对把“知识劳动者”当成数字经济的唯一主体。客服、外包、物流、制造和办公室劳动都在信息化过程中被重新标准化与远程组织。所谓cybertariat因此连接高技能与低技能、中心与外围,使数字化的阶级结构必须在全球劳动链条中理解。全球视角也能纠正“数字劳动=远程白领劳动”的偏见。高端软件产业与低薪后台服务往往处于同一价值链,前者的自动化表象可能建立在后者的标准化劳动之上。阶级分析必须同时处理这些空间上分离、功能上相互依赖的工作。数字阶级分析因此必须同时观察软件界面与供应链,把可见的知识劳动和被隐藏的支持劳动重新放入同一结构。Huws还把家庭与社会再生产纳入数字劳动视野。远程化与灵活化并没有消除照护负担,而可能把工作时间进一步渗入家庭空间。平台劳动的自由时间表因此需要与性别化再生产责任一起分析;同一种“灵活”对不同劳动者可能意味着自主,也可能意味着工作与生活边界的持续瓦解。】
- Casilli, Antonio A. Waiting for Robots: The Hired Hands of Automation. Foreword by Sarah T. Roberts. Translated by Saskia Brown. University of Chicago Press, 2025. Originally published in French as En attendant les robots: Enquête sur le travail du clic in 2019. (Read: Introduction; selected chapters on micro-work, clickwork, data labor, and automation).
【按:重点是“自动化背后仍有劳动”。micro-work、数据标注和内容审核说明AI的训练和运行经常依赖高度分散、低可见度的人类劳动。由此可以把“平台劳动”与“AI劳动”连接起来。采用2025年英译本,以便把2019年法文研究与生成式AI兴起后的“自动化”话语直接对接;但其核心仍是隐藏劳动,而不是把本书误写成对大模型的事后评论。“自动化叙事如何隐藏微任务、数据标注与维护劳动”通过Casilli Waiting for Robots: The Hired Hands of Automation进入了组织与制度层面。Casilli最重要的贡献,是把所谓“自动化”重新拆解为大量人类微劳动的组织过程。点击、标注、验证与审核并非机器智能之外的偶然补丁,而经常是模型训练和服务运行的必要条件。由此,自动化不是劳动的简单消失,而是劳动在空间、合同与可见性上的重新配置。Casilli的分析进一步说明,人类劳动常被嵌入自动化系统的反馈回路。模型需要不断标注、验证、纠错和内容判断,平台则通过众包把这些环节拆散到全球劳动力市场。所谓human-in-the-loop并不是暂时过渡阶段,而可能成为规模化AI服务的长期组织方式。自动化由此与新的劳动市场共同成长。微劳动的全球分布还使传统劳动法的雇主识别变得困难。工人可能面对多层承包商、众包平台和最终AI公司,责任链被拆散。自动化伦理因此必须重新连接供应链责任与劳动标准。】
- Gray, Mary L., and Siddharth Suri. Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass. Houghton Mifflin Harcourt, 2019. (Read: Chapters 1–3).
【按:展示AI背后的微任务、内容审核和平台化人类劳动。它可以与Casilli、Crawford对读,把“自动化”重新理解为一种劳动重新分配和劳动不可见化。合作主义也不是技术万能药。Gray Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass所揭示的“合作社、工会、劳动法与公地治理分别改变哪一层权力”并不止于文本内部的概念争论。ghost work揭示的正是自动化叙事中的基础设施悖论:系统越被描述为无需人工,维持其“智能”所需的人类判断越容易被隐藏。全球微任务市场把劳动拆成离散、按需且难以组织的单元,使技术效率建立在新的劳动不稳定性之上。这也意味着AI劳动政策不能只围绕技术岗位展开,微任务工人的工资、职业安全、申诉渠道与跨境劳动权利同样属于人工智能治理的组成部分。Gray与Suri还揭示ghost work的组织悖论:工人被分散到个人账户中,却需要通过非正式社群交换经验、识别骗局和互相帮助。平台没有提供稳定工作组织,劳动者便自行创造协作网络。数字劳动的集体性并未消失,而是从正式工作场所转移到更脆弱、更难获得制度承认的空间。这种劳动组织也说明“按需”并不意味着需求自然存在。平台通过任务分解、定价和质量控制主动塑造劳动市场,工人的可替代性是被技术和合同设计共同制造出来的。因此,劳动不可见性与技术神话之间形成了直接的制度联系。】
- Rosenblat, Alex. Uberland: How Algorithms Are Rewriting the Rules of Work. University of California Press, 2018. (Read: Chapters 2–4).
【按:Rosenblat从司机经验出发分析算法调度、价格、评价和信息不对称。它为“算法管理”提供具体民族志,使平台劳动不再仅仅被理解成抽象的资本-劳动关系。在Rosenblat Uberland: How Algorithms Are Rewriting the Rules of Work的分析尺度上,“平台劳动的任务化、算法管理、风险转移与集体行动”首先表现为一组具体关系。自动化常常通过隐藏劳动实现。数据标注、内容审核、微任务和维护工作被拆分到全球劳动力市场,使“机器完成任务”的叙事遮蔽了新的劳动组织与风险转移。Uber的算法管理并不取消管理,而是把命令分散到价格提示、任务分配、评级和信息不对称之中。司机在形式上具有自主选择,在实际工作中却持续响应平台设计的激励环境。Rosenblat的民族志因此能够具体说明“自主”与“控制”如何在同一劳动制度中并存。Rosenblat还揭示information asymmetry如何成为管理工具。平台能够同时掌握需求、价格和司机表现,而司机只能看到局部信息并据此作出选择。算法管理的力量因此不只是自动化命令,也来自对市场全局信息的垄断以及对信息披露节奏的控制。信息不对称正是算法管理区别于传统现场监督的重要维度:劳动者面对的是一个能观察自己、自己却难以观察的平台。这种信息结构使“自主劳动”呈现出明显的不对称性:选择权存在,却建立在平台更充分的信息优势之上。】
- Aloisi, Antonio, and Valerio De Stefano. Your Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labour. Bloomsbury, 2022. (Read: chapters on algorithmic management and worker protection).
【按:从劳动法与制度设计角度处理algorithmic management。重点关注“老板”不一定是一个人,而可能表现为调度模型、绩效评分、自动停权和规则系统;这能够把政治经济学讨论与劳动制度连接起来。Aloisi Your Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labour与Rosenblat Uberland: How Algorithms Are Rewriting the Rules of Work和Irani Chasing Innovation: Making Entrepreneurial Citizens in Modern India把“不稳定生活的情感依附与劳动制度之间的关系”置于不同的知识传统中。算法管理并不意味着管理者消失。Aloisi Your Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labour让“不稳定生活的情感依附与劳动制度之间的关系”摆脱了过于抽象的时代判断。自动化常常通过隐藏劳动实现。Aloisi Your Boss Is an Algorithm: Artificial Intelligence, Platform Work and Labour对“不稳定生活的情感依附与劳动制度之间的关系”的贡献,正在于让理论保持可反驳性。Aloisi与De Stefano从劳动法切入算法管理,使“老板是算法”这一流行说法重新获得制度内容。派单、绩效、定价和停权可以自动化,但劳动关系中的控制权仍由企业设计并承担法律后果。关键问题是传统雇佣分类、知情权和集体协商如何适应这种非人格化管理,而不是把责任转交给软件。】
- Irani, Lilly. Chasing Innovation: Making Entrepreneurial Citizens in Modern India. Princeton University Press, 2019. (Read: chapters on entrepreneurial citizenship, design and development).
【按:Irani研究“创新”如何成为治理和经济政策中的道德语言,并观察创业主体如何被塑造出来。与Berlant对读时,可以提出:当劳动者被要求成为“创新者”“创业者”和自我管理者,权力是否通过希望与自我实现的语言进入劳动关系?围绕Irani Chasing Innovation: Making Entrepreneurial Citizens in Modern India展开时,“自动化叙事如何隐藏微任务、数据标注与维护劳动”可以被重新拆分为若干相互关联的环节。自动化常常通过隐藏劳动实现。 从Irani Chasing Innovation: Making Entrepreneurial Citizens in Modern India来看,“自动化叙事如何隐藏微任务、数据标注与维护劳动”不能脱离制度授权与资源分配单独成立;技术形式只有与组织权力结合,才会产生稳定的社会后果。Irani研究innovation如何从经济口号转化为主体塑造机制。设计思维、创业教育和发展项目鼓励个人把社会问题重新描述为可创业解决的机会,也把结构性风险转化为自我负责。与平台劳动并置,entrepreneurial citizenship解释了为什么“不稳定”能够被重新包装为自主、灵活与创新。“创业公民”还改变了失败的归因方式。当创新被塑造成个人责任,项目失败更容易被解释为能力不足,而不是基础设施、资本与政策条件的结果。平台经济中的自我品牌、评分和自由职业文化延续了这种主体化,使劳动风险被重新描述为个人选择。】
- Schor, Juliet B. After the Gig: How the Sharing Economy Got Hijacked and How to Win It Back. University of California Press, 2020. (Read: Introduction; selected chapters on platform business models, inequality, and alternatives).
【按:Schor通过共享经济平台的经验研究区分不同平台商业模式、劳动参与方式与分配结果,避免把所有gig economy经验压成同一个剥削模型。她特别适合放在Scholz与Woodcock之后:平台合作主义和工人斗争固然重要,但平台劳动者的依赖程度、收入结构和劳动动机并不完全相同,因此替代方案也不能只靠单一组织模板。Schor After the Gig: How the Sharing Economy Got Hijacked and How to Win It Back与Irani Chasing Innovation: Making Entrepreneurial Citizens in Modern India和van Doorn "Platform Labor: On the Gendered and Racialized Exploitation of Low-Income Service Work in the "On-Demand" Economy"共同使“合作社、工会、劳动法与公地治理分别改变哪一层权力”从宏观判断下沉到机制层。合作主义也不是技术万能药。Schor After the Gig: How the Sharing Economy Got Hijacked and How to Win It Back对“合作社、工会、劳动法与公地治理分别改变哪一层权力”的贡献,在于把宏观判断压回可识别的制度机制。自动化常常通过隐藏劳动实现。Schor区分共享经济早期的社会愿景与后来平台商业化之间的分化,展示效率、社区和额外收入叙事如何被风险资本和规模扩张重新组织。其经验研究提示,平台经济不是单一路径:劳动条件、所有权和社区规则可以产生不同结果。问题因此转向何种制度安排能够把共享从营销语言转化为真实的资源治理。】
- van Doorn, Niels. "Platform Labor: On the Gendered and Racialized Exploitation of Low-Income Service Work in the 'On-Demand' Economy." Information, Communication & Society 20, no. 6 (2017): 898–914. (Read: entire article).
【按:van Doorn强调on-demand服务劳动并不是抽象的“灵活劳动”,而是高度性别化、种族化并依赖城市服务关系的劳动。文章把平台资本主义与既有低薪服务业、移民劳动和社会再生产连接起来,能够校正只从技术界面或算法调度出发的分析。“平台劳动的任务化、算法管理、风险转移与集体行动”从van Doorn "Platform Labor: On the Gendered and Racialized Exploitation of Low-Income Service Work in the "On-Demand" Economy."出发,可以获得更清楚的因果层次。van Doorn把性别化、种族化和低收入服务劳动置于platform labor中心,修正以司机或高技能自由职业者代表整个零工经济的倾向。清洁、照护、配送等工作原本就处于社会再生产与移民劳动的不平等结构中,平台化进一步将评价、匹配和支付数字化。算法并未创造这些差异,却能够重新编码并扩大它们。“按需”经济的性别与种族结构还说明技术中介不会抹平劳动市场的历史分层。客户评分、地理匹配和平台分类可能把既有偏见转换为看似个体化的交易结果。劳动公平因而既涉及算法设计,也涉及移民身份、照护责任和社会保障等平台之外的制度。平台化由此可能把社会偏见从雇主的直接判断转化为分散的客户评价与系统排序,使歧视责任更难被定位。算法化并未让不平等去身份化,反而可能使其通过看似个体化的数据反馈持续运作。】
- Graham, Mark, Isis Hjorth, and Vili Lehdonvirta. "Digital Labour and Development: Impacts of Global Digital Labour Platforms and the Gig Economy on Worker Livelihoods" Transfer 23, no. 2 (2017): 135–162. (Read: entire article).
【按:Graham、Hjorth与Lehdonvirta把数字劳务平台放入全球发展与跨境劳动市场,考察在线外包如何扩大工作机会,同时制造议价不对称、供给过剩和地域工资差。它使本节从Uber式本地gig work扩展到全球远程劳动,并提醒我们“平台劳动”内部存在不同空间尺度。算法管理并不意味着管理者消失。Graham、Hjorth与Lehdonvirta从全球数字劳动市场考察在线外包对发展与地理不平等的影响。互联网降低了跨境交易成本,却没有消除工资差异、语言等级和地域依赖。平台可以把劳动需求全球化,同时把工人留在各自不同的社会保障制度中。数字劳动由此既跨越边界,又利用边界。这种全球市场还制造了新的可替代性。雇主可以在多个国家之间快速转移任务,工人却受到本地生活成本、货币和法规约束。远程平台因此形成一种不对称的空间自由:资本和任务更易流动,而劳动者的社会再生产仍然高度地方化。跨境数字劳动的空间政治因此不是“地理消失”,而是地理差异被平台转化为成本优势和议价差异。全球数字劳动市场的“发展效应”因此具有双重性:它可以为边缘地区提供新的收入来源,也可能利用当地低工资和薄弱社会保障维持全球成本差。评价其发展价值不能只看就业数量,还要看议价能力、技能积累、收入稳定性以及劳动者能否向价值链更高位置移动。更重要的是,平台评分与全球竞争会把发展问题个体化:工人被要求通过技能和声誉改善处境,而汇率、地区工资和平台议价权等结构条件退到背景。数字发展政策需要重新把这些结构因素纳入。】
- Karatani, Kojin. The Structure of World History: From Modes of Production to Modes of Exchange. Translated by Michael K. Bourdaghs. Duke University Press, 2014. Originally published 2010. (Read: Chapter 12, "Toward a World Republic," especially the discussion of Mode D, pp. 285–308).
【按:Karatani的Mode D不是一种数字货币方案,而是对历史动力学中exchange的重新理解。他试图突破把资本主义仅仅理解为“生产方式”的方法,把资本、国家和互酬关系放入更长的交换形式历史。放进本节的意义,是让Web3、DAO和平台合作主义面对一个更高的制度问题:如果交换模式本身没有变化,那么仅仅改变技术接口是否足以改变资本-劳动关系?借助Karatani The Structure of World History: From Modes of Production to Modes of Exchange,“自动化叙事如何隐藏微任务、数据标注与维护劳动”能够从修辞性命题转化为制度分析。Mode D的规范力量正在于它拒绝把资本主义替代方案压缩为所有权形式的单一改变。交换、国家与互酬关系彼此嵌套,任何数字合作方案都必须处理公共财政、社会再生产与跨地域协调。Karatani由此使“去中心化组织”面对超出技术治理范围的世界历史尺度。Karatani的交换样式框架还迫使数字左翼重新思考国家问题。资本、国家与民族之间并不是可以单独拆除的一环,市场扩张往往与国家再分配和保护机制共同运作。任何以DAO或合作平台替代现有组织的方案,如果不处理税收、公共服务和跨共同体协调,就很难承担社会再生产的总体功能。这也说明Mode D不能被简化为技术乌托邦。真正困难的是建立能够超越资本-国家互补结构的制度形式,同时承担公共品、再分配和跨地域协调;技术只能是其中一个环节。】
- Ostrom, Elinor. Governing the Commons. Cambridge University Press, 1990. (Read: Chapter 1, pp. 1–28).
【按:Ostrom用经验研究挑战把共有资源治理简单理解为“公地悲剧”。与DAO对读时,重点是规则如何产生、如何被执行、如何修改,以及什么样的参与者结构能够维持共同资源。代码可以实现规则,但不能自动创造信任、参与和责任。合作主义也不是技术万能药。与其追求概念覆盖面的最大化,不如让理论保持可反驳性。只有在经验材料能够支持其核心机制时,这一框架才应继续延伸。Ostrom Governing the Commons最终使“合作社、工会、劳动法与公地治理分别改变哪一层权力”成为一个可以被经验研究继续拆解的问题,而不是一个预先完成的时代诊断。Ostrom的制度分析尤其适合校正“公地=无规则共享”的想象。稳定的共有资源治理依赖边界、参与式规则、监督、渐进制裁与冲突解决机制,且这些机制必须适应具体共同体。数字commons虽然复制成本不同于自然资源,但治理问题同样不能由开放访问本身解决。Ostrom的设计原则不能机械复制到数字资源。软件和数据具有非竞争性复制特征,成员边界也可能跨越全球;与此同时,服务器、维护和治理注意力仍然稀缺。真正可迁移的是她对制度多样性和自组织能力的经验主义,而不是把任何开放项目都直接称为commons。这一点使commons问题回到制度设计:资源是否可复制,并不能取消成员治理、维护义务与冲突解决的必要性。数字commons中的可复制性确实降低了资源耗尽问题,却强化了维护与治理注意力的稀缺。代码审查、服务器费用、社区管理和冲突处理都需要持续投入。Ostrom的经验主义因此提醒我们,共有资源的关键从来不只是“开放”,而是能否建立与资源特征相匹配的长期治理机制。】
第四组:一般智力与技术未来(第7–8周)
第七节:一般智力与对算力集中化的抵抗
本节把Marx的一般智力问题与当代AI基础设施重新连接起来,但这种连接并不建立在“马克思预言了大模型”之上。Grundrisse“机器片段”所处理的是社会知识、固定资本、劳动时间与价值形式之间的矛盾:科学知识和社会协作越来越进入机器体系时,直接劳动与生产力之间的关系发生变化。Virno则把general intellect重新强调为活的、公共的语言-认知能力,指出它并不等于已经凝固在机器中的知识存量。两者之间的张力为理解当代算力集中提供了比“AI是否像人”更有生产力的问题。
生成式AI把社会共同生产的文本、代码、图像、反馈和判断大规模转化为训练资源,同时又依赖芯片、数据中心、能源、资本投入与专业劳动形成高门槛的固定资本。Pasquinelli关于人工智能社会史的分析、Crawford对AI物质与劳动条件的追踪,以及知识资本主义、认知资本主义和数字劳动研究共同揭示:所谓“智能”并不是脱离社会关系的纯粹能力,而是知识、劳动、基础设施与产权安排的综合结果。算力集中因此首先是一个可经验分析的产业和制度问题,而不是一个先验的政治结论。
抵抗的尺度也不能被压缩为“开源”或“人人拥有GPU”。开放权重可能改善访问,却未必改变训练资源、云基础设施、能源成本和治理权;公共算力可以降低进入门槛,却仍需要回答数据、维护、责任与分配问题。Virno的exit、commons和非国家公共性因此提供的不是现成方案,而是一条重新组织一般智力的政治线索:社会共同的认知能力能否拥有相应的制度和基础设施,使模型、数据与计算资源不再只以私人固定资本的形式存在。
必读文献:
- Marx, Karl. Grundrisse: Foundations of the Critique of Political Economy. Translated by Martin Nicolaus. Penguin Books, 1973. Originally written 1857–58; first published 1939–1941. (Read: "Fragment on Machines," pp. 690–712).
【按:必须抵制“马克思预言了AI”这种倒推式阅读。Marx所处理的是social knowledge、general intellect、fixed capital、labour time与value之间的历史矛盾:当科学知识与社会协作越来越进入机器体系时,劳动时间在生产中的位置会发生变化。今天可以把AI看作这一问题的新的历史场景,但不能直接把LLM等同于Marx的“机器体系”。关键是追问:如果社会智力越来越凝结为私人算力基础设施,价值、劳动和所有权之间的矛盾是否以新的形式重新出现?“机器片段”要放回Marx对fixed capital和生产力的分析:机器体系吸收科学知识与社会协作,使劳动者越来越成为生产过程的监督者与调节者,由此出现劳动时间作为财富尺度与生产力社会化之间的张力。general intellect在文本中的位置并不等于后来的autonomist解释,这一差异必须保留下来。迁移到AI时,最有价值的是比较知识进入固定资本的形式:模型、软件、芯片和数据中心是否构成新的固定资本组合?训练数据与活的合作能力又有哪些无法被固定化的部分?它首先是关于社会知识、共同能力与生产关系的理论问题;当代迁移必须解释这些能力如何被固定资本化,而不能把机器参数拟人化成人类共同心智。Marx的“机器片段”首先处理固定资本、社会知识与劳动时间之间的矛盾,而不是预言人工智能。general intellect进入机器体系意味着生产越来越依赖社会积累的科学与合作能力,同时价值形式仍以劳动时间为尺度。今天的基础模型可以作为这一矛盾的新场景,但模型、数据中心和GPU只是具体历史形态,不能反过来定义Marx原有概念。】
- Virno, Paolo. A Grammar of the Multitude: For an Analysis of Contemporary Forms of Life. Translated by Isabella Bertoletti, James Cascaito, and Andrea Casson. Semiotext(e), 2004. Originally published 2001. (Read: Day One, "Common Places and 'General Intellect'"; Day Two, "Intellect as Score" and "Reason of State and Exit"; Day Three, "The Emotional Tonalities of the Multitude").
【按:Virno对Marx “一般智力”的关键推进,在于拒绝把general intellect简单理解为已经凝固在机器、固定资本和技术系统中的“知识总量”。在"Common Places and "General Intellect""与"Intellect as Score"中,一般智力首先表现为社会共同拥有的语言、知识、判断和协作能力,是活劳动的一种普遍条件。这样一来,当代AI的关键问题便不是“机器有没有获得人类智能”,而是社会已经存在的认知能力如何被平台、模型和基础设施重新组织。"Reason of State and Exit"又把这一问题推进到政治层面:如果一般智力被资本吸纳,它为什么没有自动产生解放?Virno对"exit"的思考要求我们寻找一种不以接管既有国家-企业组织为唯一目标的政治路线,而是重新组织公共合作条件。放在今天,这意味着我们不能把“抵抗算力集中”理解为简单复制一个更民主的大科技公司;真正的问题是,社会是否能够重新拥有数据、模型、计算和知识基础设施,使算力服务于general intellect,而不是让general intellect反过来成为算力资本的原料。这里必须把“算力诸众”保留为问题性概念,而不能直接当成Virno本人的术语:它用来检验一般智力、基础设施所有权与自治组织之间是否能够形成新的政治连接。尤其要把"Intellect as Score"的比喻读透:score不是已经完成的表演,而是一套只有在具体行动中才被实现的公共能力。由此general intellect不再只是储存在固定资本里的对象,而成为言说、协作、学习和判断的共同条件。"Reason of State and Exit"则把政治问题推进到国家与非国家公共性的关系:exit不是私人逃跑,也不是技术社群退出监管,而是重新组织合作并使旧关系失去支撑。把这一点带入开源模型、公共算力或federated infrastructures时,必须逐项问这些制度是否真的改变共同能力的组织权,而不是仅仅降低工具价格。Virno把general intellect从固定资本中重新拉回活的语言、判断、习惯与合作能力。一般智力因此既可能被资本组织为生产力,也包含形成非国家公共性的可能。其"exit"并非简单逃离制度,而是重新组织合作条件。放到当代算力问题中,真正的政治焦点不是机器是否拥有智能,而是共同认知能力是否能够建立自己的数据、模型与计算基础设施。】
- Pasquinelli, Matteo. The Eye of the Master: A Social History of Artificial Intelligence. Verso, 2023. (Read: Chapter 4, "The Origins of Marx's General Intellect"; Chapter 7, "The Automation of Pattern Recognition"; Conclusion, "The Automation of General Intelligence").
【按:Pasquinelli的贡献在于把AI放回劳动分工和pattern recognition的历史,而不是将其当成一个脱离社会的计算奇迹。Chapter 4重新追溯Marx的general intellect;Chapter 7讨论模式识别与自动化;结论则把这些历史线索推向general intelligence的自动化问题。对本节而言,它能够把“模型训练”从纯技术操作转成劳动与知识组织的问题:AI自动化究竟自动化的是一个人的技能、一个职业,还是社会已经形成的集体认知能力?全书的社会史线索要求把intelligence与division of labor一起理解:所谓机器智能往往从人类劳动中抽取规则、模式和统计秩序,再以自动化形式重新组织劳动。Chapter 4回到general intellect,不是为了宣布Marx早已解释AI,而是追问知识如何成为生产装置;Chapter 7对pattern recognition的历史化则拆解深度学习“从数据中自己学习”的神话。结论中的general intelligence因而首先是社会-制度问题:哪些一般化能力被形式化、谁决定模式、谁被迫适应机器所识别的模式。Pasquinelli通过劳动分工与模式识别的历史重构AI,使机器学习不再被理解为从纯数学中突然涌现的智能。自动化往往先把社会实践中的判断规则抽取、形式化,再交给机器执行。general intelligence的历史由此与组织、测量和管理史相连。AI所“自动化”的并非抽象心灵,而是已经社会化的认知操作。】
- Hardt, Michael, and Antonio Negri. Multitude: War and Democracy in the Age of Empire. Penguin Press, 2004. (Read: Part 2, selected sections on immaterial labor, biopolitical production, and the common).
【按:Hardt与Negri把multitude、immaterial labor、biopolitical production和the common连接起来,试图说明当生产越来越依赖语言、知识、情感和协作时,政治主体不必再以同质的工业工人形象为唯一前提。放在Marx—Virno之后,它提供一种更明确的政治组织想象,但也必须接受平台劳动研究的经验校正:社会合作被广泛动员并不自动意味着自治主体已经形成,资本仍可以通过产权、平台和基础设施捕获这种合作。若把Hardt Multitude: War and Democracy in the Age of Empire置于本节的整体脉络中,“从理论上的multitude/common走向可持续制度建构的困难”呈现出更具体的结构。一般智力不能直接等同于模型参数、数据集或互联网。它首先指向社会化的语言、知识、判断和协作能力;这些能力可以被固定资本吸纳,却不会因此与机器本身完全重合。Hardt与Negri的multitude强调共同生产能力与差异性主体之间的结合,试图寻找不同于同质人民或传统工人阶级的政治形式。其开放性适合连接数字协作,却也容易被网络隐喻过度美化。真正的问题是分散合作如何获得制度持续性、资源与决策结构,否则“诸众”可能只停留在描述连接的政治修辞。multitude概念与数字网络的亲和性很容易诱发技术乐观主义。共同生产并不会因为网络连接自动产生共同政治,差异主体也不会自然形成稳定意志。其真正难题在于如何建立不抹平差异的决策与资源制度,这一点恰好把诸众理论推向治理问题。】
推荐文献:
- Vercellone, Carlo. "From Formal Subsumption to General Intellect" Historical Materialism 15, no. 1 (2007): 13–36.
【按:Vercellone讨论知识、合作和社会智力进入资本积累过程。这是第六节与第七节之间的桥梁:劳动组织的问题最终会转向谁拥有一般智力及知识commons。理论上的相似并不自动意味着制度上的同一。算力集中是产业结构与基础设施问题。芯片、能源、资本和云服务的规模门槛会塑造谁能够训练、部署和维护模型,因此开放权重并不足以自动消除对私人基础设施的依赖。因此,Vercellone "From Formal Subsumption to General Intellect"对“一般智力、固定资本、认知劳动与算力集中之间的关系”的说明只有在相应制度机制能够被经验材料识别时才成立。Vercellone把general intellect与资本对社会知识的吸纳联系起来,强调价值生产越来越依赖超出工厂边界的合作与知识。认知资本主义论由此把租金、知识产权和社会化生产放到同一矛盾中。它为AI时代提供的不是“知识已经取代劳动”的结论,而是分析资本如何对难以完全封闭的共同知识重新建立稀缺性与控制。Vercellone的论证也使“租金”重新进入知识经济分析。当共同知识难以完全私有化时,资本可能更多依赖专利、平台入口和基础设施所有权获取收益。AI模型的知识封装与API访问可以据此研究,但是否构成rent仍需结合竞争结构、产权与价格机制具体判断。Vercellone对formal/real subsumption与general intellect的连接,还试图解释资本如何从直接组织劳动转向控制社会化知识条件。即使这一历史分期存在争议,它仍提出一个重要问题:当生产依赖广泛社会合作时,资本通过什么制度把分散合作重新转化为私人收益?这一问题在平台与AI基础设施中具有新的经验对象。】
- Moulier Boutang, Yann. Cognitive Capitalism. Translated by Ed Emery. Polity, 2011. Originally published in French as Le capitalisme cognitif in 2007. (Read: Introduction; selected chapters on knowledge, externalities, and cognitive labor).
【按:Moulier Boutang的cognitive capitalism命题把知识、创新、合作网络与正外部性置于价值生产中心,试图解释为什么传统工厂边界越来越难以容纳生产过程。它和general intellect之间关系密切,但不能把二者当同义词:Marx/Virno的问题涉及社会智力与资本关系,而cognitive capitalism是对特定资本主义阶段的经验-理论命名。两者对读能检验“知识成为主要生产力”究竟意味着什么。算力集中是产业结构与基础设施问题。Moulier Boutang Cognitive Capitalism所提供的并不是对“社会共同能力如何被产权、平台和基础设施重新组织”的终局命名,而是一种可被竞争性解释检验的理论切口,这种开放性本身构成其研究价值。Moulier Boutang以cognitive capitalism描述知识、创新与正外部性在积累中的突出地位,并把传统工厂之外的合作纳入价值分析。其“认知”概念不能遮蔽物质生产,但能够解释为何网络、专利、教育与知识流动成为资本竞争的关键条件。与AI结合时,应同时看到模型的认知资源与支撑这些资源的能源、硬件和劳动。认知资本主义还必须面对劳动测量问题。知识生产高度依赖协作、教育和社会交往,单个劳动者的贡献越来越难以分离;资本却仍通过工资、知识产权和企业边界分配收入。模型训练进一步放大这种矛盾,因为训练材料来自广泛社会生产,而收益集中于少数基础设施所有者。】
- Gorz, André. The Immaterial: Knowledge, Value and Capital. Translated by Chris Turner. Seagull Books, 2010. Originally published in French as L'Immatériel in 2003. (Read: selected chapters on knowledge, value, and the crisis of measurability).
【按:Gorz关注知识、能力和活的社会经验难以被传统劳动时间完整度量,却又不断被资本转化为可占有资源的矛盾。它非常适合放在general intellect谱系中,因为它把“知识”从抽象信息重新带回主体能力与社会关系,同时提出价值尺度危机。迁移到AI时,应问模型训练究竟是把知识本身商品化,还是把知识的表现、数据化痕迹和访问权转化为资产。“开放、互操作、公共所有与自治治理不是同一个解决方案”通过Gorz The Immaterial: Knowledge, Value and Capital进入了组织与制度层面。算力集中是产业结构与基础设施问题。Gorz关于知识价值的论述把认知资本主义中的核心矛盾放在可复制知识与稀缺性制造之间。数字知识越容易复制,资本越需要通过知识产权、平台封闭和服务化重新制造排他性。基础模型与API经济使这一矛盾获得新的形式:公共知识被吸收后,又以受控访问重新进入市场。Gorz对知识不可度量性的强调,使价值问题不再能够完全以单个劳动时间解释。知识往往来自长期教育、公共研究和社会交往,复制成本又趋近于零;资本必须通过专利、平台封闭和品牌重新制造稀缺。生成式AI对公共文本的大规模吸收,使这种共同知识与私有访问权之间的矛盾更加可见。Gorz的论述还可以用于理解模型访问权:资本未必拥有知识内容本身,也可以通过控制算力、接口和服务条件,把对共同知识的访问重新变成稀缺商品。占有由内容产权转向基础设施产权。】
- Roggero, Gigi. The Production of Living Knowledge: The Crisis of the University and the Transformation of Labor in Europe and North America. Translated by Enda Brophy. Temple University Press, 2011. (Read: Introduction; selected chapters on living knowledge, the university, and cognitive labor).
【按:Roggero以大学与知识劳动为中心讨论living knowledge的生产,强调认知劳动并不等于抽象的“创意阶层”,而是在教育制度、劳动市场、合作网络和冲突中形成。它能把general intellect从宏观哲学命题拉回大学、研究、培训与知识产权制度,特别适合追问生成式AI进入教育与研究后,知识共同生产与制度性占有之间的矛盾。Roggero The Production of Living Knowledge: The Crisis of the University and the Transformation of Labor in Europe and North America所揭示的“从理论上的multitude/common走向可持续制度建构的困难”并不止于文本内部的概念争论。算力集中是产业结构与基础设施问题。Roggero把"living knowledge"放入大学危机与劳动转型中,强调知识生产并非只发生在正式科研机构,而在社会合作、学习与网络中持续生成。大学既生产公共知识,也参与将知识商品化和计量化。基础模型大量吸收学术与公共文本后,这一矛盾更加突出:社会知识的开放性与私人计算基础设施之间形成新的占有关系。大学在Roggero那里并非知识资本主义之外的纯粹公共空间,而是劳动转型的实验场:研究、教学和学习越来越接受绩效计量、项目制与竞争性融资。把这一视角连接AI,可以观察学术知识如何一方面以开放文本供模型吸收,另一方面又被出版、平台和计算资源重新分层。】
- Beller, Jonathan. The Message Is Murder: Substrates of Computational Capital. Pluto Press, 2018. (Read: Introduction; selected chapters on computational capital, attention, and racialized extraction).
【按:Beller把computation放进资本、视觉文化、金融化和种族化暴力的历史,强调计算并非只有机器内部的运算,而嵌入更广泛的社会价值抽取。它能与general intellect形成必要张力:如果我们只把社会知识理解成可解放的共同能力,可能忽略这种能力的生产本身就处在差异化暴力、注意力捕获和全球分工之中。在Beller The Message Is Murder: Substrates of Computational Capital的分析尺度上,“一般智力、固定资本、认知劳动与算力集中之间的关系”首先表现为一组具体关系。Beller把计算资本主义理解为感知、注意与基础设施共同参与价值化的体系,由此扩展了数字劳动的边界。其强势论断需要经验材料约束,但它抓住了一个关键问题:计算并非中性的后台工具,而可能成为资本组织视觉、时间和社会关系的普遍基底。Beller的substrate概念能够把视觉文化、计算与物质基础连接起来。屏幕上的注意活动只有在芯片、网络、能源和劳动系统中才可能被资本化。由此,“计算资本”既不是纯粹的信息经济,也不是传统工业资本的简单延伸,而是多层基础设施共同组织价值捕获的形式。计算的“无形性”由此被重新物质化:注意力、图像和数据的价值化始终依赖底层设备、网络和劳动。计算资本的分析因此需要在感知政治和物质基础设施之间保持双重视野。两者缺一,都会把计算资本重新抽象化。】
- Tronti, Mario. Workers and Capital. Translated by David Broder. Verso, 2019. Originally published 1966. (Read: "Lenin in England," pp. 78–88).
【按:Tronti的意义在于方法论:不是先问资本想要什么,而是先分析工人斗争如何迫使资本改变。用于今天的平台经济时,应把它作为一种研究方法,而非简单证明“AI是工人斗争逼出来的”。Tronti Workers and Capital与Beller The Message Is Murder: Substrates of Computational Capital和Doctorow The Internet Con: How to Seize the Means of Computation把“社会共同能力如何被产权、平台和基础设施重新组织”置于不同的知识传统中。Tronti Workers and Capital让“社会共同能力如何被产权、平台和基础设施重新组织”摆脱了过于抽象的时代判断。Tronti Workers and Capital对“社会共同能力如何被产权、平台和基础设施重新组织”的贡献,正在于让理论保持可反驳性。Tronti的方法论提醒我们,技术变迁不能只从资本的创新战略解释。工人拒绝、流动、组织与生产冲突同样能够迫使资本寻找新的管理技术和劳动安排。把这一思路用于自动化,应追踪具体斗争与投资决策之间的历史关联,而不能把自动化本身浪漫化为工人斗争的直接成果。Tronti的workerism还强调政治知识来自斗争中的调查,而不是先验地替劳动者规定利益。用于平台和AI劳动时,这意味着需要从司机、标注员、审核员和工程师的具体冲突出发,理解技术如何被工作场所重新解释。理论不能把劳动者仅仅设定为技术变化的被动承受者。】
- Doctorow, Cory. The Internet Con: How to Seize the Means of Computation. Verso, 2023. (Read: Introduction; chapters on interoperability, switching costs, and antitrust).
【按:Doctorow把“夺取计算资料”落到interoperability、adversarial compatibility、反垄断与退出权等具体制度工具,而不是停留在所有权口号。它为本节提供极有价值的中层政策层:如果general intellect被平台基础设施围堵,抵抗是否必须先降低锁定效应、恢复互操作与迁移能力?同时也应问,这些改革能否触及数据中心、算力和资本所有权,还是主要改善市场竞争。围绕Doctorow The Internet Con: How to Seize the Means of Computation展开时,“开放、互操作、公共所有与自治治理不是同一个解决方案”可以被重新拆分为若干相互关联的环节。Doctorow所说的“夺取计算资料”把竞争政策、互操作性和用户权利重新放到基础设施政治中心。技术开放若不能改变退出成本、数据迁移和市场控制,往往只产生表面的选择。其改革主义路径与更激进的公共所有或合作治理方案可以形成有生产力的比较。互操作性在Doctorow这里具有明确的权力含义:如果用户可以带走数据、社交关系与内容,并让不同服务互联,平台锁定就会下降。与此同时,互操作也涉及安全、隐私和标准治理的复杂成本。因而“打开接口”不是无条件答案,而是一种需要公共规则支持的竞争与自治机制。互操作政策的政治意义也正在这里:它把竞争从“多几个平台”推进到用户是否真正拥有退出与迁移的制度能力。】
- Tarnoff, Ben. Internet for the People: The Fight for Our Digital Future. Verso, 2022. (Read: Introduction; chapters on privatization, public ownership, and democratic control).
【按:Tarnoff把互联网私有化的历史与公共控制问题连接起来,反对把今日平台权力只归因于少数坏企业或坏算法。他的方案更接近基础设施政治:要讨论所有权、公共投资、民主治理和非商品化。与Doctorow对读可以形成改革尺度之争:互操作与竞争政策是否足够,还是需要把互联网的某些层重新理解为公共基础设施。Tarnoff Internet for the People: The Fight for Our Digital Future与Doctorow The Internet Con: How to Seize the Means of Computation和Schneider Governable Spaces: Democratic Design for Online Life共同使“从理论上的multitude/common走向可持续制度建构的困难”从宏观判断下沉到机制层。Tarnoff Internet for the People: The Fight for Our Digital Future对“从理论上的multitude/common走向可持续制度建构的困难”的贡献,在于把宏观判断压回可识别的制度机制。Tarnoff把互联网私有化视为一个历史过程,从而打破“网络天然属于市场”的常识。公共资助、商业化和广告模式共同塑造今天的平台结构,因此去商业化也必须涉及融资机制与公共制度,而不只是呼吁企业改善伦理。互联网公共性由此成为所有权与制度设计的问题。】
- Schneider, Nathan. Governable Spaces: Democratic Design for Online Life. University of California Press, 2024. (Read: Introduction; selected chapters on governance, ownership, and democratic design).
【按:Schneider从online communities的治理实践出发,强调数字空间并非天然无政府或只能由公司管理;规则、成员资格、所有权和决策程序都可以被重新设计。它与Virno的exit和Kelty的recursive public有远距离呼应,但这里更侧重制度建构:所谓从平台权力中“退出”,如果没有可持续的治理结构,很容易只是迁移到另一个私人平台。“一般智力、固定资本、认知劳动与算力集中之间的关系”从Schneider Governable Spaces: Democratic Design for Online Life出发,可以获得更清楚的因果层次。Schneider把线上空间理解为可以被宪制化的治理空间,重点不在抽象地赞美民主,而在规则如何形成、成员如何获得代表、权力如何受到约束。数字共同体的民主化由此需要可执行的制度程序,而不是把投票功能简单嵌入平台界面。governable spaces的核心不只是建立社区规则,而是让成员获得修改规则的制度位置。平台民主因此不能等同于意见调查或投票插件;如果服务器、代码和最终执行权始终掌握在外部所有者手中,参与式程序仍可能缺乏实质约束力。数字民主的实质标准因此不是平台是否提供参与按钮,而是成员能否对规则、基础设施与执行权形成可持续约束。治理权只有能够约束基础设施所有者时,数字参与才从咨询机制转化为制度权力。民主设计最终必须能够触及执行权。】
- Söderberg, Johan. Hacking Capitalism: The Free and Open Source Software Movement. Routledge, 2008. (Read: Introduction; selected chapters on free software, labor, and political economy).
【按:Söderberg从劳动与政治经济学角度解释自由/开源软件,既看到协作生产超出传统企业边界的潜力,也分析资本如何重新吸纳免费贡献。它可以用来检验一个贯穿本节的问题:开放代码是否等于共同所有?当云算力、部署渠道和数据仍被集中控制时,软件commons的政治效力会在哪里被削弱。自由软件运动显示,对技术生产资料的控制可以通过许可证、协作规范和共享代码进行制度化。Söderberg同时指出,这种实践既可能挑战产权,也可能被资本吸纳为低成本创新模式。开放因此不是资本外部的纯净空间,而是一种持续发生争夺的生产关系。开源运动还展示了一种非国家的制度创新:copyleft通过法律工具反向保障共享,使公共资源不易被再次私有封闭。它说明“退出资本”并不一定意味着退出法律;相反,自治实践经常需要利用既有法律建立可持续边界。这一点对数据和模型commons具有直接启发。开源实践由此提示,commons并不意味着取消制度边界;恰当的法律设计反而可能成为维护公共生产的条件。对公共模型和开放数据而言,许可证与治理机制同样可能承担这种“防再封闭”的作用。开放的持续性本身就是制度成果。自由软件运动还显示,开放生产可以形成自己的物质和制度依赖。项目需要托管、维护、资金和贡献者治理,一旦这些条件重新集中于少数企业,代码开放也可能与基础设施依赖并存。对开放模型而言,同样需要区分权重开放、训练数据、算力与部署平台分别处于何种产权关系。】
- Lazzarato, Maurizio. "Immaterial Labor" In Radical Thought in Italy: A Potential Politics, edited by Paolo Virno and Michael Hardt, 133–147. University of Minnesota Press, 1996. Originally published in Italian in 1994. (Read: entire chapter).
【按:Lazzarato的immaterial labor概念试图描述信息、文化内容、沟通与主体性生产在后福特主义中的重要性。它对general intellect谱系影响深远,但今天必须以双重方式阅读:一方面它捕捉到生产向社会沟通扩散的趋势;另一方面“非物质”一词容易遮蔽芯片、矿产、物流、数据标注和身体劳动。与第六节的劳动研究对读,正好能暴露这一限度。借助Lazzarato "Immaterial Labor",“开放、互操作、公共所有与自治治理不是同一个解决方案”能够从修辞性命题转化为制度分析。immaterial labor概念抓住了语言、情感、知识与合作进入价值生产的趋势,但其边界始终需要谨慎处理。认知与沟通活动并不会因此脱离物质劳动,反而依赖设备制造、物流、能源与再生产劳动。把两者重新接合,才能避免用“非物质”遮蔽数字经济的物质基础。Lazzarato的immaterial labor概念历史上连接了后福特主义、传播与主体生产,强调劳动越来越直接生产文化内容和社会关系。其问题在于“非物质”容易遮蔽设备、物流与身体劳动。更合适的读法是把它视为劳动内容与控制方式的变化,而不是宣布物质生产已经退场。这样才能与当代数字劳动的多层结构相接。由此,immaterial labor最有价值的地方不是名称,而是它指出价值生产越来越依赖沟通、符号和主体关系。数字资本主义的问题在于这些活动与物质基础如何被重新组合,而非哪一方取代另一方。】
- Dyer-Witheford, Nick, Atle Mikkola Kjøsen, and James Steinhoff. Inhuman Power: Artificial Intelligence and the Future of Capitalism. Pluto Press, 2019. (Read: Introduction; selected chapters on AI, labor, and capital).
【按:本书从Marxist political economy讨论AI的自动化与资本关系,拒绝只以“未来机器是否有意识”来界定AI问题。它重点考察机器学习进入生产、管理与劳动市场后如何重组资本的控制能力。由此可以把general intellect与现实AI政治经济重新接上,同时也为下一节的AI哲学与未来想象设置经验边界。Inhuman Power把AI放入资本主义自动化的历史逻辑,而不是当作独立的技术主体。机器学习扩大了资本对预测、协调和监督的能力,同时也依赖既有的劳动分工与产权结构。所谓“非人力量”因此首先是一种被社会关系组织起来的技术能力。Inhuman Power还把AI与资本的“自主性”问题并置:机器系统看似获得行动能力,实际上其目标、部署和扩张仍受资本积累逻辑塑造。技术自主与资本自主并不是同一概念。区分二者有助于避免把由企业作出的投资和管理决策拟人化为“AI自己要求加速”。这一区分能够把“机器自主”重新还原为组织问题:技术能力的扩张并不会取消资本对部署方向与收益分配的决定。AI政治经济由此必须区分技术系统的能力增长与资本组织形式赋予它的社会方向。Inhuman Power的资本主义框架也能校正“AI即自主行动者”的哲学想象。算法系统的目标并不是从机器内部自然产生,而由投资、企业战略、竞争和管理需求赋予。机器能力可以表现出超出个体设计者预期的复杂性,但其社会部署仍有明确制度主体。政治责任不能因为系统复杂而被拟人化转移。】
第八节:人工智能、自动化与后人类未来
最后一节把AI置于哲学、工程、政治经济学与未来想象的交叉地带。Turing关于机器智能的经典提问、Russell对人机目标关系的讨论、Bostrom的超级智能风险框架,以及Brassier、Land、Negarestani对理性、技术和加速的不同思考,并不属于同一种论证类型。它们分别涉及操作性测试、工程控制、思想实验、规范哲学与形而上学推演,因而不能用同一证据标准相互替代。AGI也不应被当作一个已经拥有稳定定义和时间表的经验对象,而应被拆分为能力、自治性、通用性、部署条件和制度权力等不同问题。
Haraway、Hayles、Braidotti与Hui则从后人类主义、身体性、关系性与技术多元性方向重新提出“人”的边界。后人类并不等于增强人类或上传意识,也不必然导向技术乐观主义;它首先挑战把理性、主体性和政治资格固定在人类个体内部的传统想象。与此相对,Crawford、Bender/Hanna、Narayanan/Kapoor与Hao不断把宏大的智能叙事拉回数据、劳动、能源、企业组织、产业竞争和市场宣传,使“未来”重新进入现实制度。
本节因此关心未来叙事本身如何成为政治力量。产业路线图、风险话语、加速主义和解放性技术想象都会提前规定什么被视为可行、必要或不可避免。对AI的批判既不能把所有能力归结为hype,也不能因为某些系统确有显著能力就接受产业对“智能”的统一命名。现实能力、推测性风险、政治经济结构与规范性未来必须分别论证;只有在这些层次之间建立清楚的关联,关于自动化与后人类的讨论才可能超越技术崇拜与技术恐惧之间的二元摆动。
必读文献:
- Crawford, Kate. Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press, 2021. (Read: Introduction; Chapter 2, "Labor"; Chapter 3, "Data").
【按:Crawford的重要作用是把general intellect重新放回行星物质性。AI依赖的不只是算法和模型,还包括矿产、能源、数据中心、标注劳动、内容审核和全球供应链。Chapter 2让“AI的劳动”重新显形,Chapter 3则说明数据如何成为资源。与Marx和Virno对读,可以提出一个更具体的问题:如果AI把社会智力私有化,它同时也把大量物质劳动与生态成本隐藏在模型背后,那么所谓“智能”为什么总是以一种去身体化的形式出现?Crawford在这里承担的是AI现实条件的校正,而不是一般智力概念的直接延伸。Atlas of AI通过Earth、Labor、Data等尺度把AI视为extraction的产业系统:矿物、能源和空间是计算前提,标注与维护是自动化叙事隐藏的劳动,数据集又把分类历史带入模型。与Hao的组织调查和Bender/Hanna的hype critique对读,可以把“AI是什么”从抽象智能问题改写成技术能力、产业基础与制度权力的复合对象。终章必须区分现实能力判断与哲学可能性。某一系统今天做不到什么,并不能逻辑证明机器永远不可能做到;反之,一个概念上可能的超级智能也不能自动证明当前产业部署的政治优先级。Crawford的atlas方法把AI拆成矿产、能源、劳动、数据、分类和国家权力的行星级装配。所谓“人工智能”因此不是悬浮的软件能力,而是高度物质化、制度化的系统。将劳动者、矿场和数据中心重新纳入分析,可以改变风险讨论的时间尺度:AI的政治后果并非只在未来超级智能出现时才发生,而已经存在于资源开采、工作组织与分类制度中。】
- Bender, Emily M., and Alex Hanna. The AI Con: How to Fight Big Tech's Hype and Create the Future We Want. Harper, 2025. (Read: Introduction; selected chapters on AI hype, synthetic media, labor, and institutional power).
【按:Bender与Hanna把所谓"AI"首先当作产业话语、制度选择和具体技术系统来拆解,特别反对把机器学习产品拟人化为具有统一意志的智能主体。它们关心的是hype如何转移注意力:社会开始讨论遥远的机器统治,却可能忽略数据、劳动、资源、商业模式和当前已经发生的自动化伤害。作为2025年文本,它适合把讨论拉回生成式AI之后的新一轮技术叙事,并与Crawford的物质政治形成互补。"AI"是高度异质的总称。预测模型、生成模型、机器人、优化系统与假设中的AGI不应共享同一能力判断;能力判断必须明确对应具体技术对象,而不能让"AI"这个总称替代分类。对AI产业话语的批判必须始终对应具体技术对象、组织制度与可观察后果,避免以另一套宏大拟人化叙事取代技术神话。Bender与Hanna把"AI"作为产业叙事和权力装置来拆解,其批判重点不止在模型局限,还在夸大能力如何服务投资、市场集中与政策议程。与技术风险论相比,这一视角把注意力转向当下已经存在的制度后果:谁有权定义AI、谁从神话中获益,以及谁承担部署成本。产业hype之所以具有政治效果,是因为它能够改变资源分配。关于通用智能和自动化替代的宏大叙事会影响投资、人才、监管优先级和公众注意力,使某些当前问题被边缘化。批判hype并非否认技术变化,而是要求能力声明与社会决策之间建立更严格的证据关系。因此,去神话化不是降低技术的重要性,而是把技术判断重新放回可检验的能力、制度和资源关系中。】
- Narayanan, Arvind, and Sayash Kapoor. AI Snake Oil: What Artificial Intelligence Can Do, What It Can't, and How to Tell the Difference. Princeton University Press, 2024. (Read: Chapter 1, "Introduction"; Chapter 2, "How Predictive AI Goes Wrong"; Chapter 4, "The Long Road to Generative AI").
【按:Narayanan与Kapoor提供技术认识论上的“去魔法化”。它们区分predictive AI、generative AI与更广泛的AI话语,强调能力边界和社会部署条件。对于本节而言,这一校正非常重要:批判理论必须能够区分真正发生的技术能力与产业宣传中的"AI"概念。Narayanan AI Snake Oil: What Artificial Intelligence Can Do, What It Can't, and How to Tell the Difference由此为“资源、劳动、生态与组织史如何约束AGI未来主义”提供了一种有限而有效的解释尺度:它能够指出机制,但并不替代对组织、法律与物质条件的进一步分析。Narayanan与Kapoor区分predictive AI、generative AI以及围绕"AI"形成的夸张叙事,强调不同技术对象需要不同证据标准。预测失败、生成模型能力与AGI想象不能互相替代。技术批判若把所有系统都视为同一种智能,反而会复制产业营销的模糊性。去神秘化因此首先是一项分类工作。AI Snake Oil的重要方法是要求能力主张与使用场景一一对应。某一模型在基准测试中的表现不能自动证明它适合招聘、教育或医疗决策;同样,生成文本流畅也不能证明事实可靠。技术评估必须从抽象能力转向部署条件、错误成本和可替代方案。这种区分尤其适合读书会最后阶段,因为它能防止哲学未来主义与现实系统能力在同一个"AI"概念中相互替代。】
- Hao, Karen. Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI. Penguin Press, 2025. (Read: Introduction; selected chapters on OpenAI's scaling strategy, data labor, compute infrastructure, and global environmental impacts).
【按:Hao以长期调查报道重建OpenAI及其所处的AI竞赛,把模型扩张与算力、资本、数据劳动、全球供应链和环境冲突连接起来。它不是一本抽象AI哲学,而是用组织史与全球案例展示“规模化”如何成为产业战略。放在终章required的意义,是要求所有关于AGI、超级智能或后人类未来的哲学推演都必须经过现实组织与资源政治的检验:谁在建设这些系统、谁承担成本、谁能够定义未来。后人类批判也不能变成对物质政治的逃避。质疑人类中心主义很重要,但矿产、劳动、能源、生态成本与知识产权仍然由具体制度分配;关系性主体理论必须与这些结构并读。若把Hao Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI置于本节的整体脉络中,“技术未来如何在风险治理、解放想象与民主制度之间被争夺”呈现出更具体的结构。AI首先是一个高度异质的对象集合。预测模型、生成模型、机器人、优化系统与假设中的AGI具有不同能力与风险结构,把它们放在同一“智能”概念下会迅速模糊经验判断。Hao通过OpenAI的发展史把AGI理想、资本融资、组织治理与算力竞赛放到同一叙事中。其价值不只是提供公司内幕,而是展示“造福全人类”的规范愿景如何在商业扩张、人才竞争与基础设施投资中遭遇制度压力。AI未来主义由此不是脱离组织的观念史,而与公司治理和资本结构紧密相连。公司史还提示,AI治理不能只依赖企业内部使命声明。组织结构、董事会权力、投资者关系和计算资源都会改变使命能够约束商业决策的程度。公共利益若只以企业自律形式存在,就容易在规模化竞争中被重新解释。】
推荐文献:
- Turing, Alan M. "Computing Machinery and Intelligence" Mind 59, no. 236 (1950): 433–460. (Read: entire article).
【按:Turing的文章最重要的不是一句“机器能思考吗”,而是他主动改写问题:用imitation game、反驳清单与学习机器讨论什么样的行为标准可以让机器智能成为可操作的研究对象。把它放在2020年代AI话语之后重读,可以看到“智能”从来不是透明自然属性,而是由测试、任务和制度期待定义的。终章必须区分现实能力判断与哲学可能性。理论上的相似并不自动意味着制度上的同一。Turing的论文更适合被理解为关于“如何讨论机器智能”的方法论转向,而不是AGI路线图。imitation game暂时搁置了“思维本质是什么”的形而上争论,改用可操作的行为标准。这种策略既开启了计算智能研究,也留下一个持续问题:可操作测试能够测量表现,却不能自动回答理解、主体性或社会责任。Turing还预先讨论了多种反对机器智能的论证,并逐一展示其逻辑弱点。重要的不是他“证明机器会思考”,而是把争论从本质定义转向可检验问题。今天重新阅读时,也应保留这种方法论节制:模型通过某项测试,只能支持关于该测试能力的判断。这种节制也为今天提供方法论标准:行为表现可以成为证据,但不应被迅速扩张为关于意识、理解或主体性的结论。Turing在此提供的是讨论“智能”时的概念纪律,而非一份关于技术未来的预言。Turing的操作主义同时留下了一个今天依然尖锐的问题:当测试越来越多样,系统可以在某些任务上表现优异,却在另一些任务上缺乏稳健性。“智能”是否应继续作为统一总称,还是应拆分为不同能力?这一问题直接关系到生成式AI能力声明的尺度与证据标准。】
- Russell, Stuart. Human Compatible: Artificial Intelligence and the Problem of Control. Viking, 2019. (Read: Introduction; selected chapters on objectives, uncertainty, and the control problem).
【按:Russell从AI研究内部提出control/alignment问题,核心并不是预言某个确定的超级智能灾难,而是批判把机器目标预先写死为单一效用函数的设计假设。他主张让机器对人类偏好保持不确定。与批判理论文本对读时,应进一步追问:"human preferences" 如何形成、谁有权代表它们,以及社会冲突能否被还原为偏好学习。"AI"是高度异质的总称。Russell的value alignment问题把先进AI风险表述为目标设定与不确定性管理:如果机器把人类给定目标当作完全确定的效用函数,能力越强反而可能放大目标误设的后果。这个框架值得与社会科学对读,因为"human preferences"本身并非统一、透明或静态。对齐因此不仅是控制工程,也触及价值冲突、民主授权与制度代表。alignment问题进一步暴露“人类价值”并不是一个可以直接输入系统的统一对象。不同群体之间存在真实冲突,价值也会随情境变化。若把对齐仅处理为偏好学习,民主政治中的争论、权利和不可交易原则就可能被压缩为可优化变量。技术控制问题因此必然通向政治哲学。Russell的控制问题若进入现实制度,还必须处理多主体冲突。一个系统服务于医院、公司、国家或个人时,究竟以谁的偏好为准?不同价值之间不能总被压缩成同一目标函数。由此,对齐研究需要与程序正义、权利限制和民主决策相接,否则"human compatible"可能默认一个并不存在的统一人类意志。】
- Haraway, Donna. "A Cyborg Manifesto" In Simians, Cyborgs, and Women, 149–181. Routledge, 1991.
【按:Haraway的cyborg并不是简单的人机融合预言,而是一种拆解边界的政治主体理论。它让我们重新思考“人”是否必须以纯粹的生物身体和统一身份为政治基础,也为诸众提供一种不以同质性为前提的联盟想象。“资源、劳动、生态与组织史如何约束AGI未来主义”通过Haraway "A Cyborg Manifesto"进入了组织与制度层面。Haraway的cyborg政治拒绝以纯粹、统一和自然化的身份作为联盟基础。人/机、自然/文化等边界被技术实践不断重写,政治主体也由异质关系构成。它与当代AI讨论的关联不在“人机融合”预言,而在于如何在没有本质主义人性假设的情况下重新组织责任与团结。Haraway的cyborg同时是一种反本质主义联盟政治。她拒绝以纯粹女性、纯粹人类或自然身份作为政治共同体基础,而强调在技术、身体与权力交织中形成部分连接。这对AI时代的意义不在“人人成为半机器”,而在于政治主体本来就由技术和社会关系共同构成。责任与团结因此不能建立在恢复一个技术之前的纯粹人类之上。cyborg同时破坏了“技术是外部工具”的想象。身体、医疗、传播和劳动早已通过技术装置被重新构造,人类从来不是在技术之外独立存在的主体。AI治理中的“保护人类”因此必须具体说明保护何种能力、关系与生活条件,而不能诉诸抽象纯粹的人性。Haraway的意义正在于把这种混合性转化为政治资源,而不是把它理解为技术造成的身份危机。】
- Hayles, N. Katherine. Bacteria to AI: Human Futures with Our Nonhuman Symbionts. University of Chicago Press, 2025. (Read: Chapters 1–2, 6–7, and 9–10).
【按:Hayles在这本2025年著作中提出integrated cognitive framework,把认知与意义生产从人类扩展到生命体与部分人工系统,同时持续强调materiality、embodiment与relation。它比How We Became Posthuman更适合的当代AI讨论,因为它直接面对GPT-era的机器认知主张,却不简单把语言模型等同于人类心智。其价值在于提供介于强人类中心主义与“机器已经成为主体”之间的第三条理论路线。后人类批判也不能变成对物质政治的逃避。Hayles Bacteria to AI: Human Futures with Our Nonhuman Symbionts所揭示的“技术未来如何在风险治理、解放想象与民主制度之间被争夺”并不止于文本内部的概念争论。Hayles把认知分布到人类、非人生命与技术系统的关系网络中,使“智能”不再等同于孤立主体内部的能力。2025年的这一论述尤其适合连接生态与AI:技术系统既嵌入生物共生关系,也改变人类对认知边界的理解。后人类未来因此首先是关系重组,而非单一物种的升级。Hayles进一步把认知理解为跨尺度过程,从细菌到技术系统都可能以不同方式感知并响应环境。这样的认知多元主义既反对把人类理性设为唯一尺度,也不意味着所有信息处理都等同于人类智能。其关键贡献是提供一套区分不同认知形式的语言,使后人类讨论不必在“人类例外”与“机器拟人化”之间摇摆。】
- Bostrom, Nick. Superintelligence: Paths, Dangers, Strategies. Oxford University Press, 2014. (Read: Chapter 8, "Is the Default Outcome Doom?", pp. 115–142).
【按:Bostrom提供与加速主义哲学完全不同的存在性风险框架。这里无须预设接受"doom"设定,关键在于分析其论证结构:当我们假设一个高度自主、目标导向的智能体时,为什么对齐问题会成为中心问题?再与政治经济学文本比较,可以讨论“AI风险”是否可能被抽象成一个脱离产权、劳动和制度背景的纯技术问题。终章必须区分现实能力判断与哲学可能性。在Bostrom Superintelligence: Paths, Dangers, Strategies的分析尺度上,“AI能力、产业hype、哲学智能概念与政治经济现实的区分”首先表现为一组具体关系。技术未来并不是纯粹的预测对象。产业路线图、风险叙事和加速主义都会影响投资、监管与公共想象,因此“未来会发生什么”与“谁有能力规定未来”必须同时讨论。 “AI能力、产业hype、哲学智能概念与政治经济现实的区分”经由Bostrom Superintelligence: Paths, Dangers, Strategies获得了一个更严格的问题形式:哪些机制真正发生变化,哪些只是媒介转换,以及这些变化由谁承受其后果。Bostrom的superintelligence论证以极端能力增长和目标保持为前提,由此推导existential risk与control problem。其价值在于清楚展示一种未来风险推理的结构,限制则在于高度依赖关于智能体、目标和技术路径的假设。把它与当下的政治经济材料并置,可以区分远期情景分析与已经可观察的基础设施权力。】
- Brassier, Ray. "Prometheanism and Its Critics" In #Accelerate: The Accelerationist Reader, edited by Robin Mackay and Armen Avanessian, 467–488. Urbanomic, 2014.
【按:Brassier的价值不在于提供一套“AI发展路线”,而在于他把技术、理性和政治解放之间的关系问题化。阅读重点是他为什么拒绝把自然性、本真性和有机生命自动视为进步的尺度。将其置于这一问题中心时,能够迫使这一讨论区分“反技术”与“反资本主义”:如果技术本身可以被重新组织,拒绝技术是否真的能够实现解放?同时需要检验Prometheanism的危险:一种普遍理性的政治是否可能再次变成对差异和生态条件的支配?Brassier "Prometheanism and Its Critics"与Bostrom Superintelligence: Paths, Dangers, Strategies和Land "Teleoplexy: Notes on Acceleration"把“自动化、控制、对齐与后人类主体之间不同问题类型”置于不同的知识传统中。"AI"是高度异质的总称。Brassier "Prometheanism and Its Critics"让“自动化、控制、对齐与后人类主体之间不同问题类型”摆脱了过于抽象的时代判断。技术未来并不是纯粹的预测对象。Brassier "Prometheanism and Its Critics"对“自动化、控制、对齐与后人类主体之间不同问题类型”的贡献,正在于让理论保持可反驳性。Brassier的Prometheanism把技术性改造与理性自我修正联系起来,反对把任何超越既有自然或传统的实践都视为支配冲动。其论证能够挑战技术悲观主义,但也必须回答谁拥有改造世界的能力、何种制度决定技术方向。普罗米修斯主义若脱离政治经济,很容易把不平等的现实技术能力抽象成普遍“人类”的能力。】
- Land, Nick. "Teleoplexy: Notes on Acceleration" In #Accelerate: The Accelerationist Reader, edited by Robin Mackay and Armen Avanessian, 511–520. Urbanomic, 2014.
【按:Land是这一本节中最重要的反向极点。他把资本、技术和自我加速描述为一种超出人类意志的动力过程。阅读价值不在于接受其极端结论,而在于观察:当资本被想象成一个自动加速的系统,人类政治主体还剩下什么位置?与Brassier对读,可以把左右加速主义之间关于理性、资本和未来的分歧显现出来。围绕Land "Teleoplexy: Notes on Acceleration"展开时,“资源、劳动、生态与组织史如何约束AGI未来主义”可以被重新拆分为若干相互关联的环节。技术未来并不是纯粹的预测对象。 从Land "Teleoplexy: Notes on Acceleration"来看,“资源、劳动、生态与组织史如何约束AGI未来主义”不能脱离制度授权与资源分配单独成立;技术形式只有与组织权力结合,才会产生稳定的社会后果。Land把资本与技术加速描述为超出人类政治控制的自组织过程,由此把加速主义推向反人类主义极端。作为反向文本,它最有价值之处是暴露一种把资本动力自然化的思想逻辑:一旦加速被视为不可阻挡,政治就只剩适应或退出。对这一立场的批判必须重新证明制度、劳动与集体行动仍能改变技术轨迹。Land的文本也可以作为检验“不可避免性”修辞的极端案例。资本加速被写成超越政治选择的过程时,制度安排和集体行动都被预先宣布无效。批判这种思路,不是简单主张减速,而是重新区分技术速度、资本竞争与社会目标,证明它们并不存在必然一致的方向。】
- Negarestani, Reza. Intelligence and Spirit. Urbanomic, 2018. (Read: Part I, especially Chapter 1, pp. 1–52).
【按:Negarestani将intelligence从“人脑能否被机器复制”的问题转向概念、推理、自我修正与规范性的问题。它不是技术路线图,而是一种speculative philosophy。置于这一问题脉络中,可以迫使批判理论面对一个极难的问题:如果intelligence不应被定义为人类的自然属性,那么所谓“人工”智能与“自然”智能之间的政治差异应该怎样理解?Negarestani Intelligence and Spirit与Land "Teleoplexy: Notes on Acceleration"和Hui Recursivity and Contingency共同使“技术未来如何在风险治理、解放想象与民主制度之间被争夺”从宏观判断下沉到机制层。后人类批判也不能变成对物质政治的逃避。历史类比只有在机制层面才具有价值。若只是因为今天出现了算法、平台或模型便宣告时代更替,理论便失去区分连续性与断裂的能力。Negarestani Intelligence and Spirit对“技术未来如何在风险治理、解放想象与民主制度之间被争夺”的贡献,在于把宏观判断压回可识别的制度机制。技术未来并不是纯粹的预测对象。“技术未来如何在风险治理、解放想象与民主制度之间被争夺”因而仍需接受Negarestani Intelligence and Spirit之外的经验材料和竞争性解释的共同检验。Negarestani把intelligence理解为能够给出理由、修正承诺并参与规范空间的实践,而非单纯的问题求解能力。这样一来,智能不再由某种生物基质垄断,却也不能被模型性能指标轻易等同。机器是否进入规范性推理空间,涉及责任、概念使用和自我修正,比“像不像人”更为严格。】
- Hui, Yuk. Recursivity and Contingency. Rowman & Littlefield, 2019. (Read: Introduction, pp. 1–28).
【按:Hui对“只有一种现代技术道路”的假设提出挑战。recursivity与contingency让技术不再只是线性进步,而成为由不同宇宙论、文化和制度条件塑造的历史实践。对于AGI而言,这意味着我们不应默认未来只能朝着越来越集中、越来越强大、越来越自动化的方向发展;技术制度可以有不同的历史路径。终章必须区分现实能力判断与哲学可能性。“AI能力、产业hype、哲学智能概念与政治经济现实的区分”从Hui Recursivity and Contingency出发,可以获得更清楚的因果层次。因此,Hui Recursivity and Contingency对“AI能力、产业hype、哲学智能概念与政治经济现实的区分”的说明只有在相应制度机制能够被经验材料识别时才成立。Hui的recursivity并不是对技术自治的简单赞美,而是借由递归与偶然性重新打开技术历史的多路径结构。现代技术并非唯一可能的世界组织方式,不同宇宙论与制度传统可以产生不同的技术关系。AI未来由此不必被预设为同一种规模化、集中化道路。cosmotechnics的方向进一步说明,技术多样性不是给同一套设备贴上不同文化标签,而是技术目的、自然观与社会秩序可能存在不同组合。面对全球AI基础设施,这一视角能够提出比“本地化产品”更深的问题:不同社会是否有能力决定技术发展的尺度、节奏与规范目标。技术多样性因此首先是政治能力问题:不同社会是否能够真正决定自身技术制度,而非只在既定全球基础设施上选择文化风格。】
- Braidotti, Rosi. The Posthuman. Polity, 2013. (Read: Chapter 1, "Post-Humanism: Life beyond the Self," pp. 13–54).
【按:Braidotti区分posthuman subject与传统人文主义主体,为本节提供一种不同于transhumanist enhancement的后人类主义。它可以帮助我们思考技术未来是否必须以“更强的人类”为目标,还是可以重新理解生命、生态和关系性主体。"AI"是高度异质的总称。Braidotti The Posthuman让“自动化、控制、对齐与后人类主体之间不同问题类型”的分析回到证据边界:技术、制度与政治后果之间若缺少可说明的连接,理论类比就只能停留在启发层面。Braidotti The Posthuman与transhumanist enhancement之间必须保持区分:前者并不以制造“更强的人类”为中心,而是重新思考主体与非人生命、技术和环境的关系。这个差异使后人类主义能够成为AI未来主义的重要反向坐标。Braidotti的posthumanism与技术增强主义之间存在根本差异。她批判以自主、理性、欧洲男性主体为尺度的人文主义普遍性,强调主体始终嵌入生命、技术与生态关系。后人类政治因此不是追求更强大的个人,而是重新分配人与非人生命、技术系统和环境之间的依存与责任。Braidotti的affirmative politics也提醒,后人类批判不能只停留在揭露人文主义排除。主体的关系性可以成为建立新伦理的起点:承认人与动物、生态和技术系统的相互依赖,并据此重新分配责任。AI未来由此不只是能力升级问题,还涉及何种生命和环境被纳入技术发展的价值计算。】
- Srnicek, Nick, and Alex Williams. Inventing the Future: Postcapitalism and a World Without Work. Verso, 2015. (Read: Introduction; chapters on full automation, post-work politics, and institutional strategy).
【按:Srnicek与Williams把自动化放进左翼政治战略,而不是当作技术宿命。他们提出full automation、缩短劳动时间、普遍基本收入等主张,试图把技术能力从资本的利润逻辑中重新定向。放在终章可以与Land、Bostrom形成三角张力:自动化究竟是资本的自我加速、需要被控制的技术风险,还是可能被政治组织转化为减少必要劳动的制度条件?借助Srnicek Inventing the Future: Postcapitalism and a World Without Work,“资源、劳动、生态与组织史如何约束AGI未来主义”能够从修辞性命题转化为制度分析。Inventing the Future把自动化置于后工作政治的制度想象中:技术减少必要劳动只有在收入、公共服务、劳动时间和所有权同时改变时才可能具有解放意义。否则,同一自动化能力也可能转化为失业风险与管理强化。技术潜能与政治结果之间不存在自动通道。后工作政治也提醒我们,减少劳动时间与消灭有意义的活动不是一回事。自动化能够替代某些任务,但照护、教育、文化与政治参与的价值不能被简单归入效率逻辑。技术未来的规范目标应当是重新分配必要劳动与自由时间,而不是把“无工作”本身当作终点。后工作政治真正激进的部分在于把自动化与社会权利打包考虑。减少工时、基本收入、公共服务和劳动者议价权缺一不可,否则生产率提高完全可能被资本吸收为利润和失业压力。技术进步本身没有解放方向,只有制度安排能够把生产率转化为更大的时间自主。】
- Benjamin, Ruha. Imagination: A Manifesto. W. W. Norton, 2024. (Read: Introduction and chapters on imagination as a site of struggle).
【按:Benjamin把imagination当作政治资源,而不是软性的文化附属物。它非常适合与Fisher、Kornbluh和AGI未来主义放在一起:如果技术公司可以通过产品路线图塑造大众对未来的想象,那么政治抵抗也必须重新争夺“什么未来值得被想象”的能力。后人类批判也不能变成对物质政治的逃避。“技术未来如何在风险治理、解放想象与民主制度之间被争夺”因而仍需接受Benjamin Imagination: A Manifesto之外的经验材料和竞争性解释的共同检验。Benjamin把想象力从私人心理活动提升为公共政治能力。未来并不是等待技术企业交付的产品,而是在叙事、设计、制度和集体行动中被争夺的空间。将其置于终节,可以使后人类与AGI讨论重新回到民主问题:谁能够提出未来,哪些群体的经验被纳入未来设计,以及何种制度让替代方案变得可实践。想象力的政治还涉及否定现存制度的能力。技术治理常把未来限定为当前产品和商业模式的渐进升级,而Benjamin强调未来可以从被压抑的经验、历史和社区实践中重新构造。读书会最终关心的正是这种开放性:技术未来不是预测对象,也是制度创造的对象。想象力由此与制度实践相接:替代未来只有转化为组织、公共资源和可持续规则,才不会再次退回品牌化愿景。Benjamin的想象政治还可以与技术设计直接相连。数据集、产品路线图和政策场景都包含对“未来用户”和“正常社会”的想象,设计因此已经在分配可能性。扩大想象力意味着让受影响群体进入定义问题和设定目标的过程,而不是只在系统完成后提供反馈。】
All Rights Reserved © Ivan Yi-fan Hu