Introduction
There was no definitive moment when the artificial intelligence era declared its conclusion. No single collapse marked the turning from confidence to doubt. Long before quarterly earnings delivered subtle admissions and policy speeches softened their futuristic cadences, the deeper grammar of the age had already begun to fray. Assertions that once passed as inevitabilities acquired the brittle sheen of branding. Predictions that had been framed as historical certainties revealed themselves as wagers. People sensed it first in daily frictions rather than catastrophic failures. A triage system that misread a patient’s symptoms. A risk model that translated poverty into suspicion. A conversational agent whose confident answers collapsed under scrutiny. The speculative edifice did not shatter at once; it slowly deflated, revealing the contradictions it had concealed. This book begins from that after, adopting the tense of a sober dystopia in which the first collapse of artificial intelligence expectations is treated not as prophecy but as the past we now inhabit.
The collapse did not belong to one community or worldview but to a temporary coalition of incompatible imaginaries. Venture capital, seeking new forms of intangible wealth, treated artificial intelligence as an asset class whose value resided in its claim to reorganize cognition itself (Haskel and Westlake 1 to 4; Floridi 2 to 4). National security institutions invested in it as a strategic infrastructure promising anticipatory power and informational supremacy (Buchanan 1362). Platform corporations narrated it as the natural extension of datafication, the next layer of extraction built atop surveillance capitalist markets already described by Shoshana Zuboff (8 to 13). Academic futurists contributed a repertoire of eschatologies that fused existential risk with corporate aspiration. Public agencies, anxious to appear modern, integrated predictive systems into welfare administration, policing, and public health. What passed for consensus was never consensus at all. It was a convergence of interests sustained only by the promise of an imminent future in which these divergent ambitions would somehow harmonize. Their fracture marks the true beginning of the post-bubble condition.
Crucially, the bubble was never confined to financial exuberance. Its material base stretched across mines in the Democratic Republic of Congo, where cobalt and rare earth minerals were extracted under conditions whose geopolitical and human costs never appeared in pitch decks; across deserts in Arizona and river valleys in Ireland where datacenters consumed immense quantities of electricity and water; and across annotation farms in Nairobi, Manila, and Hyderabad where contract workers labeled data, filtered trauma, and trained the very models whose public narratives insisted on automation. What was marketed in Northern capitals as frictionless intelligence rested upon global supply chains structured by coercion, ecological depletion, and asymmetrical bargaining power. The collapse has not undone these infrastructures. It has merely rendered their violent coherence visible.
The concept of the human that animated the era now appears equally fragile. The dominant imaginary cast the person as a predictable locus of signals, a rational actor whose desires could be inferred through aggregation, whose vulnerabilities could be managed through optimization, and whose voice could be approximated by generative systems. Yet the category of “the human” in whose name these systems claimed to act was never stable. It was shaped by histories of empire, race, disability, and labor. Frantz Fanon demonstrated how colonialism remakes perception and selfhood from within (Fanon 109 to 140). Achille Mbembe’s account of necropolitics showed how sovereignty operates through differential exposure to death, distributing life chances unevenly (Mbembe 66 to 69). Judith Butler traced how grievability is unequally allocated, producing populations whose suffering does not register as loss (Butler 28 to 29). Artificial intelligence, in its speculative phase, inherited these distributions and encoded them into classification schemes, risk scores, and optimization routines. The bubble did not invent inequality; it automated it. Writing after the bubble therefore requires refusing any return to an innocent humanism. The task is to expand the domain of intelligibility, not to restore a figure that never existed.
Institutional life revealed the collapse before markets did. Nurses discovered triage tools whose probabilistic judgments failed to recognize symptoms that their embodied expertise identified instantly. Social workers confronted welfare eligibility systems whose hidden parameters marked entire neighborhoods as risks, and whose appeals mechanisms offered little recourse. Teachers encountered analytics platforms that predicted failure while quietly reinforcing the structural barriers that produced it. Municipal officials learned that predictive models for energy, policing, or housing tended to amplify preexisting inequities under the guise of efficiency. These scenes were not aberrations. They were the logical unfolding of architectures built to optimize populations rather than to understand them. Cathy O’Neil’s description of “weapons of math destruction” captured the basic pattern: opaque, influential, and unaccountable systems whose harms targeted those least empowered to contest them (O’Neil 3 to 5). The difference now is that the harms can no longer be absorbed into the narrative of progress.
The epistemic conditions of the era collapsed alongside the technical ones. Agnotology, the study of culturally produced ignorance, provides the vocabulary for describing how uncertainty was manufactured and maintained (Proctor and Schiebinger 3 to 5). Secrecy around training data, proprietary architectures, and safety evaluations ensured that public scrutiny remained partial. Bruno Latour warned that critique could be repurposed to undermine expertise rather than strengthen it, creating environments in which facts and fictions circulate with equal authority (Latour 225 to 28). Walter Lippmann described the pseudo-environment that mediates between citizens and the world, a construct that now expanded exponentially through algorithmically curated feeds (15). As synthetic media intensified and claims of objectivity became indistinguishable from coherent statistical hallucination, Hannah Arendt’s fear that factuality could lose its grip on public life returned with new force (Arendt 4 to 6). When the bubble finally ruptured, it was not only trust in systems that eroded but trust in the very possibility of shared reality.
The moral vocabularies of the era fared no better. Companies saturated the public sphere with commitments to fairness, transparency, and responsibility, yet these commitments operated largely at the level of branding. As Onora O’Neill argued in the context of bioethics, procedural assurances of autonomy are insufficient where conditions for meaningful agency are absent (O’Neill). Ethics became infraethical: a domain of procedural gestures that conferred legitimacy while leaving underlying practices untouched (Ott and Dabrock). The industry perfected the art of moral laundering, converting ethical discourse into an instrument of governance rather than a constraint on action. Simone Weil’s insistence that attention is the purest form of generosity, a disciplined openness to the reality of the other (Weil, Correspondance 18), stands in stark contrast to a regime that treated interior life as behavioral surplus. The collapse exposed the disjunction: ethics had been spoken everywhere and practiced almost nowhere.
Against this background, the book adopts the past tense deliberately. It writes as if the deflation of the artificial intelligence bubble has already happened because it has happened—not as an apocalyptic event but as a conceptual unmasking. The speculative enchantment has thinned. The claims of inevitability no longer compel. This narrative stance is not nostalgic. It is diagnostic. By writing from the after, the text refuses the urgency that hype demands and instead creates a vantage point from which moral, political, and epistemic reconstruction become possible. Michel Foucault’s description of critique as the art of not being governed “like that” becomes newly relevant (Foucault, Birth 63 to 70). The book’s tense is a methodological wager: that clarity becomes possible only when one ceases to inhabit the narrative of inevitability and begins instead to examine the ruins it left behind.
From this vantage, new questions open. How might institutions be built if they presupposed dignity, dependency, and plurality rather than data extractability and predictive control. What forms of public reasoning and democratic oversight become possible once artificial intelligence is no longer treated as a horizon of salvation. How might a conception of the person grounded in capabilities, care, and intelligibility reshape technology design, legal structures, and political imagination. Amartya Sen’s account of justice as a comparative, publicly reasoned practice becomes indispensable here (Sen, Idea of Justice). The chapters that follow move between diagnosis and reconstruction, tracing how artificial intelligence became a speculative economy of intelligence, how organized ignorance undermined epistemic life, how ethics was transformed into a branding instrument, and how the category of the person can be reclaimed without nostalgia or abstraction. These analyses then turn toward institutional design, proposing ways of structuring technological and political life in which attention, interdependence, and shared power become primary commitments.
We live now in the quiet after of a future that failed to arrive. The models still run. The infrastructures still hum. But the story that once animated them has emptied out. This book takes that emptiness as an opening. What remains after the bubble is not a void but a chance to build institutions capable of honoring the fragile, situated, plural lives that have too often been treated as raw material. The work that follows is an attempt to begin that reconstruction.
Chapter One
The Illusion of Progress, Hype and Disillusion
The first promise of any technological boom is that history has at last found its natural direction. In the years when artificial intelligence systems were promoted as the engine of a new age, the language of inevitability seeped into policy speeches, quarterly earnings calls, and keynote presentations. Progress was presented as a kind of slope on which there was only one acceptable movement, namely forward, and to question the pace or the destination of that slope was treated as a failure to understand the times. This chapter begins by refusing that narrative of inevitability. It traces how artificial intelligence became tied to a myth of linear progress, how that myth was institutionalized through finance, media, and governance, and how the unwinding of the bubble exposed not simply a mispricing of assets but a deeper confusion about what it means for a society to advance.
Evgeny Morozov names one of the central habits of thought that structured this period technological solutionism, the expectation that complex political and moral conflicts can be recast as technical puzzles awaiting efficient solutions (Morozov 5 to 7). In this frame, suffering, inequality, and institutional failure appear as problems of poor data or inadequate optimization rather than as outcomes of history, power, and contested values. Morozov argues that this habit narrows the range of imaginable responses, since once difficulties are framed as misalignments between reality and an ideal model, the proper response will always appear to be better modeling, more comprehensive data collection, and tighter feedback loops between prediction and intervention (Morozov 13 to 18). Within the artificial intelligence boom, this logic produced an atmosphere in which almost any collective challenge, from climate change to mental health to democratic distrust, could be addressed with a promise that more data and more computation would deliver timely, objective guidance.
The promise of technological solutionism did not emerge in a vacuum. It drew strength from older stories about progress as a single trajectory and from a twentieth century faith in expertise that equated scientific authority with moral insight. What distinguished the artificial intelligence era was not that such beliefs existed, but that they were linked to a machinery of speculation and branding that gave them unprecedented scale. Venture capital firms funded start up companies on the basis of their declared ambition to transform entire sectors by replacing human judgment with machine prediction. Public companies crafted narratives in which algorithmic systems would unlock vast reserves of economic value by eliminating inefficiencies that had supposedly persisted only because human beings were too slow, biased, or emotional. Policy makers and regulatory agencies, under pressure to appear modern and innovative, adopted the vocabulary of disruption and transformation, often repeating the slogans of industry white papers and keynote addresses. In this environment, any caution about limits or unintended consequences could be dismissed as an emotional obstacle to innovation rather than as a contribution to serious deliberation.
To understand why this configuration proved so powerful, it is useful to place it in the longer history of speculative technologies. The dot com boom at the end of the twentieth century saw fragile firms rewarded for growth without profits on the promise that network effects and first mover advantages would eventually justify extraordinary valuations. Telecom investment followed a similar pattern, with massive overbuilding of fiber networks ahead of sustainable demand. Robert J. Shiller’s analysis in Irrational Exuberance describes speculative bubbles as periods in which feedback loops between price increases and optimistic narratives lead participants to suspend ordinary judgment, since each rise in valuation appears to confirm the underlying story (Shiller 3 to 7). The artificial intelligence bubble repeated many of these features. Revenue projections were derived from assumptions about universal adoption. Valuations were justified by reference to vague categories such as potential addressable market rather than to demonstrable social benefit. Once again, stories about an inevitable future suppressed attention to the uneven and contested realities of the present.
Luciano Floridi has argued that contemporary enthusiasm for artificial intelligence should be read as another information and communication technology bubble, continuous with earlier cycles of overinvestment and disillusion yet distinctive in its claim to redefine intelligence itself (Floridi 2 to 4). For Floridi, such bubbles arise when there is a persistent decoupling between the noumenal value of a technology, namely what it can actually deliver, and its phenomenal value, namely what markets and public discourse attribute to it (Floridi 5 to 7). In the artificial intelligence case, the phenomenal value was amplified by grandiose talk of general intelligence, civilizational turning points, and existential stakes, while the noumenal value remained tied to more modest advances in pattern recognition, optimization, and data processing. The distance between these two registers could be ignored so long as investment capital continued to flow and spectacular demonstrations of narrow success could be staged. Once costs, failures, and harms became more visible, the gap could no longer be denied.
Mark Buchanan’s recent thesis column in Nature Physics uses the mathematics of complex systems to describe how such bubbles inflate and then unravel. He emphasizes that feedback loops between investor sentiment, media coverage, and apparent market success can produce runaway growth that is not grounded in underlying value and that the resulting regimes are inherently fragile (Buchanan 1362). When each upward movement in stock price is interpreted as proof that the narrative is correct, dissenting voices fade from the conversation, and small shocks in confidence can trigger cascades of selling. Applied to artificial intelligence, this analysis suggests that the collapse of the bubble is less the result of a single scandal or failure than the consequence of accumulated doubt finally outweighing the inertia of enthusiasm.
The sociology and history of technology help explain how the illusion of progress is maintained even as evidence of limitation accumulates. Bruno Latour’s work on technoscience, especially his essay “Why Has Critique Run out of Steam?,” insists that technologies are never isolated objects but elements in networks that include laboratories, financial backers, regulatory bodies, and users (Latour 227 to 30). What appears in a promotional image as an autonomous system relies in practice on continuous support, maintenance, and negotiation among these actors. The stability of a technology depends not only on internal design but also on the ongoing willingness of institutions to keep investing trust and resources. Nigel Thrift’s account of knowing capitalism further describes an economy organized around the capture, commodification, and circulation of knowledge, in which firms attempt to monetize information about markets, behaviors, and environments, and in which speculative narratives about innovation become central assets in their own right (Thrift 1 to 5). In such an order, stories about artificial intelligence do not simply report on technical achievements; they function as instruments for organizing capital, attention, and policy.
Once artificial intelligence became a favored keyword within knowing capitalism, agnotology, the study of culturally produced ignorance, took on a specific technological form. Robert Proctor and Londa Schiebinger coined this term to describe situations in which ignorance is actively created through secrecy, distraction, and strategic doubt, often to protect powerful interests (Proctor and Schiebinger 3 to 5). Within the artificial intelligence bubble, proprietary claims over data and models limited independent auditing. Companies invoked trade secrecy to avoid disclosing training sets, architectures, and evaluation metrics. Public relations campaigns emphasized successful pilot projects and carefully curated success stories while concealing failures and externalities. Policymakers were invited to rely on industry authored white papers that framed risks in narrow terms and promised that emerging problems would be handled by future technical improvements. The result was not a simple absence of information. It was a structured field in which certain questions, such as who bore the costs of errors or whose labor sustained the systems, rarely appeared as legitimate objects of inquiry.
Morozov’s language of technological solutionism helps clarify how this field of managed ignorance interacted with a deeper desire for easy answers (Morozov 23 to 27). When a complex injustice is described as a coordination problem, and when predictive systems are marketed as tools that can optimize coordination, resistance to their deployment begins to look like an attachment to inefficiency or sentimentality. The artificial intelligence boom thrived on this moralizing of efficiency. Proponents insisted that algorithmic decision tools would remove human bias by applying consistent criteria, even when those criteria were built from patterns in data that encoded past discrimination. They promised that automation would liberate workers from repetitive tasks, even when early deployments translated into speed up and intensified surveillance rather than into redistributed time and power. In such a setting ethical reflection could be framed as a luxury that might be postponed until after the supposed benefits had been secured.
The illusion of progress rested, then, on a convergence of speculative finance, cultural narrative, and organized unknowing. Investors were rewarded for identifying the next artificial intelligence breakthrough. Executives were rewarded for promising that their firms would dominate markets through artificial intelligence integration. Researchers were rewarded for benchmark beating papers that could be translated into compelling slides, regardless of whether the underlying advances were robust, generalizable, or beneficial outside narrow contexts. Journalists were rewarded with attention for stories that framed incremental improvement as revolutionary change. Across these domains, there were few incentives for public reckoning with limits. To ask what a given system could not do, or whose interests it sidelined, was to risk being cast as out of step with an era that officially admired boldness.
Disillusion did not arrive through a single catastrophe. It emerged gradually as frictions accumulated between the promised trajectory and lived experience. High profile applications in criminal justice, hiring, and welfare administration reproduced and sometimes intensified racial and economic disparities, which undermined claims of neutral efficiency. Reports about the environmental costs of training and running large models, including significant energy consumption and water use, complicated narratives of dematerialized progress. Workers who had been assured that automation would enhance their creativity found themselves subject to new forms of quantitative monitoring and performance assessment. At the same time, widely publicized failures of conversational agents and decision assistants made it difficult to maintain the fantasy that these systems possessed anything resembling common sense or contextual understanding. Each of these developments chipped away at the plausibility of the story in which artificial intelligence would simply carry societies forward along an unquestioned path.
Shiller’s account of bubbles emphasizes that when expectations finally adjust, the sense of betrayal is often as much narrative as financial (Shiller 11 to 14). People come to feel that they were misled about the direction of history, encouraged to act on a script that turned out to be unreliable. In the artificial intelligence case, the disorientation was intensified by the way in which the technology had been woven into definitions of modernity and competence. To be skeptical of artificial intelligence in its speculative phase was to risk being coded as backward looking or irrational. When the bubble began to deflate, the same institutions that had urged compliance often shifted to a defensive posture, claiming that they had always recognized the limitations and that any harms were the result of improper use. The effect was to leave many citizens with the impression that the ground beneath their sense of collective purpose had quietly shifted and that no one had been accountable for the shift.
This chapter has argued that the first collapse of artificial intelligence expectations should be read as a revelation of an underlying confusion about progress rather than as a mere correction in pricing. The bubble was economically unrealistic in that it treated public trust, regulatory patience, and human attention as inexhaustible and freely available to be expended in pursuit of speculative gains. It was morally unanchored in that it allowed organizations to celebrate their transformative vision while neglecting to ask which lives were being exposed to new harms and which voices were excluded from conversations about acceptable risk. To describe these dynamics is not to deny that artificial intelligence yielded genuine technical achievements or that some systems continue to serve useful roles. It is to insist that those achievements were bound up with a story in which history itself appeared as an asset to be leveraged rather than as a shared project to be argued over.
The chapters that follow take up the consequences of this recognition. They ask what happens to knowledge when systems of organized unknowing falter, what happens to moral language when ethics has been treated as a branding exercise, and what happens to the idea of the human when artificial intelligence has been presented as its replacement, its tutor, and its judge. If the spell of inevitability has broken, then new questions become possible about who defines advancement, how disagreement is handled, and how societies might pursue improvement without entrusting their future to speculative salvation. The end of the bubble is not the end of artificial intelligence, nor is it the end of technological change. It is an opening in which the illusion of automatic progress gives way to the harder work of confronting power, responsibility, and the plurality of ways in which lives can go well or go badly.
Chapter Two
The Speculative Economy of Intelligence
The first chapter dismantled the story of artificial intelligence as an inevitable march toward progress and revealed instead a patchwork of promissory narratives, technocratic fantasies, and cultivated forgetfulness. In this chapter I turn from narrative to balance sheet. If the previous analysis asked what people believed about artificial intelligence, this one asks what, precisely, capital believed it was buying. The answer is disarmingly simple and severe in its implications. Far more than systems that could reliably perform useful tasks, investors were purchasing a claim on the future of intelligence itself, treated as a financial asset that could be owned, leveraged, and traded. The artificial intelligence boom became a speculative economy of intelligence, in which cognition, prediction, and learning migrated from philosophical and psychological vocabulary into the language of intangible assets, valuation models, and risk instruments.
Seen from this vantage point, hype around artificial intelligence does not appear as an unfortunate distortion around an otherwise sound technological domain. It appears instead as a paradigmatic expression of contemporary technocapitalism, in which financial markets continuously seek out new frontiers of abstraction, moving from land, to industrial plant, to brands and intellectual property, and now to intelligence and attention themselves. Luis Suarez Villa describes this order as one in which technological innovation and corporate power fuse into an economic regime that systematically exploits creativity and knowledge as primary resources, with firm structures designed to capture and privatize them in the service of accumulation (Suarez Villa 3 to 7). In such a regime it is unsurprising that intelligence is recoded as an asset class and that artificial intelligence firms are valued less for what they can presently do than for what they appear to prefigure.
Karl Marx’s Grundrisse offers a vocabulary for this shift that remains unsettlingly apt. In his notebooks on money and capital, Marx treats capital as value that seeks expansion, a social relation that acquires apparent autonomy once money is advanced with the expectation of a greater sum in return (Marx 247 to 50). Capital circulates through a sequence in which money becomes commodities and then returns as more money, written as M C M prime. What matters in this sequence is not the particular use of the commodity but the expansion of value promised at the end of the circuit. In that sense the artificial intelligence firm is a textbook instance of what Marx already saw in the nineteenth century. The models, datacenters, and even engineers are not ends in themselves. They are moments in a circuit where present expenditure is justified by the expectation of future rents on automated cognition, predictive power, and behavioral control.
The intensity of the bubble around artificial intelligence came from the fact that this circuit operated twice over. First, companies promised to substitute machine learning for human labor, thereby lowering costs and raising profits in downstream sectors. Second, they promised to substitute speculative narratives for transparent accounting, allowing investors to capitalize projected future dominance long before sustainable business models existed. Luciano Floridi’s diagnosis of contemporary artificial intelligence hype identifies all the familiar markers of a technology bubble: a potentially disruptive technology, speculation racing ahead of reality, increasingly abstract valuation paradigms, proliferation of start ups, and a striking absence of proportionate regulation (Floridi 2 to 4). What distinguishes this bubble from earlier ones is that the commodity at stake is not a specific device or infrastructure but the imagined capacity to reorganize cognition at scale.
The behavior of financial markets in this period makes sense only when it is situated in the longer history of bubbles. In Manias, Panics, and Crashes, Charles Kindleberger and Robert Aliber show that speculative manias typically follow a recognizably patterned sequence: a displacement such as a new technology or market opens unexpected opportunities, credit flows into the sector, asset prices rise, expectations detach from fundamentals, and eventually a moment of revulsion or panic reverses the process (Kindleberger and Aliber 15 to 22). Artificial intelligence served as an exemplary displacement. The combination of impressive but narrow technical results, public fascination, and geopolitical anxieties around competition created a situation in which firms and governments felt compelled to invest or risk being left behind. Investment did not track the measured productivity gains of artificial intelligence systems, which remained patchy and hard to attribute. It tracked instead the fear of missing an apparently epochal shift.
Gadi Barlevy’s review of greater fool theories clarifies the logic at work in such episodes. On a greater fool account, bubbles arise when traders knowingly purchase overvalued assets because they expect to resell them to others at even higher prices; rationality lies not in the underlying cash flows but in beliefs about the behavior of later entrants (Barlevy 54 to 57). Applied to artificial intelligence, this means that many early stage valuations of model providers and platform firms were justified not by discounted future profits from specific services, but by expectations that other funds, sovereign wealth vehicles, or strategic buyers would continue to bid up anything labeled as an artificial intelligence play. The speculative economy of intelligence thrived on this recursive expectation. It was enough that everyone believed that everyone else believed in the inevitability of artificial intelligence.
To understand why investors were willing to treat intelligence in this way, one must consider the broader transformation of capitalism toward intangibles. Jonathan Haskel and Stian Westlake document how, in major economies, investment in intangible assets such as software, design, brands, organizational processes, and data has come to exceed investment in traditional tangible assets like plants and machinery (Haskel and Westlake 1 to 4). They argue that intangible assets possess distinctive economic properties. They scale easily across contexts, they often cannot be recovered once sunk, they produce substantial spillovers, and they rely heavily on synergies with other assets (Haskel and Westlake 20 to 25). These properties amplify both upside and uncertainty. An intangible rich firm may generate enormous returns if its product becomes a standard, yet its value can collapse quickly if expectations about its intangible portfolio shift.
Artificial intelligence firms concentrated all four properties. Their primary capital lay in trained models, proprietary data corpora, specialized software stacks, and organizational know how in scaling compute infrastructure. These assets were highly scalable, since a trained model could be embedded across products and markets at marginal costs that appeared negligible relative to its training expense. They were sunk, since training runs and data aggregation required substantial upfront outlays that could not be recovered if projects failed. They generated spillovers, as advances in architectures and optimization techniques diffused quickly through shared academic and open source channels. They depended on synergies, because value emerged when models, data, user interfaces, and complementary cloud services interlocked.
From the perspective of a financialized economy hungry for growth, this made artificial intelligence an ideal object of speculative enthusiasm. Capitalism without capital, as Haskel and Westlake name this order, encourages investors to seek out precisely those firms whose market value is dominated by expectations of future intangible returns (Haskel and Westlake 5 to 7). In that context, artificial intelligence appeared as the ultimate intangible. Whereas a factory has visible output and capacity constraints, and even a brand’s power can be proxied by sales, the productive potential of machine intelligence seemed bounded only by imagination. It promised to improve any process that involved pattern recognition, prediction, or optimization. Valuing such promises invited a kind of vertigo, since standard metrics of installed base or current profits seemed to underestimate the scale of possible transformation.
Floridi notes that this vertigo produced a shift from empirical to narrative valuation. Rather than anchoring prices in demonstrable, sector specific productivity gains, markets operated with stories about exponential improvement, network effects, and inevitable lock in (Floridi 7 to 9). The familiar motifs of technology hype reappeared: analogies to electricity, the internet, and general purpose technologies that would reconfigure every industry. Mark Buchanan’s short essay on “the laws of inflating the AI bubble” in Nature Physics uses the mathematics of complex systems to describe how a small number of plausible successes are amplified into evidence for a systemic revolution, while failures and limits are explained away or postponed (Buchanan 1362). The resulting picture was less an empirical estimate than a mythology of impending transformation, backed by large language models that themselves generated convincing narratives on demand.
At this point Michel Foucault’s notion of governmentality becomes indispensable. In his lectures on Security, Territory, Population and The Birth of Biopolitics, Foucault describes liberal and neoliberal government not simply as a collection of legal institutions but as an art of managing populations through economic rationalities, metrics, and forms of knowledge (Foucault, Security 108 to 13; Foucault, Birth 63 to 70). The speculative economy of intelligence can be read as a new chapter in this history. States and firms began to reason about social order and economic control in terms of models that could predict and shape behavior. Investment in artificial intelligence was therefore not only a financial bet but a governmental wager, namely that those who owned and operated predictive infrastructures would set the terms of competition and security for decades.
In that sense, the artificial intelligence sector capitalized not just future streams of profit but future capacities to govern. Cloud platforms promised to manage logistics, energy, policing, and border control through algorithmic systems. Venture capital funded companies that claimed to automate hiring, credit scoring, health triage, and classroom instruction. Even where performance remained shaky, the prospect of embedding intelligence into infrastructure aligned with neoliberal instincts to render ever more domains calculable and optimizable. The speculative premiums attached to artificial intelligence stocks and start ups therefore reflected an anticipation of power, not simply an estimation of revenue. Investors purchased a share in what they imagined would become the operating system of social and economic life.
This double anticipation of profit and power intensified the distance between valuations and fundamentals. In a traditional bubble around a commodity such as tulips or a sector such as railways, there remains some tangible relation between productive capacity and price. In the artificial intelligence bubble, that relation weakened further because the key resource, intelligence, could be invoked without clear metrics. Productivity studies struggled to isolate artificial intelligence’s contribution from broader digitalization and automation trends. Yet market narratives simply treated this ambiguity as evidence of underappreciated upside rather than as a reason for caution.
The epistemic looseness of the object invited financial alchemy. Analysts constructed models in which modest assumed improvements in efficiency, multiplied across industries and decades, justified extraordinary valuations. Consultancy reports projected trillions of dollars in added gross domestic product from artificial intelligence adoption, with methodological caveats relegated to footnotes. Institutional investors, under pressure to find yield in a low interest environment and to demonstrate engagement with frontier technologies, allocated funds to vehicles whose mandates were framed in terms of exposure to disruption rather than disciplined assessment of cash flows. The speculative economy of intelligence thrived on this cascade of institutional incentives, in which each actor could present their participation as prudent engagement with innovation while collectively inflating a structure that rested on very thin empirical ground.
Marx’s account of fictitious capital helps name what emerged. Fictitious capital describes claims to future surplus value that take on an apparent life of their own, as with stocks and bonds whose prices fluctuate in ways increasingly detached from the underlying production process (Marx 597 to 603). In the case of artificial intelligence, layers of derivatives, structured products, thematic exchange traded funds, and private fund positions accumulated on top of an industry that was still in the process of defining its stable uses. This did not mean that nothing real existed at the base. Datacenters were constructed, chips manufactured, and engineers hired. It meant that the mass of financial claims vastly exceeded any realistic estimate of near term surplus generated by artificial intelligence deployments.
The resulting configuration walked a narrow line between innovation and moral hazard. On the one hand, bubbles can leave behind useful infrastructure. Kindleberger and Aliber note that the railway mania of the nineteenth century and the dot com boom both produced networks that later generations used more soberly (Kindleberger and Aliber 153 to 60). On the other hand, bubbles redistribute resources, wealth, and political influence in ways that are never neutral. Floridi emphasizes that the hype around artificial intelligence channelled capital, research effort, and regulatory attention into narrow models of technological progress, crowding out alternative pathways and applications that might have produced more equitable benefits (Floridi 10 to 12). The collateral damage of misallocated attention and funding is not easily reversed once the bubble deflates.
Moreover, the artificial intelligence bubble did not float above existing inequalities. It rested on global supply chains of energy, rare earth minerals, and human labor. The most celebrated firms in the sector relied on contract workers labeling data, moderating content, and maintaining physical infrastructure under precarious conditions. The speculative value assigned to artificial intelligence as an asset obscured the fact that its production depended on intensified extraction of time, attention, and ecological resources in specific places. Haskel and Westlake observe that economies rich in intangibles often see widening inequality, since returns accrue disproportionately to those who already hold complementary tangible and institutional assets (Haskel and Westlake 131 to 35). The artificial intelligence boom replicated this pattern, concentrating wealth and power among existing corporate and financial elites while distributing risks and externalities across more vulnerable populations.
The speculative economy of intelligence therefore cannot be understood as a neutral mispricing that markets will quietly correct. It must be read as a political formation in which particular visions of intelligence and control acquired institutional dominance. Foucault’s account of neoliberal governmentality underscores that such formations reshape subjectivity as well as policy. When firms are rewarded for presenting themselves as purveyors of intelligence, they reorganize their internal practices to produce data, dashboards, and prediction products, even when these do not serve human flourishing (Foucault, Birth 219 to 25). When governments race to fund artificial intelligence initiatives to avoid falling behind rival states, they tacitly accept a future in which legitimacy is tied to technological prowess rather than to responsiveness and justice.
In the aftermath of the crash in expectations around artificial intelligence, what remains is a landscape of half built infrastructures and half fulfilled promises. The datacenters continue to draw power. The models continue to run. Yet the speculative aura that surrounded them has thinned, and the question that confronts us is what to do with systems that were built first to attract capital and only secondarily to serve people. The epistemic rupture described in the next chapter, in which public trust in expertise and truth erodes under the weight of manufactured ignorance, cannot be disentangled from this speculative history. Once intelligence has been treated as a bubble asset, its moral and epistemic authority becomes suspect.
The argument of this chapter has not been that speculation around artificial intelligence was uniquely irrational. Instead, it has shown that the bubble around artificial intelligence exemplifies the tendencies of a financialized, intangible rich capitalism in which promises about future cognition, control, and optimization become central vehicles for capital accumulation. In that sense, the collapse of artificial intelligence is not only a story about technology. It is a revelation of the underlying order in which technology operates as a medium for financial and governmental power. To address the harms that emerged from this period, as the subsequent chapters will argue, we must therefore attend not only to algorithms and datasets, but also to the speculative economy that treated intelligence itself as an instrument of profit and rule.
Chapter Three
Epistemic Collapse: Ignorance, Complexity, and Misinformation
The end of the AI boom was not experienced first as a market crash but as a cognitive event. What came into view was not simply a set of overvalued firms or failed products, but a much older and more pervasive pattern of not knowing, an architecture of ignorance that the AI bubble had amplified rather than invented. When the valuations fell and the forecasts were quietly revised, people did not only lose faith in particular systems; they also lost confidence in their own ability to distinguish truth from spectacle, explanation from persuasion, expertise from performance. The collapse was therefore epistemic before it was economic: a collective recognition that the conditions under which knowledge had been produced, circulated, and consumed were themselves damaged.
To name that damage we have to turn to a tradition that studies ignorance as carefully as knowledge. Robert Proctor and Londa Schiebinger’s project on agnotology insists that ignorance is not simply an absence of information but a product of cultural and political work, “the making and unmaking of ignorance” in ways that track power (Proctor and Schiebinger 1). Their anthology shows again and again that not knowing can be organized, funded, and defended, that silence and confusion can be as carefully engineered as any discovery. In a widely cited formulation, Proctor argues that ignorance must be treated as something “made, maintained, and manipulated by means of certain arts and sciences,” an outcome of institutions and interests rather than a passive vacuum waiting to be filled (Proctor and Schiebinger 3). That insight is central to understanding how the AI era ended, because what looked at first like a failure of prediction or modeling in fact revealed a deeper economy of manufactured doubt.
The AI bubble thrived in precisely the kind of environment that agnotology describes. Venture funded narratives promised precision, neutrality, and omniscience at the very moment when the wider information sphere was becoming increasingly unstable. The more AI was sold as the neutral keeper of facts, the more incentive there was to obscure the fragility and partiality of its training data, the conflicts of interest built into its deployment, and the structural blind spots encoded in its architectures. Proctor describes how ignorance can be cultivated through secrecy, selective disclosure, and the careful management of controversy, practices that have marked industries from tobacco to fossil fuels (Proctor and Schiebinger 3 to 5). AI firms, with their proprietary datasets, non disclosure agreements, and cultivated narratives of inevitability, simply updated those techniques for a new domain.
Bruno Latour’s warning about critique losing its way gives this situation a further twist. In his essay “Why Has Critique Run out of Steam,” Latour worries that traditional habits of debunking can unintentionally support the very actors who profit from manufactured doubt, since the tools that once exposed superstition are easily repurposed to question climate science, public health, or any inconvenient fact (Latour 225 to 228). The AI era unfolded in exactly this paradoxical climate. Sophisticated skepticism was available everywhere, yet the most consequential claims of the AI industry were the least examined. Public energy was directed toward debating whether consciousness could emerge in large models or whether machine learning was “really” intelligent, rather than toward the more mundane and more dangerous questions of who was excluded from training sets, who profited from opacity, and who bore the risks when systems failed. Latour’s distinction between “matters of fact” and “matters of concern” became painfully concrete: technical achievements were treated as settled facts while their social meaning and distribution remained unexamined matters that should have concerned everyone (Latour 231 to 234).
The result was not a neutral accumulation of error but an organized confusion about what counted as knowledge. Walter Lippmann diagnosed a similar pattern in the early twentieth century when he described the “pseudo environment” that intervenes between citizens and the world, a symbolic order composed of stereotypes, partial reports, and images that can never match the complexity of events themselves (Lippmann 15). In the AI era that pseudo environment was no longer limited to newspapers or radio. It was saturated with algorithmically curated feeds, recommendation systems, and generative outputs that constantly rewrote the surface of reality. The key difference was speed and scale. Where Lippmann worried about editors and propaganda bureaus, we confronted automated systems that could recombine fragments of text, sound, and image faster than any human editorial process, while their underlying logic remained hidden.
The emergence of deepfakes crystallized this danger. Early experimental tools that could replace faces in videos or synthesize speech were quickly incorporated into a wider ecology of disinformation. Reports by organizations such as UNESCO and research surveys on synthetic media documented how political actors, commercial fraudsters, and hobbyists used these tools to erode the evidentiary status of audio and visual records, making it harder to trust even direct testimony (UNESCO). The threat was not only that people would believe fabricated content, although that happened often enough, but that the very category of evidence would lose its force. When any recording could plausibly be dismissed as manipulated, those who benefited from impunity gained yet another resource. Soroush Vosoughi, Deb Roy, and Sinan Aral’s study in Science showed that false news on Twitter spread further, faster, and more broadly than true news, especially in political domains, because it triggered surprise and emotional arousal in ways that rewarded sharing (Vosoughi, Roy, and Aral 1146 to 1148). The AI boom magnified this dynamic by lowering the cost of producing plausible falsehoods and by embedding those falsehoods in increasingly personalized attention economies.
Against this background, the notion of a simple “misinformation problem” was always inadequate. The difficulty was not only that people encountered wrong statements, but that the entire ecology of communication rewarded content that was engaging rather than reliable, novel rather than measured, emotionally vivid rather than contextually grounded. When large language and image models entered this ecology, they did so as generative engines trained to maximize coherence and plausibility, not truth. Their outputs were presented as answers, yet their internal processes were fundamentally statistical. The public was asked to treat textual smoothness as a proxy for accuracy, a fragile alignment at best. Once the gap became visible, trust collapsed quickly.
The crisis was deepened by institutional decisions that weaponized complexity against accountability. Helen Nissenbaum’s work on privacy insists that what people care about is not information sharing as such but the violation of contextual norms, the ways in which data moves across roles, situations, and expectations in ways that feel inappropriate (Nissenbaum, Privacy in Context 2 to 4). In one of her most widely cited formulations, she argues that privacy is preserved when informational norms are respected and violated when they are breached (Nissenbaum 224). AI firms, however, capitalized on the opacity of their systems and the difficulty of tracing data flows, repeatedly framing architectural decisions as technically necessary rather than normatively contestable. When challenged, they offered explanations in highly specialized language that most users, and often regulators, could not decipher. Complexity became a kind of shield, an environment in which it was always possible to say that harms were unintended, that biases were emergent properties of data, that no one could really have known.
Cathy O Neil’s account of “weapons of math destruction” captures how this shield functioned at scale. She describes predictive models that are opaque, influential, and difficult to contest, especially in sectors such as policing, education, and hiring (O Neil 3 to 5). These systems convert human lives into scores and risk categories without meaningful avenues for appeal. Errors and biases propagate because those who are most harmed typically have the least access to the technical and legal resources required to challenge them. The AI era multiplied these arrangements, embedding similar logics in recommendation algorithms, credit scoring, and automated content moderation. When the bubble burst, it became evident that many of the harms had not been isolated accidents but predictable outcomes of architectures that concentrated interpretive power in the hands of system designers.
Robert Proctor warns that ignorance is not only a residue of what has not yet been studied but also something actively produced to keep certain arrangements intact. He notes that ignorance can be created by destroying documents, restricting access, inventing doubt, or framing questions so that impolite answers never arise (Proctor and Schiebinger 5, 18 to 19). In the AI context, corporate secrecy around training data, proprietary model weights, and safety evaluations functioned in precisely these ways. Whistleblower testimonies and investigative reporting revealed how internal concerns about bias or safety were often minimized or reframed as public relations problems. Research that documented harms could be downplayed or delayed, while favorable benchmarks received prominent marketing treatment. The point is not that everyone involved acted with deliberate malice. Rather, institutional incentives favored practices that maintained uncertainty about the scale and distribution of risk.
At the same time, the wider political environment increasingly normalized doubt as a tool of power. Climate denial campaigns, anti vaccine movements, and conspiratorial politics had already shown how effective it could be to flood the public sphere with conflicting claims, leaving citizens exhausted and cynical. Latour’s anxiety that critique could slide into generalized suspicion found a new illustration: methods that once exposed corporate manipulation now served to undermine any inconvenient expertise (Latour 241 to 244). In that environment, AI companies could present themselves as neutral providers of clarity, even as their systems participated in the same dynamics of fragmentation and amplification that had undermined trust in the first place. When those promises failed, the disillusionment cut doubly deep, because people discovered that the thing they had hoped would stabilize knowledge had itself been part of the destabilizing process.
Hannah Arendt’s reflections on truth and politics help explain why this felt so corrosive. In “Lying in Politics,” Arendt examines how public relations strategies surrounding the Vietnam War did not simply conceal facts but substituted an entirely manufactured image of reality, a “fictitious world” that eventually collided with events (Arendt 4 to 6). For Arendt, the danger is not only that lies will be told, but that the distinction between fact and fiction will lose its grip on public life, leaving citizens unable to orient themselves. The AI era reproduced this pattern in distributed form. Instead of a single official narrative emanating from a state, there were countless micro narratives produced by opaque recommendation engines and generative systems, each tailored to a user’s presumed preferences. The shared world that Arendt sees as the condition for politics was increasingly difficult to locate. When the bubble burst, many discovered not only that particular AI narratives had been exaggerated, but that they no longer trusted any authoritative account of what had happened.
The epistemic collapse that followed the AI crash therefore cannot be reduced to an unfortunate side effect. It exposed three interlocking structures. First, a political economy of agnotology in which institutions profit from cultivating uncertainty, confusion, and doubt. Second, a technological infrastructure that rewards virality and emotional engagement over verification, amplifying falsehood faster than truth, as Vosoughi, Roy, and Aral show in their empirical work (Vosoughi, Roy, and Aral 1146 to 1151). Third, a culture of complexity in which opacity is treated as the natural price of innovation, so that those most affected by decisions have the least capacity to understand or contest them.
To diagnose this situation is already to gesture toward the work that later sections of this book will attempt. We will have to rebuild concepts of personhood that do not reduce individuals to datapoints, develop practices of care that resist the commodification of attention, and reshape institutions so that they are answerable to those whose lives they govern. Yet those constructive projects cannot proceed without a clear account of how knowledge itself was compromised. The AI bubble was not only economically unrealistic and morally unmoored, as earlier chapters argue; it was also epistemically corrosive because it thrived in, and contributed to, a culture that treats ignorance as an acceptable byproduct of progress. Proctor’s insistence that we study the “political geography” of non knowing, Nissenbaum’s analysis of breached informational norms, O Neil’s depiction of unaccountable models, Latour’s critique of critique, Lippmann’s pseudo environment, Arendt’s worry about manufactured worlds, and the empirical mapping of misinformation cascades in platforms all converge on one lesson. If we wish to remake the human after the bubble, we must first learn how to recognize, confront, and refuse the organized ignorance that allowed the bubble to grow.
Only then can subsequent chapters turn credibly to questions of dignity, care, and solidarity. Without this prior reckoning, any proposal for ethical AI or post AI governance will risk repeating the very patterns that brought us here, decorating a damaged epistemic order with new slogans. The collapse has given us a rare opportunity to see ignorance not as a simple deficit but as a structured, contested, and in many cases deliberately crafted condition. Taking that opportunity seriously is the first step toward any more honest future.
Chapter Four
The Moral Void, Ethics as Buzzword
By the time the speculative economy of artificial intelligence began to falter, the language of ethics had already saturated the landscape. Every large platform announced new advisory boards, fairness frameworks, and responsible innovation initiatives; every earnings call affirmed commitment to privacy, dignity, and the common good. Yet when the bubble burst, what became visible was not a surplus of conscience but a deficit, a strange emptiness that ethical language had helped to conceal. The most unsettling feature of the era was not that companies lacked moral vocabulary, but that ethics itself had been absorbed into the repertoire of branding, risk management, and product design. Ethics had been spoken, cited, and publicized, yet very little in the underlying practices changed. What collapsed, therefore, was not only an industry myth about technological progress, but also a widespread faith that the mere invocation of ethics could restrain power or protect the human.
The field that named this new order of power most clearly was Shoshana Zuboff’s account of surveillance capitalism. She argues that a distinct economic logic emerged in which firms claimed human experience as raw material for data extraction and behavioral prediction, converting unstructured life into proprietary behavioral surplus that could be traded in new markets for future behavior (Zuboff 8 to 13). Surveillance capitalism, on this view, is not an accidental misuse of data but a coherent project that depends upon asymmetries of knowledge and consent. Under this regime, ethical talk about transparency and choice coexisted with business models that required users to remain in the dark about how their lives were being monitored, modeled, and monetized. The firms that narrated themselves as stewards of innovation and empowerment built their profitability on an infrastructure that treated subjectivity as a mineable resource. Ethics language functioned as a legitimating surface, a way to stabilize public trust while the underlying practices intensified.
The scandal around Cambridge Analytica makes this structure concrete. Investigative reporting, such as Carole Cadwalladr and Emma Graham Harrison’s coverage in the Guardian, documented how data from tens of millions of Facebook users were harvested and exploited without informed consent to build psychographic profiles and to target political messaging in the United States and the United Kingdom (Cadwalladr and Graham Harrison). The data practices themselves were not aberrations in an otherwise benign ecosystem; they were extensions of an economy in which microtargeting, behavioral prediction, and the sale of granular personal information had already become standard. What ethics language achieved, in this context, was not prevention but delay. The company published community standards and privacy principles, pledged to put people first, and convened internal review structures, while its basic economic commitment remained the maximal capture and use of personal data. Only when the scale of the scandal became impossible to deny did the narrative shift from celebration of data driven personalization to embarrassed promises of reform.
The same pattern holds in the case of the Facebook emotional contagion experiment. Kramer, Guillory, and Hancock manipulated the emotional content of users’ News Feeds to test whether exposure to more positive or negative posts would influence the tone of users’ own updates, demonstrating what they called massive scale emotional contagion through social networks (Kramer, Guillory, and Hancock 8788). The study was defended as minimal risk and as consistent with existing terms of service, yet it revealed a willingness to treat unconsenting users as experimental subjects whose moods could be shifted as part of routine platform optimization. The language of research ethics, of institutional review and informed consent, appeared only after public backlash. The underlying premise that the interior affective life of users could be adjusted and measured as part of product experimentation had already been normalized within the industry. Here again, ethics arrived as a public relations instrument rather than as a prior constraint.
To name this phenomenon simply as hypocrisy would be too thin. It is better understood as a structural deployment of ethics as technology. Ethics was repurposed as an instrument of governance, an apparatus for organizing legitimacy, rather than as a set of limits placed around what firms could do. Onora O Neill’s distinction between autonomy and trust in bioethics is instructive here. She observes that late modern institutions have learned to perform respect for autonomy largely by proliferating consent forms and procedural disclosures, while failing to cultivate the conditions in which agents can actually assess risks, understand consequences, and rely upon others in good faith (O Neill). In such environments, trust is not restored by more reassurances or more documentation but by demonstrable reliability and restraint. The artificial intelligence industry translated autonomy into clicks on permission boxes and acceptance of terms, while shielding the full scope of data extraction and downstream use. In the name of respecting choice, systems were designed so that most people could not see what they were choosing.
Theologian and philosopher Simone Weil offers a very different account of moral life that exposes this hollowness. In a now famous letter to the poet Joë Bousquet, she writes that attention is the rarest and purest form of generosity, a line preserved in her collected Correspondance (Weil, Correspondance 18). Attention, in Weil’s broader work, is not a sentimental feeling but a strenuous discipline of looking at another person and at the world without appropriation or self interest. To attend is to become quiet enough that the reality of the other can appear as it is, not as we would prefer it to be. When one overlays Weil’s account of attention onto the data practices described by Zuboff, the dissonance is stark. In the surveillance capitalist model, the interior life of the user is not an object of reverent attention but of instrumental gaze. The goal is not to behold but to infer, not to receive but to predict and shape. Even when companies spoke of understanding users in order to serve them better, the primary vector of that understanding was toward greater monetization and control.
Judith Butler’s analysis of mourning and vulnerability in Precarious Life helps clarify the ethical stakes of this instrumental gaze. She argues that ethical and political life depends on recognizing the precariousness of embodied beings, and she shows how some lives are systematically framed as less valuable or less grievable than others (Butler 28 to 29). The distribution of attention, in this sense, is never neutral. Media and political frames can render certain deaths worthy of national mourning and others barely mentionable, thereby organizing whose suffering counts as a public event and whose remains a private misfortune. If one reads Butler alongside the advertising infrastructures that emerged in the artificial intelligence era, a disturbing analogy appears. Platforms built elaborate mechanisms to maximize engagement and revenues, learning to track which stimuli would capture the most gaze and which populations could be most profitably targeted. Yet there was no comparable investment in perceiving whose vulnerability was being exploited, whose autonomy was being eroded, or whose political agency was being bent toward the aims of paying clients. The asymmetry between the precision of behavioral prediction and the thinness of moral regard is itself a symptom of the moral void.
It is tempting to reassure oneself that this void belonged only to corporations, while ethicists and regulators stood outside as independent critics. The historical record is more complicated. As public concern about privacy and discrimination grew, companies increasingly convened ethics boards, funded academic centers, and hired ethicists in residence. Some of this work was sincere and important, yet much of it was structurally constrained by the fact that ethics was embedded within the very organizations whose conduct it was meant to scrutinize. Scholars of organizational life have long noted how audit regimes can become rituals that certify legitimacy without altering underlying practices, a pattern Michael Power has described in relation to the audit society where verification procedures substitute for substantive change (Power 10 to 17). Within the artificial intelligence sector, ethics functions often drifted toward the same ritualization. Frameworks were drafted, principles were published, impact assessments were piloted, yet there remained remarkably few cases in which major revenue generating products were halted or fundamentally redesigned on ethical grounds. The language of responsibility multiplied even as the number of enforceable constraints remained minimal.
Tabea Ott and Peter Dabrock provide a particularly incisive diagnosis of this phenomenon in their article on transparency in artificial intelligence based health care. They argue that many current debates treat transparency as an ethical principle, when in fact it is better understood as an infraethical concept that can facilitate or hinder ethics but cannot substitute for it (Ott and Dabrock). They show how health technologies tend to render patients ever more transparent, turning their bodies and behaviors into streams of data, while the technologies and institutions that process that data remain opaque. Transparency appears Janus faced. The price of access to care is that one accepts being made legible to systems whose operations one cannot in fact inspect. For Ott and Dabrock, the central moral failure in such arrangements is not the absence of information per se but the absence of intelligibility, a term they take from Butler to name the condition under which persons can appear as recognizably human and thus as entitled to protection (Ott and Dabrock). Seen from this angle, the ethics documents of the artificial intelligence era did little to change who counted as intelligible within the system. They primarily refined how the system justified itself.
From within the industry, defenders of the status quo often appealed to complexity. Algorithms were too sophisticated to explain, competitive pressures too intense to permit full disclosure, user expectations too contradictory to satisfy. Ethics, they claimed, had to be pragmatic, balancing ideal aspirations with what was technologically and commercially feasible. There is a grain of truth here. Any serious moral philosophy acknowledges that we live amid constraints, and that political economies are not remade by decree. Yet the appeal to complexity became itself a technology of evasion. Cathy O Neil has shown in Weapons of Math Destruction how opaque mathematical models in domains such as credit scoring, employment, and criminal justice allowed decision makers to claim neutrality while in fact reproducing and amplifying social inequities (O Neil 3 to 5). The opacity of the model shielded it from contestation. In a parallel fashion, artificial intelligence companies invoked the inscrutability of deep learning systems as a reason to scale them, not to limit them. Ethics appeared in these narratives as a set of adjustable constraints on deployment rather than as a reason to question whether the underlying goals and incentives were themselves defensible.
One might object that there were, and are, genuine efforts to build alternative models. Many researchers in fairness and accountability, many activists and policy makers, pushed publicly for stronger regulation, participatory design, and robust protections for marginalized communities. There is no reason to doubt their sincerity. The claim of this chapter is not that nobody tried to act ethically, but that the dominant institutional configuration absorbed ethics into existing power structures. When corporate philanthropy funds academic chairs in artificial intelligence ethics while continuing to lobby against stringent privacy laws, when advisory boards are asked to issue recommendations but not to veto products, the function of ethics becomes ambiguous. As Avishai Margalit notes in his account of decent societies, the first demand of moral life is that institutions not humiliate the people who depend on them (Margalit 1 to 3). By that measure, any ethics architecture that preserves business models built on non consensual surveillance and behavioral manipulation fails at the most basic level. It may reduce reputational risk, but it does not prevent violation.
If one returns finally to Simone Weil’s language of attention, the hollowness of ethics as buzzword becomes even clearer. Weil insists that the capacity to give one’s attention to a sufferer is almost a miracle, because it requires suspending one’s own projects and fantasies long enough to perceive the other in their full reality (Weil, Waiting for God 105). By contrast, surveillance capitalism trained institutions to treat attention as a commodity to be captured and resold. Companies competed to fragment and occupy human attention, not to honor it. They built systems that could infer the vulnerabilities of users and deploy stimuli at the right times to keep them engaged, often exploiting loneliness, boredom, or insecurity. To name such manipulation as engagement, optimization, or user experience is to participate in the same moral laundering that transformed ethics into a marketing asset. The most intimate capacities of the person were rendered into levers.
What the collapse of the artificial intelligence bubble revealed, therefore, was not simply a mispricing of assets or an overestimation of technological capabilities. It revealed a deeper disjunction between the ethical language that institutions used and the moral realities they produced. An entire industry learned to speak of fairness, transparency, and accountability while building infrastructures that undermined autonomy, exploited vulnerability, and redistributed epistemic and political power upward. Ethics became a resource to be managed, not a vocation to be answered. The task that follows from this diagnosis, and that guides the rest of this book, is to reclaim ethics from this instrumentalization. That requires more than better guidelines or more polished mission statements. It requires a reorientation toward attention, intelligibility, and dignity as non negotiable commitments, and a willingness to allow those commitments to interrupt profitable designs. Only then can we begin to imagine technologies that do not merely perform moral concern, but that actually become sites where the human is protected rather than consumed.
Chapter Five
Interior Lives and Dignity: The Person Beyond the Data
The collapse of the AI bubble did not simply devalue a set of companies; it reordered the moral landscape in which human beings had been invited to understand themselves. For more than a decade, dominant narratives treated persons as bundles of signals, as optimization problems awaiting better data, as predictable trajectories in a space of behavioral probabilities. When the speculative edifice fell, what remained were not pure, unmediated selves waiting to be rediscovered, but exhausted subjects whose sense of interior life had already been shaped by infrastructures that translated experience into metrics. The central question for ethics after the collapse therefore cannot be whether we will have technology or not. The question is how we reclaim a robust conception of human dignity in a world where the reduction of persons to data has left deep marks on political, economic, and psychological life.
The capabilities tradition gives us the most coherent starting point for that reclamation because it begins neither from aggregate welfare nor from idealized contracts, but from a deceptively simple question: what is each person actually able to do and to be. In Creating Capabilities, Martha Nussbaum opens with the story of Vasanti, a woman seeking justice in Gujarat, in order to insist that any account of development must answer that question from the standpoint of a concrete life rather than from abstract indicators of growth (Nussbaum, Creating Capabilities 1). The capabilities list that follows is not a sentimental inventory of virtues but a juridical and political proposal: life, bodily health, bodily integrity, senses and imagination, practical reason, affiliation, relations to other species, play, and control over one’s environment are treated as threshold entitlements that political institutions must secure for every person as a matter of justice (Nussbaum, Creating Capabilities 2). In Frontiers of Justice, Nussbaum extends this approach to those whom traditional contractarian pictures routinely marginalize, namely people with disabilities, noncitizens, and nonhuman animals, and she does so precisely by treating them as bearers of interior lives whose dignity does not depend on their market productivity or their fit with an ideal of independent rational agency (Nussbaum, Frontiers of Justice 69). When we speak in this chapter about reclaiming the interior life, we are therefore not invoking a private realm beyond politics. We are describing the normative content of dignity as a demand that institutions protect and expand each person’s substantive opportunities to live, feel, think, relate, and act.
Once we set the question of dignity in these terms, the limitations of the AI era’s anthropological assumptions become clear. The speculative economy of intelligence that earlier chapters described depended on a thin model of personhood. Datafied infrastructures treated human beings as nodes that generated traceable preferences and measurable risks. Even well meaning ethical frameworks often conceded that model, focusing on consent checkboxes, privacy settings, or fairness constraints while accepting that the primary representation of a person would be a profile of attributes in a database. By contrast, the capabilities approach insists that the subject of justice is not a dataset but an embodied and situated life, whose value cannot be derived from its contribution to aggregate efficiencies. Amartya Sen’s account of freedom as both process and opportunity makes exactly this point. Freedom is not exhausted by the absence of interference; it requires that people have real capability to pursue valuable functionings, which in turn presupposes education, health, and social recognition as enabling conditions, not optional extras (Sen, Idea of Justice). To speak of dignity after the collapse is therefore to speak of the infrastructures that make interiority materially livable: schools that cultivate imagination, health systems that respect vulnerability, social protections that prevent destitution from becoming a permanent horizon.
Yet any appeal to dignity risks sliding into a generic humanism that floats above the concrete dependencies and asymmetries through which lives are actually lived. Here Eva Feder Kittay’s work on dependency forces a necessary correction. In Love’s Labor, Kittay argues that our theories of justice and personhood have been built around the fantasy of the independent subject who stands as the norm of citizenship, while relations of care and dependency are relegated to the margins as unfortunate exceptions (Kittay, Love’s Labor 23 to 49). Against that fantasy, she insists that human dependency is a structural fact, not a temporary deviation, and that those who provide intense care, often women, occupy positions of moral labor that standard liberal theories ignore. The interior life cannot be reclaimed if it is imagined as the reflective space of a self who has been abstracted away from caregiving, disability, and economic precarity. Instead, Kittay’s analysis shows that the content of dignity includes the right to receive care without humiliation and the right of caregivers to be recognized as citizens whose labor is a condition of everyone else’s agency. A post AI ethics that takes dependency seriously must therefore treat the time and affective energy of caregivers as central political goods rather than as invisible background resources.
This correction also reshapes how we understand autonomy. The AI era often treated autonomy as the capacity of individuals to manage their own data flows, toggle their privacy settings, or rationally consent to terms of service. Feminist ethicists have long shown how impoverished that picture is. Carol Gilligan’s In a Different Voice undermined the idea that moral maturity consists in ascending from particular relationships to universal principles; by listening carefully to women’s moral reasoning, she demonstrated that attentiveness to context and relationship can express a form of ethical intelligence no less demanding than rule based reasoning (Gilligan 72 to 79). Nel Noddings, developing an ethic of caring, similarly argues that the fundamental moral question is not “what rule applies” but “how can I respond as one who cares,” with the relation between one caring and cared for serving as the basic unit of analysis (Noddings 79). If we put these insights alongside Nussbaum and Kittay, autonomy begins to look less like self sufficient control and more like what we might call situated self direction: the capacity to shape one’s life projects in the presence of others who care and in institutions that do not exploit vulnerability. After the collapse of systems that equated autonomy with data sovereignty, we need an account that honors people’s need for privacy while refusing to make the ability to stand alone the measure of personhood.
The interior life, on this picture, is not a sealed interior cavity but a dense field of perception, memory, affect, and imagination that is always already relational. Iris Marion Young’s work on structural injustice helps to clarify why that field is so often misrecognized. Young argues that oppression in modern societies operates not only through overt exclusion but through background structures of decision making and distribution that position some groups permanently on the receiving end of risks and burdens (Young, Inclusion and Democracy). For those who live at such structural edges, the sense of self is shaped by repeated experiences of blocked agency, stigmatizing surveillance, or bureaucratic indifference. Charles Taylor’s account of modern identity, especially his analysis of how recognition becomes a condition for the experience of authenticity, similarly suggests that persons come to know themselves through webs of meaning and regard that they do not fully control (Taylor, Sources of the Self). To invoke “interior lives and dignity” after the AI collapse therefore cannot mean a return to an inner sanctuary untouched by power. It must mean a program for rebuilding social and technological structures so that people who have been systematically misrecognized can finally be seen and heard as the authors of their own lives.
Judith Butler’s influence, mediated here through Ott and Dabrock’s notion of intelligibility, deepens this point. In their argument about transparency in AI based health technologies, Ott and Dabrock note that many regulatory discourses treat transparency as the main ethical remedy, assuming that if patients understand how systems work, their autonomy will be protected (Ott and Dabrock). They counter that this form of transparency is infraethical: it operates at the level of procedural conditions while leaving untouched the deeper question of whether people appear within these systems as intelligible subjects or as objects of measurement. Drawing on Butler’s claim that to be intelligible is to be recognized as a bearer of a life that matters, they insist that the transparent system can still produce the transparent human, a person stripped down to data points and rendered available for unlimited observation. In the context of our argument, this means that reclaiming interior lives cannot be reduced to explaining algorithms. It requires a reorientation in which systems are designed so that individuals enter them as agents with histories, vulnerabilities, and projects, not as disaggregated variables. Intelligibility is an ethical and political achievement, not a default.
Margaret Urban Walker’s account of moral understandings as socially embodied practices helps to guard against the temptation to treat these insights as purely philosophical. Walker argues that moral knowledge lives in the narratives, habits, and shared expectations through which communities teach one another how to interpret harm, responsibility, and repair (Walker, Moral Understandings). When AI systems were treated as quasi authoritative moral assistants, many organizations outsourced those interpretive practices to pattern recognition engines, which silently translated complex situations into risk scores or engagement metrics. The result was not merely a technical misfire; it was a displacement of the shared work of moral understanding. To rebuild that work, we will need collective processes through which communities articulate what counts as dignity in concrete domains such as work, health, education, and care. This is where Sen’s emphasis on public reasoning as the heart of justice becomes indispensable. The content of capabilities cannot be fixed once and for all from above; it must be argued out, revised, and contested in open forums where those most affected have real voice (Sen, Idea of Justice). Interior lives become politically real when the people who live them can shape the norms that govern their possibilities.
At this point, one might object that any emphasis on interiority risks reinstating a liberal individualism that prioritizes personal fulfillment over collective responsibility. The worry is not unfounded. Romantic invocations of the inner life have sometimes served to distract from material injustice, encouraging privileged subjects to invest in self discovery while others shoulder the burdens of care and exploitation. The framework we are building, however, resists that diversion precisely through its dependence on capabilities, care, and dependency. Nussbaum’s capabilities list already includes affiliation and control over one’s environment, which tie dignity to social and political conditions rather than to private introspection (Nussbaum, Creating Capabilities 2). Gilligan and Noddings make care and responsiveness to others central to moral maturity, not optional supplements. Kittay explicitly argues that justice requires restructuring social and economic institutions so that caregivers and dependents are not relegated to poverty and marginal status (Kittay, Love’s Labor 23 to 49). When these lines of thought converge, interior life appears not as a private luxury but as a socially produced and collectively protected good. It flourishes only where labor, care, and recognition are distributed in less exploitative ways.
The more dangerous objection invokes technology itself. Some will argue that the practices of data collection and behavioral prediction have so thoroughly infiltrated our institutions that talk of interior life has become nostalgic. If the state, employers, platforms, and insurers can already infer patterns of attention, preference, and vulnerability from traces we cannot realistically withhold, what remains of any interior that can be ethically protected. Here the lesson of the collapse is again instructive. Earlier chapters showed that the AI bubble depended on a belief in total legibility, a belief that the right combination of sensors, models, and compute could render the human fully knowable. That belief failed not only because of technical limitations, but because persons continue to exceed the categories through which they are measured. Even the most invasive data regimes misread, misclassify, and overlook significant dimensions of experience. The task is not to restore a mythical opacity to all aspects of life, but to decide politically what kinds of opacity we will defend as conditions of dignity. Nussbaum’s insistence on bodily integrity, for instance, can be extended to include protections against certain forms of biometric and emotional surveillance; Kittay’s focus on dependency can ground limits on invasive monitoring in caregiving contexts; Young’s analysis of structural injustice can inform restrictions on algorithmic systems that reproduce stigmatizing visibility for already marginalized groups (Nussbaum, Frontiers of Justice 69; Kittay, Love’s Labor 23 to 49; Young, Inclusion and Democracy).
On this basis, we can name the positive project that will guide the rest of Section II. We call it critical personhood, not to valorize critique for its own sake, but to mark a conception of the person that is at once normatively thick and historically attentive. Critical personhood affirms, with Nussbaum and Sen, that each human life has a claim to capabilities and respect that cannot be traded away for aggregate gains (Nussbaum, Creating Capabilities 2; Sen, Idea of Justice). It affirms, with Gilligan, Noddings, and Kittay, that relations of care and dependency are constitutive of agency, not obstacles to it (Gilligan 72 to 79; Noddings 79; Kittay, Love’s Labor 23 to 49). It affirms, with Young and Taylor, that identity is shaped by recognition and misrecognition within structures of power, so that the boundaries of interior life are themselves political achievements (Young, Inclusion and Democracy; Taylor, Sources of the Self). And it affirms, with Walker, Butler, and Ott and Dabrock, that intelligibility and moral understanding are collective practices that can and must be redesigned when they have been captured by technologies that flatten difference (Walker, Moral Understandings; Ott and Dabrock). To speak of critical personhood is to refuse both the thin data subject of the AI era and the abstract, disembodied self of older liberal theories.
The rest of this section will develop the implications of this conception for identity, care, solidarity, and justice. For now, the central claim of this chapter can be stated plainly. Ethics after the bubble cannot content itself with procedural safeguards around systems that still treat people as bundles of data. Nor can it retreat to a generic language of human values that leaves existing structures intact. It must instead take as its primary object the interior lives of persons understood in their dependence, plurality, and vulnerability, and it must ask what forms of policy, design, and collective practice can protect those lives from being hollowed out by extraction and misrecognition. The failure of AI’s grand narratives has opened a space in which that question can be heard again. Whether we use that space to rebuild institutions that honor dignity will depend on whether we are willing to reimagine the human subject not as an input to systems, but as the one for whom and with whom any just system must be built.
Chapter Six
Difference and Identity: Beyond Algorithmic Uniformity
The collapse of the AI bubble did not only deflate valuations and embarrass predictions. It exposed something far more pervasive and less easily undone: a culture that had quietly accepted a model of the human in which difference was noise and identity was an optimization variable. For more than a decade, the dominant imaginary of machine intelligence treated people as interchangeable nodes in a graph, whose value could be inferred from patterns of past behavior and whose future trajectories could be nudged toward profitable ends. Personalization rhetoric masked a deeper homogenization. Recommendation engines promised uniqueness while converging users toward the same narrow band of monetizable tastes. Risk scoring promised objectivity while forcing heterogeneous lives into a small repertoire of predefined profiles. In the aftermath of collapse, what remains is a field of subjectivities that have been repeatedly translated into data types, scored through categories never chosen by those subjected to them, and governed through infrastructures that learned to treat difference as a problem to be managed rather than a condition of justice.
To understand why this particular form of uniformity feels both historically familiar and technologically novel, it is helpful to return to Frantz Fanon’s account of colonialism as a total project. Fanon insists that colonial power does not merely occupy land or extract resources; it seeks to reconstitute the very horizon of personhood, such that language, body, and world are reorganized around the colonizer’s image of the human (Fanon). When Fanon writes that to speak a language is to participate in a world, he describes an environment in which the colonized subject becomes legible only through categories fashioned elsewhere, categories that convert difference into deficit. In the age of AI, platform infrastructures and machine learning models quietly extend this logic. The “default user” around which many systems are designed still bears the imprint of a historically dominant subject, typically presumed to be Western, affluent, able bodied, and racially unmarked. Everyone else is interpolated in relation to that default, either as a deviation to be corrected or as an anomaly whose data can be harvested without seriously altering the governing template. The result is not simply bias in a narrow technical sense. It is a re installation of colonial habits of perception inside the everyday machinery of classification.
Ruha Benjamin names this convergence of seemingly neutral computation and entrenched racial hierarchy the New Jim Code, a configuration in which technical systems “hide, speed up, and even deepen discrimination, while appearing to be neutral or benevolent when compared to the racism of a previous era” (Benjamin 8). The phrase matters because it refuses the comforting belief that discrimination is a residue that technology will gradually wash away. Instead, Benjamin shows how design choices, training data, and institutional incentives work together to encode racialized and gendered assumptions into ostensibly generalized systems. Risk assessment tools in criminal justice, automated hiring platforms, predictive policing software, and credit scoring algorithms all promise even handed efficiency while reproducing long standing inequities in more opaque, faster, and more scalable forms. The central problem here is not only that certain groups are treated worse by these systems. It is that the systems themselves are built on an ontology in which persons are reduced to variables whose primary meaning lies in their predictive correlation with institutional goals. Difference appears as a factor in a model rather than as a claim to political voice.
Achille Mbembe’s account of necropolitics sharpens the stakes of such technical simplifications by shifting attention to how regimes decide which lives may flourish and which will be exposed to premature death. For Mbembe, the ultimate expression of sovereignty lies in “the power and the capacity to dictate who may live and who must die” (Necropolitics 11). This formulation does not only apply to spectacular forms of violence. It also clarifies how infrastructures that allocate risk, debt, exposure, and care distribute vulnerability along familiar lines of race, class, and geography. When credit scoring systems systematically assign higher default probabilities to residents of particular neighborhoods, when border technologies categorize some bodies as threats and others as legitimate travelers, when health algorithms assign lower risk scores to Black patients because past spending on their care was lower, they participate in a necropolitical ordering of the world. These systems rarely declare that some may live and others must die. Instead, they silently intensify the likelihood that some will be denied access, refused treatment, criminalized, or burdened with unpayable debt. Algorithmic uniformity becomes a technology of power precisely because it hides the political decision to treat some lives as standard and others as marginal.
Against this homogenizing logic, Black feminist thought has long articulated a different understanding of identity, one that begins from the experiences of those positioned at the intersections of multiple oppressive systems. Patricia Hill Collins famously analyzes “controlling images” of Black women, such as the mammy, the jezebel, the welfare mother, and the matriarch, as ideological instruments that both justify and reproduce race, gender, and class inequality (Collins, Black Feminist Thought). These images do not simply reside in media; they structure institutional expectations, shape employment opportunities, and organize welfare policy. In a datafied environment, controlling images migrate into the training corpora and labeling practices that shape machine perception. Search engine results that auto complete degrading stereotypes, content moderation systems that flag Black vernacular as abusive, and photo tagging systems that misclassify Black faces all demonstrate how computational models inherit and amplify controlling images instead of undoing them. The promise of generic fairness cannot address this problem, because what is at stake is not just unequal treatment within a stable taxonomy but the taxonomy itself, the very grid through which personhood is recognized.
bell hooks’s insistence that feminist theory must be written from the margin to the center further deepens this critique. hooks argues that early second wave feminism took as its subject the experiences of relatively privileged white women and universalized them, thereby erasing the realities of women of color, poor women, and women outside the metropole (hooks). The language of sisterhood became another homogenizing discourse that obscured hierarchies under the banner of shared struggle. The analogy to contemporary discourses of “the user” is precise. User centered design and human centered AI speak as if there were a singular human whose needs can be captured by aggregated metrics of usability and satisfaction. Yet in practice these models often track and prioritize those users who are already legible to dominant institutions. What hooks demands of feminist theory, a reorientation that treats marginal lives as epistemic starting points rather than exceptions, must be demanded of technical design as well. A system that aspires to justice cannot simply correct its parameters while retaining a generic user; it must be rewritten from the vantage point of those whom existing infrastructures have rendered peripheral.
Hannah Arendt offers a vocabulary that helps name what is destroyed when algorithmic governance treats difference as error. In The Human Condition she describes plurality as “the condition of human action because we are all the same, that is, human, in such a way that nobody is ever the same as anyone else who ever lived, lives, or will live” (Arendt 8). Plurality here does not refer to simple diversity in the sociological sense but to the ontological fact that a shared world is built out of distinct perspectives that cannot be collapsed into one comprehensive view. Political action, for Arendt, depends on this simultaneity of sameness and irreducible difference. Algorithmic systems that pursue optimization over large populations, especially when aligned with economic and security imperatives, erode plurality by incentivizing convergence. They reward behaviors that fit predictive profiles, penalize those that deviate, and increasingly mediate the spaces in which people appear to one another. Even personalization can function as anti plurality when it reduces encounter to the careful feeding of isolated individuals with content tailored to sustain engagement rather than to expose them to otherness. The problem is not that systems notice patterns. It is that they quietly reorganize the field of possible action in ways that narrow the space for genuinely plural beginnings.
Stuart Hall’s analysis of encoding and decoding in mass communication clarifies how power operates within such mediated environments. Hall shows that producers encode messages with preferred meanings aligned with dominant interests, while audiences decode them in negotiated or oppositional ways (Hall 128 to 130). The key insight is that communication is never a neutral transfer of information; it is a struggle over meaning within a structured set of positions. When machine learning systems learn from historically encoded media representations, they inherit this structure of dominance. They do not simply describe the world; they participate in re encoding it, offering a new set of preferred readings that are then operationalized in credit decisions, policing practices, hiring filters, and content rankings. Moreover, unlike traditional mass media, algorithmic systems do not straightforwardly present a text to be decoded. They present individualized, dynamically assembled environments that make oppositional readings more difficult to coordinate. The encoded viewpoint appears as the natural order of things because it arrives through computation rather than overt editorial choice.
Postcolonial and decolonial thinkers such as Walter Mignolo help articulate why these dynamics must be understood as part of a broader history of global hierarchy. In Local Histories Global Designs, Mignolo argues that modern knowledge systems have consistently elevated Europe as the site of universal reason while relegating other epistemologies to the status of local culture or tradition (Mignolo). The result is a form of epistemic coloniality in which conceptual frameworks and normative standards travel outward from the metropole and organize how other parts of the world are known. AI infrastructures extend this pattern by embedding these epistemic assumptions into data schemas, model architectures, and evaluation metrics. When facial recognition benchmarks are constructed primarily from images of lighter skinned populations, when natural language processing systems treat English as the implicit norm and other languages as resource constrained additions, when platform policies assume a set of legal and cultural categories drawn from North Atlantic democracies, the global design of computation reproduces colonial divisions of knowledge. Local histories and ways of knowing appear only as deviations that must either be translated into the dominant schema or ignored.
Kevin Donovan’s work on financial technology in East Africa offers an empirical image of how these abstract patterns play out in everyday life. In his account of mobile lending in Kenya, Donovan and Emma Park describe an emergent regime of “algorithmic intimacy” in which firms translate granular data about family life, social ties, and daily routines into credit scores and targeted offers (Donovan and Park 120 to 124). The intimacy here is not mutual recognition but asymmetrical exposure: lenders know their clients in ways that clients do not know lenders. The categories through which Kenyan borrowers are evaluated are developed largely elsewhere and imported under the banner of inclusion. Yet this inclusion often amounts to what Donovan and Park call predatory inclusion, a pattern in which marginalized communities are brought into financial networks on terms that deepen rather than alleviate vulnerability. Algorithmic models, tuned for profit and risk management, compress heterogeneous livelihoods into a narrow range of credit profiles that reproduce existing inequalities in access and default. The margins are not erased; they are operationalized as sites of extractive opportunity.
Ruha Benjamin’s exposition of the New Jim Code situates these examples within a broader analytic of discriminatory design. The New Jim Code is not only a matter of biased datasets or flawed algorithms. It is a schema in which anti Blackness and structural racism shape which problems technologists find interesting, which metrics they deem relevant, and which harms they consider acceptable collateral damage (Benjamin 30). This schema manifests when facial recognition tools with high error rates for darker skinned women are deployed anyway, when predictive policing systems concentrate surveillance in neighborhoods already overpoliced, when risk models treat previous incarceration as a neutral predictor rather than a symptom of racialized criminalization. Under these conditions, the promise of algorithmic uniformity, a promise of equal treatment under machine rules, becomes a cover for the continuation of unequal structures under a new technical guise.
If we take seriously this constellation of critiques, the task after the bubble cannot be to restore a purified version of AI that achieves neutral optimization. Instead, it must be to reimagine identity and difference in ways that refuse the reduction of persons to points in feature space. This does not require a naïve return to pre digital notions of the self. Fanon, Collins, hooks, and Benjamin all resist such nostalgia. Rather, it requires acknowledging that subjectivity is always historically situated, relational, and contested, and that any system which claims to know who people are must be accountable to those people’s own accounts of themselves. Intersectional frameworks that start from the experiences of those at the sharpest edges of oppression provide not only a more accurate description of social reality but also a more demanding standard for technical design. A lending platform designed from the standpoint of Kenyan borrowers trapped in cycles of debt will look very different from one optimized for investor yield. A content moderation system accountable to Black women who are routinely harassed online will prioritize different risks than one accountable primarily to advertiser comfort.
Finally, a post uniformity ethics of identity must recover something like Arendtian plurality as a design principle. If plurality names the condition in which humans share a world without collapsing into sameness, then attentive technologies must be built to protect and cultivate this condition rather than erode it. This would mean, at minimum, recognizing that there is no view from nowhere in either data collection or model building, making explicit the positionality of designers and the situatedness of training data, and opening interpretive space for those subjected to algorithmic decisions to contest, revise, or refuse the categories imposed on them. It would also mean designing infrastructures that foster encounters across difference rather than enclosing people within personalized corridors of content. Such a project does not harmonize easily with prevailing commercial incentives, which favor scale, predictability, and engagement. Yet if the collapse of the AI bubble has revealed anything, it is that an ethic of efficiency without plurality leads to brittle systems and wounded lives. To move beyond algorithmic uniformity is therefore not a matter of technical refinement alone. It is a struggle over the very meaning of the human, conducted in code, in institutions, and in the contested space where people insist, against every model that flattens them, that they are more than what the system can currently see.
Chapter Seven
Ethics of Care: Reclaiming Attention
When the AI bubble finally burst, commentators rushed to frame the collapse in terms of mispriced risk, speculative exuberance, and failed prediction. Yet underneath the financial language, a different wound had been quietly forming. What the era of attention maximizing platforms and predictive services corroded was not only trust in expertise or confidence in technical forecasts, but the everyday capacities through which people care for one another. Time was reorganized around notification streams, presence was fractured across competing feeds, and institutions learned to treat the inner lives of their users as a resource to be managed rather than a reality to be honored. In this sense, the crisis that followed the collapse was not simply epistemic or economic. It was a crisis of care, in which the basic ability to notice another person, to remain with their difficulty without converting it into data, and to act in ways that protect their dignity had been repeatedly subordinated to optimization.
The modern ethic of care first emerged as a philosophical counterpoint to abstract rule based morality, and that history remains instructive in the present moment. Carol Gilligan’s In a Different Voice appeared as a direct challenge to Lawrence Kohlberg’s model of moral development, which valorized a justice centered orientation grounded in rights, rules, and impartial principles. Working with interviews from women considering abortion and other life shaping decisions, Gilligan argued that these subjects did not simply fail to reach a higher stage of moral maturity; they articulated a different moral logic, one oriented toward preserving relationships, preventing concrete harm, and holding conflicting responsibilities in view at the same time (Gilligan). Rather than asking what universal rule applied, they asked what response would maintain connection without betrayal. Gilligan’s intervention matters for a post AI world because the AI era had quietly elevated a Kohlbergian ideal as its moral template. Technical systems, and the institutions that commissioned them, expressed their ethics in the language of fairness metrics, rights compliance, and impartial rule enforcement, while treating the relational context in which people actually live as secondary. Care ethics does not reject justice, but it insists that there is something morally significant about the situated work of tending to particular others that cannot be captured by formal procedures alone.
Nel Noddings deepened this reorientation by arguing that ethics begins not from hypothetical contracts among equal rational agents, but from the experience of natural caring. In Caring: A Feminine Approach to Ethics and Moral Education, she develops the figure of the one caring and the cared for, emphasizing that moral life grows from concrete encounters in which one person receives another’s need as a claim on their attention and action (Noddings). For Noddings, memory of having been cared for and the ongoing desire to sustain caring relations form the real ground of ethical response, whereas the language of duty and abstract obligation often appears as an attempt to reconstruct morality once these relations have frayed. The AI boom appropriated the vocabulary of care in a very different sense. Recommendation engines promised to care about what users liked, chatbots promised empathic conversation, and corporate manifestos announced that products had been designed with care for the user in mind. Yet the actual logic of these systems treated users less as participants in caring relations and more as sources of engagement whose psychological regularities could be monetized. The ethic of care that Gilligan and Noddings describe begins from listening and presence; the ethic of “care” that guided many AI deployments began from prediction and extraction.
Joan Tronto makes explicit that care is not simply a set of private virtues, but a political practice that structures institutions. In Moral Boundaries, she describes care as “a species activity that includes everything that we do to maintain, continue, and repair our world so that we can live in it as well as possible,” and she analyzes care through a sequence of phases and corresponding virtues: caring about as attentiveness to need, taking care of as responsibility, care giving as competent work, and care receiving as responsiveness to the perspective of those who are cared for (Tronto). The lesson is that good care is never simply a matter of benevolent sentiment. It is a structured practice requiring institutions that enable people to notice needs, take responsibility, act with skill, and remain open to critique from those they intend to help. The AI era repeatedly fractured this sequence. Systems often operated with massive stores of data about human behavior, yet they lacked real attentiveness to what counted as need from the standpoint of the affected person. Responsibility was displaced downward onto end users instructed to manage complex consent settings and privacy dashboards. Competence was equated with statistical performance that ignored broader harms. Responsiveness was undermined by opaque decisions that could be neither understood nor effectively appealed. Reclaiming care in the aftermath of collapse therefore requires more than new interfaces. It requires rebuilding institutions so that these four dimensions of care are structurally supported rather than short circuited.
Avishai Margalit offers a complementary lens by defining a decent society not as one that maximizes abstract justice, but as one whose institutions do not humiliate those under their authority (Margalit). Humiliation for Margalit involves giving people sound reason to consider their self respect injured, for example by treating them as less than fully human or ignoring their claims in ways that deny their standing. When one reads Margalit after tracing Tronto’s account of care, a simple but demanding connection emerges. A society that neglects the work of care, or delegates it to the least empowered under precarious conditions, inevitably generates humiliation. People experience their encounters with bureaucracies, platforms, and agencies as a sequence of small injuries to their dignity, especially when those systems purport to be neutral or benevolent. Algorithmic infrastructures intensified this pattern by introducing new forms of impersonal rejection. A person whose loan application is denied by an opaque model, whose welfare benefits are terminated by an automated fraud detection system, or whose content is removed by a platform without explanation is not only inconvenienced. They are humiliated by being forced to confront a power that sees enough of them to discipline but not enough to understand. Margalit’s criterion of decency therefore aligns with an ethic of care: institutions that care well do not humiliate, and institutions that chronically humiliate have, by that very fact, failed at care.
Sara Ahmed’s ethnography of diversity work shows how institutions can appropriate the vocabulary of care and inclusion while quietly preserving patterns of neglect. In On Being Included, she describes how diversity policies and committees often function as what she calls nonperformative speech acts: statements that announce commitment while leaving underlying structures untouched (Ahmed). Diversity workers, in her account, spend their days navigating a paradoxical environment in which the language of care, inclusion, and support saturates official documents, yet actual experiences of racism and inequality remain unaddressed. The institution declares that it cares, and in so doing creates the impression that the problem has been solved even as those who continue to experience harm are rendered complaint vectors rather than co authors of change. The AI ethics industry reproduced this pattern with remarkable fidelity. Statements of principle, fairness frameworks, and ethics guidelines proliferated, while the workers raising concerns about the harms of deployed systems often faced marginalization or dismissal. A genuine ethic of care in the post bubble era will require refusing this nonperformative pattern. Care cannot remain a word in policy documents; it must be instantiated in the redistribution of attention, authority, and resources toward those who bear the costs of technological decisions.
Eva Feder Kittay extends the ethic of care to the very structure of justice by insisting that any adequate political theory must take dependency work as central rather than peripheral. In Love’s Labor she argues that traditional accounts of justice, including John Rawls’s famous social contract model, implicitly assume an image of the citizen as an independent adult, free of disabling conditions and free of caring responsibilities that would limit their participation (Kittay). This image obscures the fact that every life passes through states of profound dependency, and that some lives are constituted by ongoing needs that require sustained care. Kittay therefore highlights dependency workers, often women and often poorly compensated, whose labor makes social cooperation possible yet is rarely recognized as a political concern. In the context of AI, dependency labor appears in several guises. Content moderators who absorb traumatic material so that platforms remain usable, data labelers who perform repetitive cognitive tasks to make models function, and caregivers whose work is increasingly squeezed by scheduling and monitoring technologies all represent forms of labor that sustain the technical systems others rely upon. An ethics of care requires that these forms of dependency work be brought into view as sites of justice rather than treated as incidental support.
The phenomenology of illness developed by Havi Carel adds another dimension by showing how vulnerability transforms perception and calls for specific forms of attention. In Phenomenology of Illness, Carel argues that illness is not only a biological condition but a fundamental modification of one’s way of being in the world, altering spatiality, temporality, and the felt possibilities of action (Carel). Illness reveals, with particular clarity, the ways in which institutional practices can either support or undermine an ill person’s efforts to live meaningfully. Elsewhere, she notes that illness can function as an invitation to philosophical reflection, “a call to explore one’s life, its meaning, priorities, and values,” and she warns that medical settings often generate epistemic injustice when clinicians discount patient testimony as unreliable or overly emotional (Carel 11). Technologies deployed in healthcare during the AI era often reproduced this pattern by centering data about disease while sidelining narratives of illness. Risk prediction tools, triage algorithms, and remote monitoring platforms prioritized measurable indicators, while the felt experience of fatigue, pain, anxiety, or stigma remained secondary. A care ethic informed by Carel’s phenomenology would insist that any system mediating health must be accountable to the ill person’s own account of their life. Attention here is not a matter of processing more data points; it is the disciplined practice of receiving a narrative that cannot be reduced to metrics.
Attention becomes a unifying thread across these thinkers. Tronto explicitly names attentiveness as the first virtue of care; Ahmed describes diversity work as a way of attending to what institutions prefer not to see; Kittay insists that just societies must attend to dependency rather than abstract from it; Carel sees phenomenological sensitivity as a training of attention to lived experience. The AI economy, by contrast, built its profitability on the capture and redirection of attention. Platforms studied how long users lingered over images, where their gaze fell on a screen, which notifications could interrupt them at what times. This attention was then sold, packaged, and used to fine tune behavioral predictions. In the process, attention was moved away from other human beings, from shared spaces, and from the subtle demands of care. It was drawn into circuits of consumption and performance that left little surplus for patient listening or sustained accompaniment. The ethic of care that must follow the collapse therefore cannot simply encourage people to be more caring in their private lives while leaving attention economies intact. It must treat the organization of attention as a matter of justice.
Michael Sandel’s account of the moral limits of markets helps describe what went wrong when attention was subjected to market logic. In What Money Cannot Buy he argues that the expansion of markets into spheres of life traditionally governed by nonmarket norms does not only raise questions of fairness; it can also corrupt the meaning of the goods at stake by crowding out the values appropriate to them (Sandel). When friendship becomes instrumentalized for networking, civic participation becomes a branding opportunity, or learning becomes a means to a salary, the goods themselves change. During the AI boom, markets entered domains of care under the promise of efficiency and personalization. Care work in hospitals, schools, and homes was increasingly organized around metrics of time saved, throughput achieved, and costs reduced. Digital platforms offered companionship, emotional support, and mental health check ins packaged as subscription services or engagement products. The value of attention shifted from being a way of honoring the reality of another person to being an input in a revenue model. Sandel’s argument suggests that reclaiming care will require not only redistributing resources but also re establishing zones of life where market norms do not rule. Attention in such zones cannot be bought or sold; it must be offered as a gift and protected as a shared capacity.
If one reads these strands together, a post bubble ethic of care begins to take shape as both a personal and institutional project. Personally, it involves cultivating the capacity to linger with the particular, to resist the impulse to convert every discomfort into a problem to be solved quickly, and to notice how one’s own goals and distractions can crowd out responsiveness to others. Politically, it requires designing infrastructures that do not reward constant self exhibition at the expense of presence, and that do not hide the labor of care behind the frictionless surface of automated services. Care ethics, in this register, is not sentimental. It is an exacting demand that societies direct their best technical and institutional ingenuity toward supporting those who give and receive care, rather than treating care as a residue beneath the real work of innovation.
To speak of reclaiming attention, then, is to propose a reordering of moral priorities after the collapse. In place of systems that track what people look at in order to sell that gaze, we can imagine systems that protect the time people need to attend to one another in families, neighborhoods, and workplaces. In place of ethics frameworks that begin from compliance, we can imagine frameworks that begin from the question of whether those most affected by a technology feel seen and heard within it. In place of a culture that measures the value of care by its efficiency, we can imagine one that measures institutions by their decency, asking with Margalit whether they leave those under their authority humiliated or upheld. The ethic of care that emerges from Gilligan, Noddings, Tronto, Margalit, Ahmed, Kittay, Carel, and Sandel does not offer a simple recipe for humane technology. It offers something both more modest and more demanding: a way of orienting attention so that human vulnerability is neither ignored nor exploited, and a way of judging institutions by how well they sustain the fragile, time consuming work of tending to one another in a world where collapse has made that work visible again.
Chapter Eight
Solidarity and Justice, The New Political Subject
The collapse of the AI bubble did not only expose a defective technology story. It exposed the poverty of the political subject that story presupposed. For a decade, public imagination was trained to picture social life as a field of isolated users, customer journeys, and individual risk scores, all of whom could be managed through behavioral nudges and predictive profiles. The citizen was quietly rewritten as a data point, the neighbor as a node in a network, the public as an aggregate of user metrics. What vanished in that translation was the older intuition that justice depends on a people who can act together, argue together, and hold institutions to account together. After the bubble, the question of justice returns immediately as a question of solidarity. Who is the “we” that can still appear in public, decide something, and bear responsibility for its consequences.
In this chapter I argue that any ethic adequate to the post AI world must be anchored in a renewed idea of the political subject as a bearer of solidarity rather than as an isolated chooser. The previous chapters have already prepared this ground. Chapter Five insisted that human dignity includes an interior life that cannot be reduced to data. Chapter Six argued that difference and identity are historically structured by race, gender, and colonial power, not neutral attributes that systems can harmlessly index. Chapter Seven shifted the moral lens from abstract obligation to concrete care, asking how people sustain one another through dependency. Solidarity and justice now gather these threads into an explicitly political register. The question becomes: how can institutions cultivate subjects who are capable of standing with others in the face of injustice, not simply consuming services or optimizing their own exposure to risk.
Martha Nussbaum’s work on education and democracy offers a direct way to name what was lost and what needs rebuilding. In Not for Profit she warns that when education is reduced to economic productivity, societies begin “producing generations of useful machines, rather than complete citizens who can think for themselves” (Nussbaum, Not for Profit 2). The AI boom amplified exactly this pattern. Education policy, corporate training, and even public culture were reshaped around employability, resilience, and entrepreneurial adaptability to technological disruption. The promised reward was competitiveness in a data driven economy. The hidden cost was a population discouraged from seeing itself as a source of public judgment and public resistance. Nussbaum’s insistence that the humanities and arts are indispensable because they cultivate narrative imagination and the capacity to “see the world through another person’s eyes” must be read, in this context, as a theory of solidarity, not as a nostalgic defense of culture. A society that has lost the ability to imagine the pain, anger, and hopes of those unlike oneself will find it nearly impossible to mount any collective response when systems fail.
If the AI bubble eroded the imagination of solidarity, it also intensified structures of injustice that had already fractured the “we” of democratic life. Iris Marion Young’s work on inclusion and structural injustice helps clarify this terrain. In Inclusion and Democracy she argues that justice cannot be reduced to the fair distribution of goods among already recognized individuals. Rather, injustice often takes the form of social processes that systematically place whole groups under threat of domination or deprivation, even when no single actor intends this result (Young, Inclusion and Democracy 92 to 102). This analysis maps directly onto the world that AI systems helped produce. Predictive policing algorithms that over target Black neighborhoods, credit scoring models that quietly disadvantage those without formal banking histories, and content moderation regimes that silence activists under the guise of neutral enforcement all express what Young calls structural injustice. They do not simply miscalculate; they reproduce group based vulnerability.
Young’s concept of communicative democracy is therefore not a decorative supplement to institutional design but a practical proposal for how solidarity can be formed under these conditions. She insists that democratic communication must welcome storytelling, greeting, and rhetoric alongside abstract argument, because those forms allow marginalized groups to articulate experiences that dominant frameworks ignore (Young, Inclusion and Democracy 77 to 80). In the wake of AI’s collapse, with trust in systems already frayed, this communicative pluralism becomes a condition of possibility for any renewed public. People who have been reduced to data categories must be able to reintroduce themselves in their own voices. Solidarity, on this view, is not a sentimental feeling of togetherness but a relation forged when structurally different groups hear one another’s accounts of harm and are willing to revise their own positions in response.
Amartya Sen’s account of justice deepens this picture by clarifying the role of public reasoning. The dominant theories that framed the AI era often treated justice as a matter of designing the right rules and letting systems optimize within them. Sen’s The Idea of Justice famously resists this approach. Instead of seeking a perfectly just institutional blueprint, he argues that justice requires comparative assessments of how actual arrangements treat people, guided by an “unrestricted public reasoning” that remains open to new evidence and argument (Sen 390; Sen, The Idea of Justice x to xix). Public reasoning here is not a ritual of expert consultation, nor is it the thin consultation exercises that many technology firms staged to legitimize their products. It is a continuous practice in which citizens, especially those exposed to injustice, challenge policies, present counterexamples, and insist that institutional designs be revised in light of their lived effects.
Once we take Sen’s emphasis on public reasoning seriously, the political subject after AI cannot be the passive recipient of decisions made in opaque technical forums. Nor can it simply be the individualized rights bearer who occasionally litigates against a platform. The subject of justice is a participant in an ongoing conversation about how institutions ought to be structured, a conversation that must remain open to the voices that have historically been excluded. Solidarity emerges when those with relative privilege recognize themselves as answerable to this wider audience, not only as private moral agents but as members of institutions that either support or undermine others’ capabilities.
Axel Honneth’s theory of recognition adds a further normative dimension. In The Struggle for Recognition he argues that modern societies depend on institutionalized forms of recognition in three domains: love in intimate relations, legal respect as bearers of rights, and social esteem for one’s contributions (Honneth 92 to 135). When recognition in these domains is denied or distorted, subjects experience not only material harm but an injury to their self relation. The AI era produced new modes of misrecognition. Automated moderation labeled certain dialects as abusive. Risk scores treated entire neighborhoods as inherently suspect. Workplace monitoring systems inferred distrust from every deviation from expected patterns. Each of these practices conveyed a message about whose presence counted, whose word could be believed, whose labor could be trusted.
In a post AI ethical framework, Honneth’s analysis implies that solidarity requires institutions that materially and symbolically recognize people as coauthors of the social world. Legal reforms that grant data protection and algorithmic transparency are important, but without corresponding changes in recognition they risk being experienced as hollow. If a community acquires the formal right to challenge a system yet still finds that its testimony is discounted, solidarity will fracture. The new political subject must therefore be conceived as engaged in struggles for recognition that seek not only inclusion within existing institutions but transformation of those institutions so that they genuinely register the experiences and claims of those previously marginalized.
Hannah Arendt offers a final conceptual anchor for this transformation. In The Human Condition she describes action as the human capacity to begin something new in concert with others and insists that plurality, the fact that “men, not Man, live on the earth and inhabit the world,” is the condition of political life (Arendt 7 to 8). The AI bubble’s fantasy was that technical systems could manage plurality by prediction and control, eliminating the unpredictability that makes politics risky. From an Arendtian perspective, this was not only a philosophical mistake but an attack on the very condition that makes freedom possible. The collapse of that fantasy opens the way to reimagine politics not as optimization of given preferences but as the unpredictable, sometimes conflictual, process of people acting together in public.
Solidarity, in this sense, is bound to natality rather than to conformity. It is the willingness to begin something together that none of the participants could have foreseen alone. When communities organize around environmental justice, police abolition, or data sovereignty, they do not simply demand better services. They enact a new public that had not previously existed, one that tests the limits of existing categories and forces institutions to respond. Arendt reminds us that such action is fragile and easily destroyed by bureaucracy, violence, or cynicism. Yet it is also the only way in which a genuinely common world can be rebuilt after the disillusion of technological salvation.
If solidarity is to have material force rather than remain an attractive concept, it must confront the concrete histories of race and class that shape who can act together in practice. Cornel West’s Race Matters, written in the early nineteen nineties after the Los Angeles uprisings, still speaks with uncomfortable clarity to the post AI world. West argues that the United States suffers from a “nihilistic threat” rooted in the lived meaninglessness experienced by many Black Americans, a condition produced by structures of poverty, violence, and cultural contempt, not by any deficiency of character (West 14 to 22). The AI bubble did not invent these structures, but predictive policing, risk based sentencing tools, and employment algorithms layered new forms of violence upon them. Solidarity that ignores this history risks becoming another universal rhetoric that floats above the actual distribution of suffering.
West’s vision of prophetic democracy suggests a different orientation. For him, democratic renewal requires what he calls “prophetic criticism,” a form of speech that names injustice without euphemism, rooted in love for those who suffer, and willing to unsettle entrenched power (West 122). Translated into the post AI context, this implies that the new political subject of solidarity must be capable of both analysis and denunciation. It must understand how data infrastructures reproduce racial hierarchies and be prepared to say so in public even when institutions prefer technocratic language about bias mitigation. Solidarity in this key means standing with those whose lives are treated as expendable by both old and new regimes of control and refusing the comfort of neutral vocabulary when the reality is structured abandonment.
David Graeber’s account of contemporary bureaucracy helps illuminate the institutional obstacles such a subject will face. In The Utopia of Rules he describes our era as one of “total bureaucratization,” in which public and private bureaucracies intertwine so thoroughly that they become the main mechanism for extracting profit and enforcing order (Graeber). AI systems were quickly folded into this bureaucratic landscape, promising frictionless compliance, automated eligibility determinations, and efficient risk management. Yet, as Graeber emphasizes, bureaucracies are sustained by the threat of force and by an often unspoken demand that citizens accept opaque procedures without real voice. Solidarity within this environment must therefore confront not only policies but the everyday micro humiliations through which people learn to lower their expectations of justice.
From this vantage point, the new political subject cannot be defined simply by possession of rights or by participation in formal democratic rituals. It must be composed through practices that cultivate capacities for public reasoning, for recognition, and for resistance to bureaucratic inertia. Educational institutions that take Nussbaum seriously would prioritize not only critical thinking but cross cultural narrative engagement, teaching students to recognize how their own welfare is bound up with that of distant others (Nussbaum, Not for Profit 95 to 98). Civic institutions informed by Young would redesign public hearings, citizen assemblies, and digital consultations so that storytelling and testimony are treated as legitimate modes of argument rather than as emotional noise (Young, Inclusion and Democracy 52 to 80). Legal frameworks guided by Sen would create enforceable spaces for public contestation of algorithms and regulations, with special attention to those who lack conventional resources to participate (Sen, The Idea of Justice 320 to 324).
At the same time, solidarity must resist the temptation to imagine a single, harmonious public that erases difference in the name of unity. The histories traced in earlier chapters show that appeals to national or human unity have often served to silence demands from colonized, racialized, or otherwise marginalized groups. Young’s account of structural injustice and group differentiated citizenship warns that justice sometimes requires institutionalizing differences rather than dissolving them, for example through mechanisms of group representation or targeted redistribution (Young, Inclusion and Democracy 133 to 164). Solidarity in this light does not mean that everyone agrees or feels the same. It means that groups recognize their entanglement in shared structures and take responsibility for changing them, even when their immediate interests diverge.
The political subject that emerges from these converging lines of thought is neither the rational chooser of liberal economic theory nor the passive data source envisioned in many AI governance schemes. It is a relational subject who understands itself as implicated in networks of dependency, recognition, and power. This subject is enabled, not exhausted, by institutions that respect its interior life, protect its basic capabilities, and make room for its voice in public reasoning. It is also a subject that remains unfinished, since new struggles will continuously reveal previously invisible forms of injustice. Arendt’s emphasis on natality implies that there will always be new beginnings, new movements, and new claims that exceed existing categories (Arendt 175). A just order therefore cannot aim at closure. It must institutionalize the possibility that the “we” of politics can be redefined by those who were formerly outside its boundaries.
The task of this chapter has been to recover the idea that solidarity is not a luxury to be pursued after technical and economic questions have been settled. It is the condition under which questions of technology and economy can be addressed at all without reproducing the injustices that made the AI bubble so damaging. By weaving together Nussbaum’s defense of democratic education, Young’s theory of structural injustice and communicative democracy, Sen’s account of public reasoning, Honneth’s analysis of recognition, Arendt’s concept of action and plurality, West’s prophetic democratic critique, and Graeber’s archaeology of bureaucracy, we arrive at a conception of the new political subject as a bearer of solidarity who is capable of remaking institutions from within and without. The subsequent sections of the book will translate this conception into normative frameworks and design principles. For now it is enough to insist that after the AI bubble, justice can no longer be imagined without this kind of subject, and that any attempt to rebuild our technological and political infrastructures that does not strengthen solidarity in this sense will only prepare the ground for the next cycle of disillusion.
Chapter Nine
Biopolitics and Surveillance: The Technologies of Control
When the speculative dream of artificial intelligence shattered, what remained was not a neutral technical landscape but an already assembled architecture of observation, prediction, and control that had long preceded the boom and would outlast it. The collapse of the AI bubble did not dismantle these architectures so much as expose their genealogy. The same infrastructures that had once promised optimization and personalized assistance now appeared as instruments for sorting populations, managing risk, and distributing vulnerability. In this sense, the post AI moment is not a void but a clearing in which the long history of modern power becomes visible again. To understand how, we have to return to the conceptual vocabulary that first named the intimate relation between bodies, life, and political sovereignty, and then trace how that vocabulary is transformed when computation and data saturate everyday life.
Michel Foucault’s account of modern power remains indispensable because it refuses the comforting fiction that surveillance is a late and accidental property of digital systems. In Discipline and Punish he describes the emergence of a disciplinary order in which power is exercised not through spectacular violence but through continuous observation, minor corrections, and the production of docile bodies within schools, factories, barracks, and prisons (Foucault, Discipline and Punish 195 to 228). The figure of the panopticon is not merely an architectural curiosity; it condenses a new rationality of power in which visibility becomes a technique for shaping conduct and for internalizing authority. The prisoner does not obey because a guard is present but because the possibility of being observed is continuous and unverifiable. Foucault insists that this reorganization of visibility produces a “micro physics of power” that diffuses through institutional life rather than remaining concentrated in a single sovereign center (Foucault, Discipline and Punish 26 to 27). If we transpose this analysis into the digital present, it becomes clear that algorithmic monitoring is not an aberration from liberal norms but an intensification of a long standing disciplinary tendency.
In his later work, Foucault names a further transformation. In the first volume of The History of Sexuality he argues that the classical “right of life and death” associated with sovereign power has been displaced by a political rationality that “fosters life or disallows it to the point of death” (Foucault, History of Sexuality 136 to 138). Power, he suggests, relocates from the moment of execution to the continuous administration of biological processes: birth rates, health indicators, longevity, and the circulation of bodies within labor markets. He distinguishes between an anatomo politics of the individual body and a biopolitics of the population, arguing that modern states seek both to discipline individual bodies and to regulate the statistical life of populations (Foucault, History of Sexuality 139 to 144). Biopolitics thus appears wherever institutions collect, compare, and manipulate information about populations in order to normalize behaviors and optimize flows. Digital infrastructures that track credit scores, movement patterns, consumption habits, or health data extend this biopolitical logic into finer and more continuous registers. After the AI bubble, what remains is precisely this apparatus of life management, stripped of its utopian rhetoric but still capable of shaping who receives care, who is exposed to harm, and whose risks are ignored.
Achille Mbembe’s notion of necropolitics sharpens this picture by insisting that the administration of life always includes a politics of death. In his seminal essay “Necropolitics” he argues that “the ultimate expression of sovereignty resides, to a large degree, in the power and the capacity to dictate who may live and who must die” and that to exercise sovereignty is to exercise control over mortality, defining life as the deployment and manifestation of power (Mbembe, “Necropolitics” 11 to 13). Biopolitics, in other words, cannot be understood solely as the benign optimization of life; it also constructs “death worlds,” zones where populations are abandoned to slow destruction or targeted for active annihilation (Mbembe, “Necropolitics” 39 to 40). When we place Mbembe’s analysis alongside Foucault’s, the post AI surveillance infrastructure appears as a mixed regime in which some populations experience finely tuned optimization while others encounter the negligent or deliberate production of unlivable conditions. Data centers, predictive models, and sensor networks become tools not only for managing traffic or optimizing supply chains but also for deciding which neighborhoods receive police attention, which borders are militarized, and which communities are left to environmental degradation.
The spread of digital surveillance cannot be separated from the consolidation of what C. Wright Mills called the power elite. In his mid twentieth century diagnosis of American society, Mills argued that effective power had concentrated in an alliance of political, economic, and military institutions, whose leaders occupy “command posts” that allow them to make decisions of national and international consequence (Mills, The Power Elite 3 to 5). Later commentators summarize his thesis by noting that the power elite is “composed of political, economic, and military men” whose coordination derives not from conspiracy but from shared institutional interests and social backgrounds (Mills, The Power Elite 276). The digital infrastructures that matured during the AI boom deepen this configuration. Intelligence agencies, platform corporations, and defense contractors share data, personnel, and objectives to an unprecedented degree. Even after speculative capital withdraws from AI startups, the infrastructures they helped justify remain embedded in this triad of state, corporate, and military power. The result is a surveillance order that cannot be reduced to any single actor, since authority is distributed across legal frameworks, technical standards, and commercial incentives.
Shoshana Zuboff offers a contemporary vocabulary for this configuration through her concept of surveillance capitalism. She describes it as “a new economic order that claims human experience as free raw material for hidden commercial practices of extraction, prediction, and sales,” emphasizing that behavioral data are appropriated without meaningful consent and transformed into predictive products for markets of business and government clients (Zuboff 8 to 10). This model, she argues, depends on the construction of “behavioral futures markets” in which the likely actions of individuals and groups become objects of speculation and control (Zuboff 16 to 20). During the AI bubble, this logic was obscured by narratives of intelligent assistance and personal convenience. After the crash, what is left is the basic economic machinery: continuous monitoring of behavior, algorithmic construction of profiles, and sale of predictions to those who wish to target, nudge, or govern populations. Zuboff’s analysis extends the Foucauldian insight that power operates through knowledge, but it also insists that this power is organized as a specific form of capital accumulation.
Sheila Jasanoff’s work on co production clarifies why these arrangements are so difficult to dislodge. In States of Knowledge she argues that scientific and technical orders and social orders are produced together; ways of knowing the world are inseparable from ways of organizing and controlling it (Jasanoff, States of Knowledge 1 to 4). The architecture of digital surveillance therefore cannot be treated as a neutral tool awaiting ethical guidelines. Regulatory agencies, corporate governance structures, national security doctrines, and technical standards have evolved in tandem with the systems they regulate. Legal categories such as “metadata,” “anonymized data,” or “legitimate interest” are not mere descriptions; they are instruments that stabilize particular distributions of visibility and vulnerability. When intelligence alliances share data collected by private companies, or when cities enter contracts with platform firms to manage mobility, what is being co produced is a political order in which surveillance is normalized as basic infrastructure. The collapse of AI hype may alter valuations and product names, but it does not dissolve this mutual reinforcement between knowledge practices and governance.
If we follow Foucault, Mbembe, Mills, Zuboff, and Jasanoff together, the technologies of control that persist after the AI bubble can be described as a layered regime of biopolitical, necropolitical, and economic power. At one layer, bodies are rendered legible through identifiers, location traces, and behavioral logs, allowing institutions to discipline conduct and administer access to work, credit, and services. At another layer, populations are sorted into those who are optimized and those who are neglected or targeted, producing death worlds at borders, in prisons, and in zones of environmental sacrifice. At yet another layer, behavioral futures are commodified and traded, so that the capacity to foresee and shape human action becomes concentrated in a small set of institutions. None of these layers can be understood in isolation, and none can be reduced to “technology” in a narrow sense. They are forms of sovereignty that operate through servers and sensors but are anchored in law, policy, and capital.
The concept of the surveillant assemblage elaborated by Kevin Haggerty and Richard Ericson helps explain why these forms of power are so pervasive and so hard to contest. They argue that contemporary surveillance operates less as a single centralized panopticon and more as a rhizomatic assemblage in which diverse surveillance systems converge by abstracting human bodies into flows of data fragments that can be reassembled for different institutional purposes (Haggerty and Ericson 606 to 610). Biometric identifiers collected at a border can later be matched with social media activity or credit information; medical records can be cross referenced with geolocation data for purposes far beyond health care. After the AI crash, governments and corporations may retreat from certain ambitious machine learning projects, but the underlying assemblage remains: a distributed, constantly expanding network of databases and sensors that can be recombined to pursue new objectives, including intensified policing, migration control, and financial exclusion.
One consequence of this assemblage is that lines between security, welfare, and commerce blur. Data gathered to prevent fraud can be repurposed to profile protesters; tools developed for targeted advertising can be tuned to influence elections; health surveillance deployed during a pandemic can become a template for permanent tracking. These shifts are often justified in the language of risk management and efficiency, but they also reconfigure who must live under continuous scrutiny. Marginalized communities, migrants, and political dissidents experience surveillance not as an abstract concern but as a tangible threat to employment, mobility, and bodily safety. Mbembe’s emphasis on the racialized character of necropolitics is relevant here: the creation of death worlds is not accidental but follows existing patterns of racial and colonial domination (Mbembe, “Necropolitics” 23 to 26). Digital surveillance does not invent these hierarchies, but it provides new means of enforcing them, including predictive policing systems that disproportionately target communities of color and border technologies that render certain lives permanently precarious.
The post AI landscape also exposes the limitations of procedural transparency as a safeguard. Making code or policies visible does little if the underlying distribution of power remains unchanged. Foucault already warned that modern power often operates not through explicit prohibition but through normalization; the norm becomes internalized and reproduced without overt coercion (Foucault, Discipline and Punish 177 to 184). In the context of surveillance, this means that individuals are encouraged to consent to monitoring through convenience, fear, or social expectation, while institutions retain wide discretion to interpret, combine, and monetize the resulting data. Zuboff’s analysis of surveillance capitalism underscores that the core problem is not secrecy alone but the unilateral appropriation of human experience and its transformation into behavioral surplus (Zuboff 93 to 102). Without addressing this asymmetry, transparency becomes another instrument through which institutions legitimize their practices without relinquishing control.
At the same time, the very density of contemporary surveillance invites renewed democratic critique. Jasanoff’s notion of co production implies that if science and social order are made together, they can also be remade together (Jasanoff, States of Knowledge 13 to 19). A Foucauldian analysis that once seemed diagnostic can become prescriptive, not by promising a world without power but by clarifying where counter power might be exercised. If surveillance systems are assemblages, they can be interrupted, reconfigured, or subjected to new norms. If necropolitical zones arise where certain populations are exposed to social or physical death, then a politics of resistance can focus on dismantling those conditions and insisting that the same infrastructures be repurposed for care rather than abandonment. The point is not to imagine a return to pre digital innocence but to insist that infrastructures of visibility answer to institutions that are themselves visible, contestable, and accountable.
C. Wright Mills closed The Power Elite with a call for what he termed a “sociological imagination,” the capacity to link personal troubles with public issues and to see biography and history together (Mills, The Power Elite 7 to 10). In the post AI era, such imagination must extend to the technical substrates of collective life. People cannot meaningfully deliberate about democracy or justice if they cannot perceive how their communications, movements, and decisions are continuously recorded and modeled. The technologies of control described in this chapter are not abstractions; they are the concrete means by which future chapters on data colonialism and democratic regeneration must proceed. Any ethics that aspires to remake the human after the bubble must therefore confront biopolitics and surveillance not as secondary topics but as the central terrain on which dignity, sovereignty, and life itself are negotiated.
The challenge, then, is to reverse the asymmetry that has characterized the AI era. Instead of transparent subjects and opaque infrastructures, we need institutions that can be interrogated and constrained while individuals retain zones of opacity necessary for thought, dissent, and intimacy. Foucault’s analyses show that such a reversal will never be a simple juridical reform, because power circulates through habits, expectations, and technical routines as much as through laws. Mbembe reminds us that life and death are already distributed unequally along racial and colonial lines. Zuboff exposes the economic incentives that reward extraction and prediction. Jasanoff reveals the mutual making of knowledge and order. Mills names the structural convergence of political, corporate, and military authority. Taken together, these thinkers suggest that after the AI bubble, the remaking of the human will depend on whether we can transform the very architectures of surveillance that the bubble helped finance. The rest of this section will build on that premise by following those architectures into the terrains of data colonialism and democratic regeneration, asking not only who is watched and governed but who has the power to decide what forms of watching and governing are possible at all.
Chapter Ten
Data Colonialism and Race: Remaking Territory and Commons
The collapse of the artificial intelligence bubble did not erase its infrastructures. Instead, it clarified what those infrastructures had become. Long before the crash, critics from the global South and from decolonial traditions warned that data extraction was not simply a new business model but the latest chapter in an older project of domination. Nick Couldry and Ulises Mejias propose the phrase “data colonialism” to name this formation. They argue that contemporary data practices combine the extractive logic of historical colonialism with the quantifying power of computation, so that “the appropriation of human life through data” becomes central to a new stage of capitalism (Couldry and Mejias, “Data Colonialism” 1 to 2). After the AI boom imploded, the violence of this arrangement became easier to see. What had been marketed as innovation revealed itself as a systematic remapping of territory, subjectivity, and the commons, in which the most vulnerable bodies and communities again provided the raw material for someone else’s future.
To understand why data colonialism matters, the chapter must begin with an older vocabulary. Aníbal Quijano describes coloniality as a pattern of power that outlives formal empire. In his account, European conquest did not only seize land and labor; it also produced a hierarchy of knowledges and races that sorted humanity into those who govern and those who can be governed without consent (Quijano 171 to 177). Frantz Fanon names colonialism a total project that saturates the psyche and the body, a world that divides itself into zones of security and zones of exposure, with different rules for colonizer and colonized (Fanon 3 to 5). Achille Mbembe’s concept of necropolitics extends this to the management of death, where sovereignty expresses itself by deciding “who may live and who must die” (Mbembe, Necropolitics 66). Data colonialism emerges at the intersection of these lineages. It does not replace territorial conquest with something more benign. It extends conquest into the informational fabric of life, encoding colonial distinctions in infrastructures that decide who is watched, whose labor is cheap, whose movement can be interrupted by a shutdown, and whose future is collateral damage.
Couldry and Mejias insist that data are never simply “there.” They must be appropriated. They describe “data relations” as new social relations that enable the extraction of data for commodification. Through these relations, “social life all over the globe becomes an ‘open’ resource for extraction that is somehow ‘just there’ for capital” (Couldry and Mejias, “Data Colonialism” 2). The point is not metaphorical. What historical colonialism did to land, minerals, and bodies, data colonialism does to communication, gesture, and attention. In the article’s middle sections, they describe how ordinary interaction begins to “contribute to surplus value” as platforms and devices capture every trace, transform it into data, and route it into circuits of prediction and control (Couldry and Mejias, “Data Colonialism” 12 to 15). Their book The Costs of Connection repeats the argument in wider scope, presenting data colonialism as “the colonization of human life by extracting data from it” so that everyday activities become part of a “continuous flow of value” for others (Couldry and Mejias, Costs of Connection 3 to 5). In the aftermath of the AI boom, this analysis allows us to see that predictive models were not simply technical artefacts that failed. They were instruments in an appropriation that treated whole populations as information mines.
Race is not an afterthought in this story. Ruha Benjamin frames contemporary automation as a “New Jim Code,” a set of technologies that “reflect and reproduce existing inequities, but that are promoted and perceived as more objective or progressive than the discriminatory systems of a previous era” (Benjamin 8). Safiya Umoja Noble demonstrates that search engines do not neutrally index the world but encode the racist histories and commercial incentives of those who design and monetize them. Her study of Google search results for terms like “Black girls” shows how corporate logics and advertising markets produce sexualized and demeaning representations that then circulate as seemingly natural knowledge (Noble 1 to 5). When these systems are exported to the global South and installed as the default interface to information, they transport the racial hierarchies of North Atlantic capitalism into new territories. Data colonialism is therefore not an abstract exploitation of “users.” It is a racial project that privileges some lives as sources of profit and relegates others to experimental material.
Concrete examples make this visible. Facebook’s Free Basics program presented itself as a philanthropic effort to provide free internet access in “emerging markets,” especially across Africa. In practice, it offered a zero rated walled garden dominated by Facebook and a small set of partner sites, while violating net neutrality and sidelining local content producers. Researchers and activists have described Free Basics as a form of “digital colonialism,” a “scramble” to capture new users by controlling the on ramp to connectivity and by channeling their data into Western corporate platforms (Nothias 1381 to 1383). In many contexts, including India and several African countries, civil society coalitions fought back, arguing that the program reproduced older colonial logics in which the metropole supplies infrastructure on its own terms and extracts value in return.
The labor that sustains AI offers another window into territorial and racial asymmetries. Investigations have revealed that major AI firms contracted intermediaries such as Sama to employ workers in Nairobi, Kenya, to label toxic content for systems like ChatGPT, paying wages as low as two United States dollars per hour to confront traumatic material including sexual abuse, torture, and hate speech. These workers describe the experience as a continuation of colonial patterns in which African labor absorbs the psychological and bodily costs of technological revolutions that enrich others. Brookings and human rights organizations now describe such “data work” and content moderation as a new “factory floor” of exploitation, where global South workers provide the emotional and cognitive labor that makes Northern AI products marketable while lacking basic protections or bargaining power. When Kenyan advocates describe this situation as “a continuation of slavery and colonialism,” they are not indulging in hyperbole; they are naming the way that control over infrastructure, law, and capital allows foreign firms to dictate the terms on which local life becomes raw material.
Territory in this chapter is not only physical. It includes the informational terrain on which communities move. Internet shutdowns provide a stark illustration. Access Now’s KeepItOn reports show that in 2023 governments around the world implemented at least 283 internet shutdowns, often during protests, elections, or armed conflicts, with the result that “tens of thousands of lives have been taken” under cover of digital darkness, from Palestine and Myanmar to Sudan and Ukraine. Shutdowns do not extract data; they weaponize its absence. Yet they function within the same colonial pattern. Authorities who control network infrastructure decide which populations are allowed to speak, organize, and document abuses. Those who suffer most from the withdrawal of connectivity are often the same groups whose data are extracted without consent in other contexts. Data colonialism therefore includes both over exposure to surveillance and forced invisibility when repression demands secrecy. In both cases, sovereignty over digital territory is exercised without meaningful participation from those affected.
Scholars of “digital colonialism” have traced these patterns beyond single platforms or scandals. Daniel Coleman describes a “twenty first century scramble for Africa” in which technology companies extract, analyze, and own user data as a new form of resource control, echoing the nineteenth century scramble for land and minerals (Coleman 365 to 369). Simeon Nothias reconstructs how Free Basics became a paradigmatic case in this debate, showing that activists from the global South were among the first to articulate digital colonialism as a framework, not simply as a metaphor. Rediet Abebe and colleagues complicate narratives of data sharing in Africa by showing how well meaning open data projects can reproduce colonial hierarchies when they treat African communities as data sources while allowing non African institutions to set research agendas, define problems, and reap benefits (Abebe et al. 1 to 5). Together, these analyses show that data colonialism is not only corporate overreach. It is a transnational configuration in which states, firms, and even some development organizations participate in an unequal ordering of who generates data, who interprets it, and who gains from the infrastructures that organize it.
Race structures every layer of this ordering. Mbembe’s necropolitics remains essential here. In his account, postcolonial regimes govern populations through differential exposure to death, often by harnessing the tools of late modern technologies (Mbembe, Necropolitics 66 to 79). Data colonialism extends this differential management into predictive registers. Automated credit scoring, predictive policing, welfare fraud detection, and biometric border control constitute a distributed apparatus that sorts persons according to risk and worth. Benjamin shows how such systems “fix” inequality in place by rebranding discrimination as optimization, rendering Black communities and other marginalized groups hyper visible to systems of control while erasing their claims to dignity and political voice (Benjamin 9 to 14). Noble’s work adds that even the basic informational environment is skewed, so that racially biased representations shape what is thinkable long before any policy or intervention is proposed (Noble 113 to 135). In the post bubble landscape, where many AI promises have collapsed, these infrastructures remain active. The bubble’s burst does not undo the racialized distributions of visibility and harm that it helped to crystallize.
If the story ended here, the chapter would only describe domination. Yet colonial histories also generate traditions of resistance that point toward another politics of data and territory. Indigenous data sovereignty movements insist that communities have the right to govern data about themselves according to their own values and protocols. Tahu Kukutai and John Taylor’s collection on Indigenous data sovereignty argues that self determination requires control over the collection, interpretation, and use of Indigenous data, not simply symbolic recognition (Kukutai and Taylor 1 to 5). The CARE Principles for Indigenous Data Governance propose that data initiatives must foreground Collective benefit, Authority to control, Responsibility, and Ethics, thereby refusing the assumption that openness or sharing are always inherently good (Carroll et al. 1 to 3). On the African continent, the African Declaration on Internet Rights and Freedoms articulates a regional vision of an open, secure, and rights respecting internet, while explicitly calling on states and companies to respect equality, privacy, and participation (African Declaration). These efforts do not offer a simple template for all contexts. They show that the governance of data and infrastructure can be anchored in collective political projects rather than in market logics alone.
Remaking territory and commons in a post AI world therefore involves more than regulating individual harms. It requires a transformation of ownership and authority over digital infrastructures. The language of the commons is not a metaphor in this regard. Elinor Ostrom’s work on shared resources showed that communities can design durable institutions for managing common goods without inevitable tragedy, provided that rules are clear, participatory, and enforceable (Ostrom 88 to 102). Applied to data, this suggests that community data trusts, municipal platforms, and regional cooperatives could hold and steward data on behalf of citizens rather than treating it as property of platforms or as an endlessly exploitable asset of the state. Experiments in municipal data trusts, open city platforms, and community networks already hint at such arrangements. When combined with Indigenous and decolonial frameworks, they point toward data commons that respect difference and self determination rather than subsuming all life under a single extractive standard.
In the context of race, this transformation must include material redress. If data colonialism has prolonged historical patterns in which Black, Indigenous, and global South communities supply both the raw material and the sacrificial labor of digital modernity, then decolonization cannot be achieved merely by new consent forms or audits. It requires redistribution: of infrastructure investment, of governance power, of the benefits generated by data driven research and innovation. That may involve legal frameworks that enshrine data sovereignty, reparative funding for communities whose data and labor were exploited, and strong protections against shutdowns and surveillance that disproportionately target marginalized groups. Access Now’s documentation of shutdowns underscores that such interventions are not theoretical. They are urgent requirements for any polity that claims to value human rights.
The chapter closes by returning to the question of the human that structures this book. Data colonialism remakes territory by annexing the everyday into an extractive order. It also remakes the commons by redefining shared life as a field for technical management. The racial and colonial genealogies traced here show that this order did not arise from innocent innovation. It emerged from long standing projects of domination that treat some lives as disposable and others as entitled to govern. Any attempt to imagine ethics after AI that ignores these histories will repeat their violence in subtler forms. The next chapter turns to democratic regeneration and institutional design. Its premise is simple. There can be no just future for technologies that manage knowledge and attention unless the people whose lives they touch are recognized as sovereign over their data, their infrastructures, and their collective futures.
Chapter Eleven
Democratic Regeneration: Decolonizing Institutions
The collapse of the AI myth did not simply reveal faulty models or overconfident engineering; it exposed a deeper constitutional problem in how contemporary societies authorize power, define membership, and distribute the right to speak. The institutions that governed AI were not aberrations sitting on top of otherwise healthy democracies. They condensed and automated what had long been true of the liberal order itself: that certain subjects were silently treated as standard bearers of reason and universality, while others were treated as data, risk, or noise. To imagine a future after the bubble is therefore not to add better regulatory layers to the same underlying contracts. It is to ask whether the very terms of political belonging that structured AI’s ascent can be repaired. Democratic regeneration, in this sense, means more than institutional reform; it names a project of decolonizing the basic agreements that have historically defined who counts, who decides, and whose harms are legible.
Charles Mills’s diagnosis of the modern polity as a racial contract clarifies what is at stake. In his opening pages he insists that “white supremacy is the unnamed political system that has made the world what it is today” and that this system operates not as an accidental deviation from the social contract but as its real content, governing epistemology, morality, and law at once (Mills 1 to 3). In the AI era, the racial contract acquired a computational extension. Algorithmic systems were trained on histories structured by that unnamed system, outsourced decisions were made by tools that could not see outside its categories, and institutional responsibility was diffused across technical infrastructures that presented themselves as neutral. When that regime collapsed under the weight of its own false promises, what came into view was not simply the failure of one industry but the durability of a contract that had always distributed recognition and vulnerability along racial, colonial, and economic lines.
If the racial contract has become infrastructural, then democratization cannot be confined to periodic elections or procedural compliance. Nancy Fraser’s notion of participatory parity provides a more demanding standard. In her essay on redistribution and recognition she defines justice as requiring social arrangements that allow all members of society “to interact with one another as peers” and she insists that this norm sets the “nonsectarian” core of a theory of justice (Fraser 7 to 9). She then specifies two conditions: first, that the distribution of material resources must be sufficient to secure independence and voice; second, that institutionalized cultural patterns must express equal respect and make possible equal social esteem (Fraser 9). Democratic regeneration, understood through this lens, means refounding institutions so that both conditions are satisfied for those who were structurally excluded in the AI order, including those whose labor, data, and territories were treated as extractable inputs rather than co equal sources of judgment.
Fraser’s account matters here because it refuses both a narrow economism and a narrow culturalism. Maldistribution and misrecognition are distinct yet intertwined, and no democratic renewal can succeed if it treats one as a secondary effect of the other (Fraser 8 to 12). AI governance failures were never confined to biased datasets or unfair contracts alone. They were anchored in status orders that made it possible for some groups to appear as natural experiment subjects, content moderators, click workers, or surveillance targets. Regenerating democracy therefore requires institutions that treat the norm of participatory parity not as an aspirational slogan but as a test: do the rules, forums, and infrastructures of decision making actually permit those who bear the greatest cost to act and speak as peers, or do they silently reproduce stratified orders beneath the vocabulary of inclusion.
Jane Mansbridge’s work on democratic practice helps to explain why this is so difficult. In her analysis of modern democracies she distinguishes between adversary models, which assume fixed interests and legitimate contestation among them, and unitary or consensus models, which presume a common good that can be discovered through deliberation (Mansbridge 4 to 10). Much of liberal institutional design has relied on the adversary picture, with parties, interest groups, and competitive elections seen as sufficient. At the same time, technocratic governance imported the worst of the unitary ideal by assuming that experts could approximate the public good without sustained engagement with those subjected to their decisions. AI governance intensified this pattern: corporate and governmental elites invoked public interest while consolidating power in opaque committees and proprietary architectures. Regeneration demands a different synthesis, one that accepts persistent conflict and divergent interests, yet still orients institutions toward mutual justification rather than market or executive fiat.
Chantal Mouffe’s account of agonistic democracy sharpens this point. For her, what she calls the democratic paradox lies in the tension between equality and liberty that underpins modern constitutional orders; these values cannot be finally reconciled, yet they must be held together in a framework that treats adversaries as legitimate opponents rather than enemies to be destroyed (Mouffe 96 to 103). Mouffe argues for what she names a conflictual consensus, where there is agreement on the basic ethico political principles of liberty and equality for all, yet ongoing disagreement about their interpretation and institutionalization (Mouffe 102 to 103). Translating this into the post AI context means recognizing that the goal is not a harmonious information society in which algorithms settle disputes, nor a purely procedural state that treats all preferences as interchangeable. Instead, democratic regeneration requires institutional forms that sustain conflict without allowing that conflict to slide into racialized, colonial, or exterminatory logics, precisely the logics that AI infrastructures often made easier to enact at scale.
If Fraser gives us a norm of parity and Mouffe names the ineradicable tension within democratic projects, Amartya Sen and Martha Nussbaum supply a complementary account of how public reasoning and capability can anchor institutional redesign. Sen’s Idea of Justice criticizes what he calls transcendental institutionalism and argues instead for a comparative approach focused on removing remediable injustices through public reasoning (Sen 1 to 3). In the chapter on democracy as public reason he treats democracy not as a fixed set of institutions but as an ongoing practice in which “public reasoning is central” to judgments about justice and in which the reach of that reasoning must not be confined to already powerful groups (Sen 321 to 337). Nussbaum’s work, especially Frontiers of Justice and Creating Capabilities, similarly resists purely resource based metrics, arguing that political orders should be evaluated by the real opportunities people have to live lives worthy of human dignity and that this requires constitutional guarantees of capabilities for all, particularly those historically marginalized (Nussbaum, Frontiers 69 to 75; Creating Capabilities 23 to 26).
When we read these projects together, democratic regeneration after the AI bubble cannot mean a technocratic optimization of existing frameworks. Instead, it suggests at least three intertwined commitments. First, institutional evaluation must be capabilities based and parity oriented; the relevant question is not whether an institution looks formally inclusive but whether those who were previously treated as data points possess material conditions and cultural standing sufficient to exercise voice. Second, decision making must be organized as public reasoning rather than private negotiation between firms and states; this entails opening AI and data governance to processes in which arguments can be made, contested, and revised in public forums where those most affected can participate. Third, the design of those forums must acknowledge agonistic conflict; procedures that sanitize disagreement in the name of consensus often simply reinstall dominant groups as arbiters of reason.
If the preceding chapters have shown how AI infrastructures extended biopolitical and colonial forms of control, then democratic regeneration must also be decolonizing in content and method. Mills’s racial contract, Fanon’s analyses of colonial subject formation, and contemporary accounts of data colonialism demonstrate that modern political orders have been built through exclusions that were juridical and epistemic at once (Mills 11 to 18; Fanon, Wretched 36 to 40; Couldry and Mejias 3379 to 3382). To decolonize institutions is not to search for a pure, pretechnological democracy free of these histories; it is to accept that current constitutional frameworks were written under conditions of conquest and racial domination and to treat that fact as a mandate for structural transformation. This involves, at minimum, institutionalizing forms of representation and veto that give colonized and racialized communities real power to shape, block, or redirect technological agendas that affect land, labor, and life.
Sasha Costanza Chock’s work on design justice shows how such a shift can begin from the level of practice. Design justice is described as an approach that “rethinks design processes, centers people who are normally marginalized by design, and uses collaborative, creative practices to address the deepest challenges our communities face” and insists that design decisions should be led by those who are most affected by them (Costanza Chock 1 to 5). In contrast to user centered design frameworks that still assume elite control over agenda setting, design justice treats marginalized communities as co designers and co governors, linking questions of interface and infrastructure to larger struggles over land, policing, and economic survival. When these principles are transposed into institutional architecture, they suggest that democratic regeneration requires not only citizen consultation but co production of rules, metrics, and enforcement mechanisms, especially in domains like data governance and AI deployment where harms have been concentrated among those already dispossessed.
Sheila Jasanoff’s account of co production in science and technology policy adds a further dimension. Her work has shown that scientific facts and legal norms are co produced; that is, they emerge together through institutional practices that distribute authority and credibility (Jasanoff 2 to 6). In the AI era, co production occurred in ways that privileged corporate research agendas and national security logics over public deliberation. A decolonizing democratic regeneration would invert this priority. Instead of asking how public concerns can be accommodated within existing technical roadmaps, institutions would begin with public and planetary needs and treat technical research as one element in a broader negotiation over collective futures. Co production would become a site of democratic struggle rather than an opaque background process.
Concrete proposals follow from this theoretical frame. First, data and AI governance should be re anchored in participatory bodies where those most affected by algorithmic systems hold real power. Examples include community data trusts governed by assemblies of local residents, worker councils with binding authority over workplace surveillance technologies, and Indigenous data governance frameworks that recognize collective sovereignty over information about lands and peoples. Such structures resonate with Fraser’s parity norm, since they directly address both material distribution and cultural status, and with Sen’s comparative justice, since they are oriented toward removing specific injustices rather than achieving a fully ideal order (Fraser 9; Sen 321 to 327).
Second, the practice of democratic deliberation must be restructured to correct the epistemic injustices described earlier in this book. Miranda Fricker’s analysis of testimonial and hermeneutical injustice showed how prejudicial credibility deficits and gaps in interpretive resources systematically silence marginalized knowers (Fricker 28 to 30). In the AI era, those deficits were often deepened when predictive models treated some communities as data sources but not as epistemic agents. Regenerating democracy means designing forums that not only invite testimonies from affected communities but also change procedures so that those testimonies reshape the questions, categories, and metrics by which decisions are made. This goes beyond representation in existing parliaments or committees; it requires reconfiguring issue framing, agenda setting, and evidentiary standards.
Third, institutional accountability must be tied to transnational norms that already articulate duties across borders, including the United Nations Guiding Principles on Business and Human Rights. These principles affirm that states have a duty to protect against human rights abuses by third parties and that corporations have a responsibility to respect rights and to provide remedies for harms (United Nations 13 to 24). In the post AI context, this implies that firms cannot treat algorithmic harms as externalities of innovation nor can states offload responsibility to voluntary ethics codes. Democratic regeneration would embed these guiding principles within binding frameworks for algorithmic impact assessment, cross border data transfers, and supply chain governance, with particular attention to communities in the global South and to racialized minorities within wealthier nations who have borne disproportionate costs.
Crucially, none of these reforms can be understood as neutral technical adjustments. They constitute a struggle over sovereignty. Achille Mbembe’s account of necropolitics, the power to dictate who may live and who must die, revealed how late modern sovereignty operates through differential exposure to death, precarity, and slow forms of violence (Mbembe 11 to 17). AI infrastructures made that power more tractable and scalable, for example by sorting migrants into risk categories, directing policing resources toward already over policed neighborhoods, or allocating welfare on the basis of opaque risk scores. Democratic regeneration must therefore include explicit mechanisms to return sovereignty to those who have been subject to these necropolitical arrangements. This may take the form of community veto powers over specific technologies, reparative redistribution funded by those who profited from AI extraction, and new forms of transnational representation for those whose lives are affected by decisions taken in distant corporate or governmental centers.
At the same time, the agonistic perspective warns against any illusion that a single institutional blueprint will permanently settle these questions. Mouffe’s insistence that conflict cannot be absorbed by rational consensus alone should temper the temptation to craft universal governance frameworks that claim to speak for all futures (Mouffe 101 to 103). Instead, a decolonizing democratic regeneration will likely involve overlapping and sometimes competing institutions: citizens assemblies, workers councils, Indigenous governance bodies, municipal cooperatives, and international treaties that must be continually renegotiated. The task is not to eliminate conflict but to structure it so that those who were historically cast as objects of governance become subjects of decision, while the resort to violence, racial hierarchy, and colonial domination is institutionally delegitimized.
The language of regeneration risks suggesting a return to an earlier, purer democracy. The histories that Mills, Fanon, and Mbembe recount make it impossible to sustain that picture (Mills 1 to 3; Fanon, Wretched 36 to 40; Mbembe 11 to 17). What is possible instead is a forward looking reconstruction in which institutions are deliberately re founded with an awareness of their own genealogy. This involves education systems that teach the racial and colonial history of law, governance experiments that normalize shared authority between state and nonstate actors, and constitutional reforms that embed capability and parity requirements directly into foundational documents. Nussbaum’s proposal for constitutional essential capabilities can be read in this light as a concrete way to inscribe the dignity of those previously excluded at the heart of legal orders (Nussbaum, Frontiers 69 to 75). Sen’s emphasis on public reasoning can similarly be institutionalized through requirements that significant technological or economic decisions undergo deliberative processes with real power to alter outcomes (Sen 321 to 337).
Democratic regeneration, then, is not a final stage after the AI collapse but an ongoing practice of decolonizing institutions through participatory parity, agonistic contestation, and public reasoning. It requires acknowledging that the bubble did not simply distort markets or misallocate capital; it entrenched a particular vision of who counts as a subject of knowledge and a bearer of risk. To undo that vision, institutions must be remade in ways that center those who have been treated as peripheral, treat their knowledge as indispensable, and grant them not only seats at the table but the authority to reshape the table itself. The remainder of the book will build on this chapter’s normative architecture, asking how ethical theories, epistemic practices, and design choices can sustain this demanding, unfinished project of democratic life after the bubble.
Chapter Twelve
Virtue, Freedom, and Justice: Rethinking Moral Philosophy
The collapse of the artificial intelligence dream did not only expose a set of misguided engineering projects or overstated business plans. It exposed a moral vocabulary that had, for decades, been content to borrow its substance from the very institutions and calculative logics it should have been evaluating. The rhetoric that framed algorithmic systems as neutral optimizers of social life relied on thin notions of value, instrumental accounts of rationality, and procedural theories of fairness that could be translated easily into code and compliance reports but hardly at all into lived moral experience. When the bubble burst, what failed was not only a technology but a style of moral thinking: an inheritance in which ethics had been reduced to rules, metrics, and risk management rather than understood as the ongoing formation of character, the protection of freedom, and the shared pursuit of justice. In the aftermath, the central task is not to add a more sophisticated checklist to future systems but to reconstruct the moral frameworks themselves.
This chapter returns to three families of moral thought that the AI era invoked in slogan form but rarely engaged with depth: the virtue tradition that asks what kind of people we become through practices and institutions; the traditions of freedom and justice that ask what is owed to each person and how social structures should be arranged; and the pragmatist insistence that moral reasoning is a living practice of inquiry in changing contexts rather than a static code. By revisiting Aristotle, Rawls, and Sen alongside Martha Nussbaum, Alasdair MacIntyre, Hilary Putnam, Christine Korsgaard, and Elizabeth Anderson, we argue that ethics after the bubble must be reassembled as a dense fabric of character, capability, and public reasoning, not as a menu of constraints for optimizing algorithms. The point is not to romanticize older theories but to see how their more demanding insights were flattened when imported into technical discourse, and how those deeper strands can now guide the remaking of institutions, infrastructures, and everyday life.
Virtue ethics is the most obvious casualty of the translation of moral philosophy into engineering requirements. In the virtue tradition, moral reflection is not primarily a matter of choosing among actions at a moment in time, but of asking what the human good consists in and what kind of life enables that good to be realized. Aristotle formulates this through the notion of a function: if there is a characteristic activity that is distinctively human, then the good for human beings will consist in excellent performance of that activity. He famously concludes that the human good is “activity of soul in accord with virtue” over a complete life, and not simply the accumulation of pleasurable episodes or successful transactions (Aristotle, Nicomachean Ethics 1097b22 to 1098a18). The virtues are not rules external to that activity; they are dispositions that allow reason, desire, and perception to be harmonized so that one reliably sees and chooses the right thing in the right way. They are taught and learned through upbringing, imitation, and participation in shared practices.
The AI era borrowed the language of virtue without its substance. Companies spoke of “trustworthiness,” “responsibility,” and “fairness” as desirable attributes of systems, and sometimes even of organizations, but treated these qualities as outputs of compliance pipelines rather than as sustained achievements of character, cultivated through institutional design and habits of attention. To put the point in Aristotelian terms, the dominant discourse treated virtue as an external constraint on a pre given aim, rather than as internal excellence in the very activity of designing, deploying, and governing technology. Once virtues are detached from the teleology of a whole life and from the social practices in which they are formed, they become decorative labels for what are in fact thin rule sets.
MacIntyre’s own diagnosis of modern moral culture helps explain this transformation. In After Virtue he argues that contemporary societies speak in the language of rights and utility while lacking a shared narrative about the human good or about the practices that sustain it (MacIntyre 2 to 5). Moral judgments, in his terms, come to resemble expressions of preference in a world governed by managerial expertise. What the AI boom added was a new species of managerialism that claimed not only to administer social life but to automate its evaluation. Moral concepts were imported into this sphere as branding devices for institutions that did not accept the disciplines of virtuous practice: confession without conversion, to use an older theological metaphor.
MacIntyre’s constructive alternative has two components that matter especially for a post AI ethics. First, he defines virtues as those acquired qualities “the possession and exercise of which tend to enable us to achieve those goods which are internal to practices” and which also sustain the practices themselves across time (MacIntyre 187). Practices in his sense include not only traditional arts or crafts but also complex forms of cooperative activity such as scientific research, medicine, and democratic deliberation. Second, he insists that the intelligibility of a life depends on the narrative unity that connects its episodes and commitments (MacIntyre 204 to 212). Institutions that fragment experience into isolated tasks and metrics threaten both dimensions: they make it harder for practitioners to sustain internal goods against external pressures and harder for individuals to live lives that are more than a sequence of disconnected roles.
A post AI virtue ethics, then, must treat technological design and governance themselves as practices whose internal goods include truthful representation, respect for persons, and the preservation of human capacities for judgment. That means evaluating engineers, executives, and regulators not predominantly by short term performance indicators but by the quality of the practices and communities they help to sustain. It also means attending to the narrative structures that lives can still bear: whether institutions permit people to pursue coherent projects that they can recognize as their own, or whether they are constantly re sorted, re scored, and re incentivized in ways that fracture agency.
If virtue ethics speaks to the formation of character and the integrity of practices, liberal theories of justice speak to the basic structure of institutions. In A Theory of Justice, Rawls famously proposes that “justice is the first virtue of social institutions, as truth is of systems of thought” (Rawls 3). His device of the original position invites us to imagine free and equal persons choosing principles of justice behind a veil of ignorance that hides their contingent advantages. The conclusion is a pair of principles that protect basic liberties, secure fair equality of opportunity, and regulate inequalities in income and wealth so that they benefit the least advantaged members of society (Rawls 52 to 65).
The AI boom drew selectively on this tradition. The language of fairness, equality of opportunity, and non discrimination was translated into algorithmic constraints and audit metrics. Yet Rawls himself was clear that his theory is not a set of technical criteria to be hard coded but a way of thinking about the moral basis of the basic structure. The principles of justice are meant to express “the appropriate distribution of the burdens and benefits of social cooperation” under conditions where persons regard one another as free and equal (Rawls 4 to 5). What collapsed in the AI era was not only adherence to those principles, but the idea that the institutional sphere itself should be answerable to a prior conception of justice articulated through public reasoning.
Sen’s sustained critique of what he calls “transcendental institutionalism” sharpens this point for our present context. In The Idea of Justice he argues that a theory that focuses on describing perfectly just basic structures is of limited help in dealing with the urgent injustices of actual societies (Sen 7 to 10). Rather than asking what an ideally just order would look like, Sen urges us to ask how we can reduce manifest injustice here and now by comparing lived arrangements and evaluating them in public reasoning. This comparative, realization focused approach is grounded not in an abstract measure of resources but in capabilities: effective opportunities to be and do what one has reason to value. The task of justice becomes the expansion of these capabilities through democratic deliberation and institutional reform (Sen 231 to 235).
The failures of AI era institutions can be read directly through this lens. Systems designed around narrow performance targets amplified some capabilities while undermining others, often for those already disadvantaged. To call a content ranking system fair because it avoids a specific statistical bias, while it simultaneously shrinks citizens’ capacities for informed deliberation or meaningful privacy, is to mistake a partial constraint for a standard of justice. A Sen inspired reorientation would insist that justice requires asking what people are actually able to do and to be within the socio technical environments in which they live, including their abilities to participate in public reasoning about those very systems. It would require that technological infrastructures be scrutinized not only by experts but by those whose capabilities they most directly shape, especially marginalized communities.
Nussbaum develops this capabilities approach into a more explicit theory of justice grounded in human dignity. In Frontiers of Justice she describes her project as seeking “a theory of social justice that can guide us to a richer, more responsive approach to social cooperation” for those excluded by traditional social contract models, especially people with disabilities, non citizens, and non human animals (Nussbaum, Frontiers of Justice 1 to 4). Against accounts that treat justice as a matter of distributing resources or formal rights, she argues that societies must guarantee each person a threshold level of central capabilities such as bodily integrity, practical reason, affiliation, and control over one’s environment (Nussbaum, Frontiers of Justice 70, 278). In Creating Capabilities she describes this list as a set of “political goals” that articulate “what it is to respect the equal dignity of all people” (Nussbaum, Creating Capabilities 23 to 25).
This threshold conception has two implications for ethics after AI. First, it treats dignity not as an abstract status but as a claim on institutions: to respect a person’s dignity is to ensure that she has effective opportunities to live a life worthy of that dignity. Second, it insists that capabilities are plural and in some tension with one another; there is no single metric that can collapse them into a single number. AI era practices, by contrast, often collapsed diverse human goods into uniform objectives such as engagement, efficiency, or risk reduction, then treated those metrics as proxies for well being. A capability focused justice would require that infrastructures be evaluated by how they affect a whole set of interlocking freedoms, especially for those whose capabilities have historically been denied by racism, sexism, disability, or colonial domination.
Virtue and justice traditions, however, can become abstract or formalist if they are not reconnected to the lived sources of normativity and to the epistemic conditions under which moral claims are made. Here pragmatist and post positivist work helps repair some of the damage done by a thin, technocratic picture of value. Hilary Putnam’s attack on the fact value dichotomy is instructive. In The Collapse of the Fact Value Dichotomy he argues that even the most seemingly factual judgments are pervaded by evaluative concepts and that our moral concepts, in turn, are shaped and refined through rational inquiry (Putnam 28 to 33). The attempt to cordon off value from rational argument leaves us saddled with an impoverished picture in which technology appears value free and ethics appears irrational. AI era discourse exploited precisely this picture: engineers claimed to be delivering neutral tools while moral concerns were relegated to public relations and ethics boards.
On the other side, Korsgaard’s work on the sources of normativity reminds us that moral obligations are not external impositions but expressions of our self conception as agents. In The Sources of Normativity she argues that reflective beings must be able to endorse the principles that guide their actions; to act well is to act in ways that can be justified to oneself and to others who are owed equal respect (Korsgaard 99 to 104). This Kantian insight cuts directly against architectures that treat users as manipulable points in a behavioral graph. Systems that rely on opacity or on the covert exploitation of cognitive vulnerabilities treat persons as objects to be managed rather than as agents who can share in the justification of the rules that govern them. A post AI moral framework must return to the thought that legitimacy depends on justifiability to those who are bound by the norms, which in practice means transparency of institutions, spaces for contestation, and respect for the capacities of those whom decisions affect.
Elizabeth Anderson’s work on equality gives this Kantian concern a social and political dimension that is indispensable in light of the distributive and racialized harms of AI. In “What Is the Point of Equality,” she argues that the aim of egalitarian justice is not to level all differences or to distribute resources according to a single formula, but to end “oppressive social relations” that undermine people’s standing as equal citizens (Anderson 289 to 295). She criticizes distributive theories that focus on individual holdings while ignoring hierarchies of status and power. The lesson for our context is direct. AI era ethics often concentrated on individual harms or on distributional fairness among users, while leaving intact the large scale structures through which corporations and states exercised power over information, labor, and attention. A justice framework informed by Anderson would insist that we ask whether socio technical arrangements humiliate, marginalize, or subordinate some groups to others and would measure success by the degree to which all can participate as equals in shaping common life.
Taken together, these strands point toward a conception of ethics after AI that is at once Aristotelian, Kantian, and democratic, while also being resolutely non ideal and attentive to plural histories. It would inherit from Aristotle the idea that questions about right action cannot be separated from questions about the good life and about the virtues that make that life possible. It would accept from Rawls and Nussbaum that institutions must be structured to secure basic liberties and capabilities for all, with particular attention to those excluded or injured by existing orders. It would learn from Sen that justice must be pursued through public reasoning about actual injustices and comparative improvements rather than through fantasies of perfectly just systems. It would echo Putnam and Korsgaard in rejecting the separation of value from reason and in grounding obligations in the self conception of agents who must be able to endorse the norms that bind them. It would incorporate Anderson’s insistence that equality is about ending oppressive relations, not simply equalizing shares.
Such a framework directly challenges the idea that ethics can be outsourced to technical procedures. It asks instead what kinds of persons, institutions, and forms of life we want to sustain under conditions of technological saturation. It asks whether engineers and executives are being formed into agents who can see and resist injustice, or into specialists who know how to optimize metrics under any value assumptions that pay. It asks whether our public cultures still have the capacities for reasoning together about justice, or whether those capacities have been eroded by decades of information architectures that reward outrage and distraction. Above all, it insists that any future design of intelligent systems must answer to standards of virtue, freedom, and justice that do not originate in those systems and cannot be collapsed into their objective functions.
If the first phases of AI ethics treated moral philosophy as a quarry from which to extract convenient slogans, the task now is to restore its structural role. That means reading Aristotle, Rawls, Sen, Nussbaum, MacIntyre, Putnam, Korsgaard, Anderson, and others not as lists of principles to be implemented, but as interlocutors in an ongoing argument about what it means to live well together. In later chapters we will connect this reconstruction to questions of epistemic justice, recognition, and the public good. Here the conclusion is simpler. Ethics after the bubble must be thickened again. It must return to the work of cultivating virtues, protecting capabilities, and building institutions that treat persons as agents with stories, commitments, and claims, rather than as data points in a landscape of speculation.
Chapter Thirteen
Epistemic Justice and the Politics of Knowledge
When the speculative economy of artificial intelligence finally imploded, what failed was not only an industrial fantasy but also a regime of knowledge about who counts as a knower, whose experience becomes evidence, and which voices are permitted to define reality. The collapse exposed that the AI era had never simply been about computation or data; it had been about the organization of credibility, the distribution of intelligibility, and the quiet violence of exclusionary epistemic practices. The earlier chapters have traced how financial speculation, moral hollowing, and surveillance infrastructures produced a world in which technical systems claimed to know more about people than people could authoritatively say about themselves. In the ruins of that claim, the central question becomes unavoidable. If a society is to repair itself after such a failure, it must ask not only what is true, but whose truths have been systematically discounted, mistranslated, or rendered unsayable.
Miranda Fricker names this terrain with rare clarity when she defines epistemic injustice as a wrong done to someone in their capacity as a knower, a wrong that is distinctively epistemic because it attacks a capacity that is integral to human value (Fricker 10 to 11). Her account distinguishes two primary forms. Testimonial injustice arises when prejudice leads a hearer to grant a speaker less credibility than they would otherwise deserve. Hermeneutical injustice arises earlier, when gaps in a communitys shared interpretive resources prevent some people from making sense of their own experience or from rendering it intelligible to others (Fricker 10 to 16). These are not limited to interpersonal mistakes. They crystallize social patterns in which some groups are persistently believed less, or not believed at all, and some forms of suffering lack the concepts required for recognition. During the AI boom, both forms were intensified. Testimonial injustice appeared when communities already marked as deviant or irrational had their testimony overridden by data profiles and predictive scores. Hermeneutical injustice appeared when new harms created by automated systems arrived faster than the language to describe them, leaving users with a sense of violation they could not yet name.
Frickers argument matters here because it insists that epistemic relations are never purely cognitive; they are saturated with social power. She shows that the practices of listening, believing, and interpreting are structured by what she calls identity power, those socially shared images of who people are that guide the distribution of credibility and intelligibility (Fricker 13 to 15). In the high period of AI, identity power was encoded into datasets, optimization targets, and classification schemes. The result was an epistemic order in which already advantaged groups were further amplified as reliable sources of knowledge, while racialized, gendered, disabled, and economically marginalized communities were often treated as noise, edge cases, or security risks. When the bubble burst, it was therefore not only a technical failure. It was the exposure of an epistemic hierarchy that had long depended on injustice. Any post AI ethics that hopes to be more than a rebranding of old authority must begin from this recognition.
Feminist standpoint theory deepens this analysis by refusing the comfortable fiction that all knowers are situated in the same way. Sandra Harding argues that the sciences and their epistemologies emerge from particular historical and political projects, and that science is politics by other means even as it produces reliable knowledge about the world (Harding 10 to 12). This is not a dismissal of science, but a demand for honesty about its social location. Because observers and observed occupy the same social, economic, and political plane, the most authoritative accounts of reality tend to reflect the interests of those with the power to fund research, set agendas, and define what counts as a worthwhile question (Harding 11 to 12). In the AI era, corporations and security states played the role of privileged observers. They defined which problems were worth solving and which harms were tolerable collateral. After the collapse, standpoint theory offers a different starting point. If knowledge is always socially situated, then communities historically subjected to technological violence possess indispensable perspectives on both the failure of AI and the possibilities for its transformation.
Patricia Hill Collins provides a model of what this reorientation can look like in practice. In the prefaces to Black Feminist Thought, Collins explains that she wrote in order to help empower African American women by documenting a collective knowledge that had been sustaining survival and resistance all along (Collins 10 to 11). She describes how everyday activities, from caring for children to mentoring students and participating in community groups, became sources of theoretical insight once she refused the false opposition between theory and the ordinary lives of Black women (Collins 8 to 9). For Collins, empowerment cannot occur in a social context marked by oppression and injustice; what matters is not domination over others, but the creation of knowledge that fosters both group empowerment and broader social justice (Collins 10 to 11). Her work thereby recasts epistemic authority as something that emerges from the lived struggles of those whom dominant institutions have treated as objects of study rather than as authors of interpretation.
Read together, Fricker, Harding, and Collins challenge every aspect of the AI era’s epistemic self image. Fricker reveals that many of the glamorous innovations of AI were built on unresolved testimonial and hermeneutical injustices; predictive policing tools that misclassify entire neighborhoods, content moderation systems that violently misrecognize minoritized speech, and decision engines that treat certain life trajectories as inherently risky all manifest identity prejudices translated into computational form. Harding reminds us that these systems did not arise from nowhere; they are products of highly specific alliances between technologists, investors, militaries, and bureaucracies, whose political projects shaped which forms of knowledge were amplified and which were quietly discarded (Harding 10 to 12). Collins shows that the very communities most harmed by such systems have long been producing rich epistemologies of survival, care, and critique. Any serious attempt at epistemic justice after the bubble must therefore move those subaltern epistemologies from the margins to the center.
This reorientation has institutional consequences. Sheila Jasanoff’s work on science and public reason demonstrates that democratic states do not simply receive scientific facts and then apply them neutrally; they actively construct forms of public reasoning, including the evidentiary standards and argumentative practices that citizens are expected to accept as legitimate (Jasanoff, Science and Public Reason). She calls these shared repertoires civic epistemologies, the culturally specific ways in which societies ask who may speak with authority, what counts as reliable evidence, and how disagreement is handled. In the AI era, civic epistemologies were quietly reshaped by the insistence that predictive models, proprietary datasets, and risk scores represented a higher, more objective form of knowledge than contested public narratives. After the collapse, Jasanoffs insight forces a hard question. If civic epistemologies have been colonized by a narrow technocratic elite, then epistemic justice requires more than transparency or open data. It requires redesigning the very procedures through which societies decide what counts as knowledge in the first place, including who sits at the table, which forms of experience receive presumptive standing, and how uncertainties are named.
Seyla Benhabib’s account of democratic discourse provides a complementary axis. In Situating the Self, she defends a non relativist ethics rooted in the idea that each person must be able to justify the norms that bind them within a public sphere structured by reciprocity and equal respect (Benhabib, Situating the Self). The point is not simply that everyone should be heard, but that the very criteria for rational argument must be open to transformation when confronted by marginalized experiences that earlier frameworks could not comprehend. In the context of post AI reconstruction, Benhabib’s work insists that epistemic justice is inseparable from democratic legitimacy. A polity that treats entire populations as data sources while denying them meaningful participation in the design and evaluation of the systems that govern their lives does not merely fail ethically. It fails epistemically, because it refuses the full range of perspectives needed to understand the consequences of its own actions.
Within this combined framework, epistemic injustice after the AI collapse can be named in three interconnected registers. At the micro level, it appears whenever individuals experience the familiar harms that Fricker describes: being systematically disbelieved, interrupted, or treated as unintelligible in the face of automated classifications and policy decisions. At the meso level, it appears in institutional procedures that codify whose testimony counts, whose models are funded, and which harms are deemed too anecdotal to merit investigation. At the macro level, it appears in civic epistemologies and public spheres that normalize the idea that complex technical systems stand above contestation, while historically oppressed communities are cast as subjects to be managed rather than partners in reasoning. Epistemic justice in this expanded sense requires intervention at all three levels. It calls for cultivating virtues of testimonial and hermeneutical justice in everyday life, redesigning organizational processes to recognize marginalized expertise, and remaking the public cultures through which evidence and argument acquire authority (Fricker 13 to 17; Harding 10 to 12; Collins 10 to 13).
Feminist standpoint theorists have long argued that those who occupy subordinated social positions often possess what Collins calls a collective knowledge that fosters empowerment precisely because it is forged in the struggle to survive intersecting oppressions (Collins 10 to 11). Harding notes that when we take seriously the idea that observers and observed share the same social field, a new kind of historical agent of knowledge appears, one whose situated awareness can reveal patterns that dominant perspectives overlook or suppress (Harding 11 to 13). In a similar register, Black feminist, Indigenous, disability, and decolonial thinkers have shown that communities treated as epistemically inferior by modern institutions often develop finely tuned sensitivities to risk, harm, and resilience that more powerful groups neither need nor desire to cultivate. The politics of knowledge after AI therefore cannot be satisfied with adding a token diversity of viewpoints to existing technocratic structures. It must recognize that epistemic advantage is often inversely distributed with social privilege, and that institutional authority must be reorganized accordingly.
This recognition destabilizes the familiar picture in which experts generate knowledge and publics receive it. From a standpoint perspective, expertise is always partial and dependent on background conditions that are themselves contested. Fricker’s analysis of hermeneutical injustice illustrates how entire conceptual frameworks may lack the resources to render certain experiences intelligible, with the result that those who suffer are deprived not only of recognition but also of the tools needed for self understanding and political action (Fricker 15 to 16). When AI systems were deployed in criminal justice, welfare, housing, and employment, people often experienced harms that did not fit existing vocabularies of discrimination, because the mechanisms of injury were distributed across code, policy, and institutional design. The belated emergence of terms such as algorithmic oppression, data colonialism, or surveillance capitalism should be read as attempts to repair hermeneutical gaps that had already done real damage. Epistemic justice in this sense entails a deliberate commitment to conceptual innovation, guided by those who have had to live inside such gaps.
The politics of knowledge that emerges here is not a simple inversion where marginalized groups are declared infallible and experts are dismissed. Harding warns that both conventional science and its countercultural critiques contain progressive and regressive tendencies, and that the task is to develop conceptual frameworks rich enough to think what appear at first to be contradictory truths: that science is both politics and a source of reliable knowledge, that feminism can both challenge and reproduce domination, that subaltern standpoints can both illuminate and obscure (Harding 10 to 12). Epistemic justice therefore involves ongoing, conflictual negotiation over the credibility of claims, not the replacement of one unquestioned authority with another. What changes is the background distribution of suspicion and trust. Instead of automatically privileging the views of those closest to capital and state power, institutions must cultivate structural habits of deference to those who have historically borne the costs of epistemic error.
Jasanoff’s notion of civic epistemologies suggests practical directions for this transformation. If societies already possess culturally specific repertoires for assessing knowledge claims, then those repertoires can be redesigned. Public hearings on the deployment of complex technologies can be organized in ways that foreground testimony from affected communities, with expert models treated as one form of evidence among others rather than as unquestionable benchmarks. Funding structures for research and development can be reoriented toward questions posed by marginalized groups. Regulatory bodies can embed mechanisms that require systematic attention to hermeneutical gaps, asking not only which harms have been documented, but which experiences still lack adequate names. More fundamentally, civic education can be reimagined so that citizens are trained not only to consume expert knowledge but also to interrogate its conditions of production and to contribute their own situated insights.
Benhabib’s emphasis on public justification adds a normative spine to such reforms. If norms are to bind free and equal persons, then they must be defensible to all affected through processes that respect their capacities as knowers and speakers. In a world reshaped by AI and its failure, this means that technological infrastructures cannot legitimately govern people who have been systematically excluded from the discourses in which those infrastructures are evaluated. Epistemic justice becomes a condition for democratic authority, not a charitable supplement. When entire communities are treated as data sources without meaningful access to the arenas in which data practices are debated, the resulting policies lack not only moral legitimacy but epistemic adequacy, because they cannot draw on the full range of relevant knowledge about their consequences.
Within the architecture of this book, Chapter Thirteen therefore performs a double task. Conceptually, it brings together Frickers analysis of epistemic injustice, feminist standpoint theory, and theories of public reason to articulate a thick account of epistemic justice as both a virtue of persons and a property of institutions. Politically, it argues that the remaking of the human after the AI bubble depends on a radical redistribution of epistemic authority. The human cannot be re centered simply as an abstract bearer of rights. It must be re situated as a plurality of differently located subjects whose capacities to know and to speak have been historically organized by power. The chapters that follow will turn to recognition and moral attention, but they presuppose the terrain mapped here. Before one can attend to another, one must first accept that the other is a knower, that their experience has something to teach, and that the structures within which we think together must be rebuilt so that such teaching becomes possible.
Chapter Fourteen
Recognition and Moral Attention: The Ethics of Being Heard
The preceding chapter argued that the collapse of the artificial intelligence order must be understood as an epistemic failure, a regime of knowledge that decided in advance who could be believed and what kinds of harm could be named. Once that order falters, a further question presses forward. It is not enough to correct patterns of credibility or to refine our vocabularies for injustice. We must ask what it would mean for persons to be recognized as subjects whose voices possess standing before institutions, and what kind of moral attention would be required for such recognition to become more than rhetoric. The central claim of this chapter is that ethical life after the bubble depends on an architecture of recognition and attention in which people are not simply visible to systems but are addressable and heard by other subjects through those systems. The aim is not a sentimental celebration of empathy, but a precise account of how acknowledgment, intelligibility, and attention become structural obligations rather than private virtues.
Axel Honneth’s theory of recognition provides one indispensable starting point. In The Struggle for Recognition he argues that personal identity and self respect are formed through patterns of recognition in three domains. In intimate relationships, love and care provide the basic assurance that one’s needs can be voiced and answered. In legal relations, the attribution of rights recognizes each person as a bearer of moral autonomy who must be respected as an equal partner in social cooperation. In the sphere of social esteem, shared cultural values confer recognition on particular contributions and ways of life, enabling individuals to experience themselves as worthwhile members of a community (Honneth 92 to 135). When recognition is withheld or withdrawn in any of these domains, the result is not only psychological hurt but what Honneth calls moral injury, a denial of the conditions under which persons can develop a practical relation to themselves as capable and worthy (Honneth 129 to 134).
During the AI era, institutions frequently invoked rights and fairness while leaving these recognitional conditions unaddressed. Automated decision systems might avoid certain statistical forms of discrimination while still rendering people practically unintelligible to the agencies that governed their lives. A welfare recipient could be denied benefits by a risk model without any forum in which her narrative could be received as reason, a gig worker could be deactivated by a platform’s fraud detection without any space in which to confront another subject and demand justification. In such cases, the legal form of recognition persisted in principle, yet the lived experience of being recognized as a rights bearer or as a valued contributor evaporated. Honneth’s framework allows us to name this as a structural form of misrecognition. Even where explicit rights are acknowledged, institutions can still inflict moral injury by undermining the very practices through which persons ordinarily have their claims heard and answered.
Judith Butler’s work on precarity and grievability clarifies how such misrecognition is distributed along lines of power. In Precarious Life she argues that every life is vulnerable to injury and loss, but that not every life is apprehended as a life that can be grieved, protected, or even counted (Butler 19 to 25). Frames of perception, often sustained by media, law, and institutional routines, decide in advance whose suffering will register as a public event and whose will pass as background noise. Butler notes that when a population is constituted as threat, collateral, or mere statistic, its members’ deaths are less likely to be marked, mourned, or investigated; they become, in her terms, ungrievable (Butler 32 to 39). The ethical question for her is not simply whether we can acknowledge that others are vulnerable, but whether we can allow our own frames of reference to be disrupted by their claims upon us.
In the AI epoch, frames of grievability were coded into the very architectures of information. Predictive policing systems marked certain neighborhoods as inherently risky, news feeds elevated some stories and buried others, and risk models assigned higher tolerance for harm to those whose lives were already discounted by racial and economic orders. When decision making is mediated through such frames, it is not only physical vulnerability that is unequally distributed. The basic capacity to be heard, to have one’s voice register as a claim on others, is stratified. Butler’s account of precariousness therefore intersects Honneth’s theory of recognition at a crucial point. Recognition is not a purely interpersonal gesture of esteem. It is patterned by frames that make some lives intelligible and others disposable. To design for recognition after AI is thus to design for the disruption of those frames, so that those who were previously ungrievable or inaudible become subjects whose speech exerts ethical pressure.
Tabea Ott and Peter Dabrock’s work on transparency and intelligibility in health care technologies positions this insight within debates about technological governance. In their analysis of artificial intelligence in clinical contexts they argue that calls for transparency are often infraethical, meaning that they address the preconditions for ethical evaluation without yet securing ethical relations themselves. Making an algorithmic system more interpretable may be necessary in order for professionals to scrutinize its outputs, but transparency by itself does not guarantee respect for patients’ dignity if their voices remain peripheral to the design and use of the system (Ott and Dabrock 3 to 6). For them, the key question is not only whether models are open to audit, but whether human subjects can appear as intelligible to the institutions that rely on those models. Their invocation of intelligibility draws explicitly on Butler’s work, insisting that ethical life requires that people be recognizable as bearers of claims that can be heard and answered, not simply as entries in a database.
The concept of infraethics is helpful here because it allows us to distinguish between conditions that make ethical evaluation possible and the substance of ethical practice. In the AI boom, transparency was frequently treated as the final answer to ethical concerns. Publish the model card, document the dataset, disclose error rates, and the institution could declare itself responsible. Ott and Dabrock remind us that these are only background conditions. The ethical event arrives when persons address one another across asymmetries of power and when institutions respond not only to metrics but to speech. In other words, recognition is not achieved when a system is explainable in principle. It is achieved when those affected can confront others in forums where explanation can be demanded, contested, and revised, and where the explanation itself is constrained by a duty to treat the other as a subject of justice rather than as an object of management.
Simone Weil’s reflections on attention deepen this distinction from another angle. In her essay on the use of school studies she describes attention as a form of waiting, a disciplined openness in which one suspends the rush to solution in order to receive the reality of another person or problem (Weil 105 to 110). For Weil, attention at its highest degree becomes a form of prayer, not because it is addressed to a particular doctrine, but because it consists in the patient refusal to substitute one’s own projections for the presence of what is actually there. To pay attention is to allow the other to appear, in her words, without attempting to appropriate or dominate that appearance. If we transpose this into a secular register, attention becomes a moral act in which we withhold the impulse to classify, fix, or exploit another’s situation long enough to be instructed by it.
Weil’s conception matters in our context because the AI era was characterized by the opposite habit. Technical systems were trained to classify as quickly and exhaustively as possible, to turn every gesture and pattern into a prediction that could be acted upon. Human agents learned to trust these classifications in advance of hearing from those whom they concerned. Recognition requires a different posture. To attend to a person is to delay closure, to permit their self presentation and speech to modify one’s initial categories. Ott and Dabrock’s intelligibility, Butler’s demand that frames be open to disruption, and Honneth’s emphasis on the formative role of recognition all presuppose this capacity for attention. Without it, even the most transparent system will simply confirm what institutions already think they know.
Joan Tronto’s ethic of care offers a further articulation of attention as a public practice rather than a private virtue. In Moral Boundaries she defines care as a species of activity that includes everything we do to maintain, continue, and repair our world, so that we can live in it as well as possible (Tronto 103 to 104). She distinguishes several phases of care: caring about, in which we come to recognize that someone has a need; taking care of, in which we assume responsibility for meeting that need; caregiving, in which concrete work is done; and care receiving, in which we attend to the response of the cared for and adjust our actions accordingly (Tronto 105 to 110). The last phase is crucial. Without listening to the response of those we intend to help, care becomes paternalism or domination rather than a relation of mutual recognition.
Applied to the governance of technology, Tronto’s schema suggests that institutions must not only identify harms and assume responsibility for mitigating them. They must create conditions in which those affected can speak back and have their responses taken as authoritative data for future action. A risk assessment that never leaves the modelers’ desks, an ethics board that never meets with communities subject to surveillance, or an oversight mechanism that does not track whether policies are experienced as just, all fail at the phase of care receiving. They therefore fail at recognition. Moral attention, in this sense, is institutionalized when feedback from those affected is not treated as optional input but as a constitutive moment in the cycle of ethical action.
Stanley Cavell’s distinction between knowing and acknowledging gives this point yet another register. In The Claim of Reason he argues that in ordinary life the response called for by another’s suffering is not a demonstrative proof that we have perceived it correctly, but an acknowledgment that expresses our readiness to be addressed and obligated by it (Cavell 340 to 345). To refuse acknowledgment is not simply to express doubt about a fact; it is to withdraw from relation, to deny the other’s standing as one who can make a claim upon us. Cavell famously observes that skepticism about other minds is less a theoretical puzzle than a moral temptation to deny the demands that others place on us (Cavell 329 to 340). When one says that another’s pain is exaggerated, invented, or irrelevant, one is not merely mistaken about evidence; one is refusing a form of moral attention.
In the context of automated systems, this refusal can become routinized. A person presents testimony that a model has miscategorized them, that a risk score has destroyed their livelihood, or that a content moderation system has silenced their political speech. The institution responds with a reference to aggregate error rates or to the impossibility of manual review. The fact of harm is acknowledged as a data point yet never rises to the level of acknowledgment in Cavell’s sense, because the person’s voice is not treated as a source of obligation. The ethics of being heard requires that institutions reclaim the distinction between knowledge and acknowledgment. Technical expertise can clarify what has happened, but only acknowledgment can inaugurate the relation in which repair or transformation becomes possible.
Bringing these strands together, we can now say more precisely what recognition and moral attention demand after the collapse of AI. First, they require that persons appear in institutional life as more than objects of prediction. Honneth insists that recognition confers on individuals a practical self relation in which they can understand themselves as loved, as bearers of rights, and as socially esteemed contributors (Honneth 129 to 135). That self relation cannot emerge where decisions that matter most are made in ways that do not permit narrative, contestation, or response. Second, they require that frames of perception be open to transformation by those who suffer. Butler’s analysis of grievability shows that some lives are rendered unworthy of public mourning by the very categories through which institutions see (Butler 32 to 39). Designing for recognition means designing for the possibility that such categories will be unsettled when voices long excluded begin to speak.
Third, they require that transparency be subordinated to intelligibility and that intelligibility be measured not by what experts can interpret but by whether those affected can use available concepts and forums to render their experiences communicable. Ott and Dabrock’s critique of infraethical transparency points in this direction, as does Fricker’s analysis of hermeneutical injustice. An interface that exposes how a model works, but that does not allow users to articulate new harms or to contest its categories, remains infraethical. Moral attention appears when those same users can bring novel experiences into view and oblige institutions to revise their practices in light of those experiences.
Fourth, they require disciplines of attention at both personal and institutional levels. Weil and Tronto remind us that attention is not mere curiosity. It is the willingness to allow another’s reality to recalibrate our own priorities and plans (Weil 105 to 110; Tronto 105 to 110). In bureaucratic contexts, this might mean refusing to let metric targets fully dictate what counts as success, and instead building into evaluation processes qualitative accounts from those on the receiving end of policies. In design contexts, it might mean structuring workshops and participatory processes so that those most affected by systems speak first and define the terms of discussion, rather than being invited to react to preformed proposals.
Finally, recognition and moral attention require that acknowledgment be built into the core logic of institutions, not relegated to the domain of personal virtue. Cavell’s distinction suggests that there must be public sites in which individuals and communities can confront those who wield power over them, present their claims, and receive more than a statistical reply (Cavell 340 to 345). This could take the form of standing citizens panels with authority over particular technologies, of ombuds offices empowered to investigate and remedy algorithmic harms, or of legal doctrines that recognize the right to a hearing whenever automated systems significantly affect one’s prospects. In each case, the point is the same. To be recognized is to be able to speak in a forum where one’s words can obligate others.
Within the architecture of this book, Chapter Fourteen marks a shift from diagnosing the failures of knowledge and justice to articulating the positive conditions under which human subjects can live as beings who are heard. The ethics of being heard is not an ornament added after more fundamental questions have been resolved. It is the form that those questions take once we accept that persons are embodied, vulnerable, and historically situated, and that they require more than abstract respect. They require structures of recognition and attention that can absorb their speech without neutralizing it. The next movement will draw these threads toward the idea of the public good, but its possibility depends on the work done here. A public can exist as a moral reality only where individuals and communities can speak into shared institutions and find that their words do more than echo.
Chapter Sixteen
Transparency and Opacity Revisited: Designing for Trust
When the speculative edifice of artificial intelligence finally collapsed, what remained was an infrastructure that still processed people while no longer convincing them that it deserved to. The crisis was not confined to economics or to epistemology. It was a crisis of trust in the deepest sense, a recognition that the very architectures through which people moved, spoke, worked, and received care had been designed to see through them while remaining obscure themselves. The earlier volume proposed a simple reversal as a design maxim, that infrastructures must become transparent while human beings require zones of opacity in which their interior lives, vulnerabilities, and histories are not turned into raw material. The intervening years have shown both the power and the limits of that slogan. Transparency turned out to be more complicated than disclosure, and opacity more demanding than secrecy. In this chapter I return to that pair and argue that trust after the bubble depends on an ethics of design in which institutional transparency is treated as a public obligation and human opacity as a basic entitlement, yet both are refracted through contextual norms, democratic governance, and the practice of practical wisdom rather than technical maximalism.
Tabea Ott and Peter Dabrock provide a precise vocabulary for this rethinking in their analysis of artificial intelligence in health care. They point out that the call for transparency has a Janus faced character. On the one hand, patients are asked to make themselves ever more visible to systems, donating data and accepting pervasive monitoring as the condition for receiving care. On the other hand, the systems that process those data, from proprietary algorithms to institutional decision pathways, often remain difficult or impossible to scrutinize (Ott and Dabrock, “Transparent Human” 1 to 3). They argue that transparency in such contexts is not itself an ethical principle but what they call an infraethical concept, a condition that may enable ethical evaluation but that does not yet decide what is just or unjust, respectful or degrading. Transparency can be mobilized in the service of dignity when it allows patients and citizens to understand and contest how they are governed, but it can just as easily serve domination when used as a pretext for stripping away privacy while leaving institutional power untouched (Ott and Dabrock, “Transparent Human” 4 to 6).
Taking their warning seriously requires that we refine the earlier maxim. To say that infrastructures must be transparent while people remain opaque cannot mean that all mechanisms and data flows must be equally available to everyone in every context, any more than it can mean that persons can never be asked to disclose what is needed for care, coordination, or mutual accountability. It must instead mean that transparency and opacity are distributed according to a moral grammar in which the direction of justification runs from institutions toward persons rather than the reverse, and in which requests for personal visibility are always answerable to publicly contestable reasons. The AI era largely inverted this relation. People were asked to explain themselves to platforms, insurers, and security regimes that did not reciprocate with intelligible explanations of their own structure. After the bubble, designing for trust requires reversing that priority. Systems must be legible in their purposes, capacities, and limits before they are granted authority, while persons must be allowed to disclose themselves only in ways that accord with meaningful consent and with the norms of the particular situations in which they live.
Helen Nissenbaum’s theory of contextual integrity clarifies why this reversal cannot be accomplished through general slogans alone. In her account, privacy is not the protection of a mysterious inner core against any information flow, but the protection of appropriate information flows within and across social contexts. She defines contextual integrity as a condition in which information gathering and dissemination conform to the norms of a given context, including who is entitled to know what, under what roles and relationships, and with what limitations on further transmission (Nissenbaum, “Privacy as Contextual Integrity” 119 to 122). Public surveillance and many data intensive practices violate privacy not simply because they gather much information, but because they rip information from one context and redeploy it in another without regard for the expectations and values that govern each setting (Nissenbaum, “Privacy as Contextual Integrity” 127 to 130).
From a contextual integrity perspective, the problem with making human beings transparent is not that any disclosure is always wrong. It is that indiscriminate visibility destroys the patterned differences among spheres of life that people rely on to shape relationships, pursue projects, and form identities. Nissenbaum writes that there are no arenas of life where anything goes in relation to information flows. Every context, whether intimate friendship, political association, or medical care, is governed by more or less articulate norms about what may be shared and with whom (Nissenbaum, “Privacy as Contextual Integrity” 119 to 120, 122 to 124). Technical architectures that treat all data as fungible undermine those norms and thereby corrode trust. To design for trust is therefore to design for contextual integrity, which means that infrastructural transparency must be tailored to specific practices. Systems must disclose how they handle information in ways that map onto existing or aspirational norms, rather than appealing to abstract notions of openness while continuing to disembed personal data from their contexts.
The history of the World Wide Web offers a suggestive counterexample of what a different orientation to transparency might look like. Tim Berners Lee insists that the web was from the beginning a social creation more than a technical one, designed to help people work together rather than as an autonomous technical toy (Berners Lee, Weaving the Web 123 to 126). He notes that he worked under the auspices of CERN, an institution committed to collaborative scientific inquiry, and that he insisted the core protocols and standards be shared freely so that the web could work for everyone rather than becoming a proprietary system (Berners Lee, Weaving the Web 52 to 55, 135 to 140). The transparency that mattered here was not primarily that every line of source code be available to every layperson, though the openness of standards was significant. It was that the rules governing interaction were public, interoperable, and subject to modification through collective processes rather than dictated by a single firm. As later consolidation and data extraction practices demonstrate, such openness is no guarantee against enclosure, but it remains an instructive example of transparency oriented toward shared empowerment rather than unilateral control.
Andrew Feenberg’s critical theory of technology helps us articulate what is at stake in these design decisions. He argues that modern technologies are not neutral tools waiting to be deployed for good or ill, but condensations of social choices about control, participation, and meaning (Feenberg, Transforming Technology 3 to 8). For Feenberg, technical codes embody particular political rationalities. Under what he calls a technocratic or instrumental model, efficiency and control are privileged, and design tends to exclude users from meaningful influence. Under a more democratic model, technologies can be redesigned through processes of what he terms democratic rationalization, in which users and affected publics intervene to reshape technical choices so that they better express shared values (Feenberg, Transforming Technology 49 to 55, 84 to 90).
In this light, the AI era’s preference for opaque architectures was not an accident of complexity. It was a political decision encoded in technical form. Proprietary models, secret datasets, and restricted interfaces served a rationality of centralized control and speculative profit. Designing for trust in the aftermath requires a different technical code, one that presupposes public scrutiny, participatory modification, and the possibility that those affected may contest and redirect the purposes of a system. Transparency then becomes not a marketing term but a design constraint. Code, data practices, and institutional arrangements must be fashioned so that they can bear the weight of democratic rationalization, which means that they must be understandable enough to be criticized and flexible enough to be changed.
At the same time, the tendency to treat transparency as a universal remedy for collective problems has its own hazards. Garrett Hardin’s famous essay on the tragedy of the commons argues that when individuals pursue their own advantage in a shared resource, the aggregate result can be depletion that harms everyone. He concludes that some commons problems do not admit of purely technical solutions and instead require what he calls an extension in morality, collective arrangements that restrain individual action for the sake of the whole (Hardin 1243 to 1248). Hardin’s pessimism fed a narrative in which centralized control or privatization appeared as the only alternatives to ruin.
Elinor Ostrom’s work on self governing commons offers a crucial corrective that is directly relevant for the design of digital infrastructures. In Governing the Commons she examines empirical cases in which resource users have crafted durable institutions that prevent overuse without relying on either Leviathan or complete privatization. She identifies a set of design principles for such institutions, including clearly defined boundaries, collective choice arrangements that allow most individuals affected by rules to participate in modifying them, monitoring that is accountable to users, graduated sanctions, and conflict resolution mechanisms that are accessible and low cost (Ostrom 88 to 102, 178 to 181). These principles show that transparency, in the sense of shared monitoring and open rule making, must be coupled with local control and participation if it is to support trust rather than surveillance. Applied to data and algorithmic infrastructures, Ostrom’s insights suggest that common resources such as public data sets, communication platforms, and even certain models can be governed as commons, but only if those most affected have real authority in setting and revising the rules.
E. F. Schumacher’s call for appropriate technology adds a further dimension, that of scale and human comprehensibility. In Small Is Beautiful he argues that the dominant model of economic and technical progress treats bigger, faster, and more centralized systems as inherently superior, with the result that technologies often outgrow the scale at which people can understand and creatively engage them (Schumacher 159 to 162). He recommends instead methods and equipment that are inexpensive, suitable for small scale application, and compatible with human beings’ need for meaningful work and creativity (Schumacher 159 to 161). When we transpose this to digital infrastructures, a similar tension appears. The AI era favored massively centralized models trained on planetary scale data, which few outside a narrow technical and corporate elite could understand or influence. After the bubble, an ethic of trust points toward architectures that can be grasped, questioned, and modified by communities at the scale at which they live, whether that means municipal data trusts, local cooperative platforms, or sector specific consortia. Opacity here is not only a matter of privacy. It is a symptom of mismatched scale, where systems have grown so complex and distant that laypersons are forced into radical dependency on experts. Transparency then means not complete exposure of every parameter, which would overwhelm, but right sizing systems so that their core workings and consequences can be brought within human reach.
Aristotle’s account of phronesis in Nicomachean Ethics offers an account of the intellectual virtue needed to navigate these design tensions. In Book VI he distinguishes practical wisdom from both theoretical wisdom and technical skill. Phronesis concerns action in its own right and aims at good action, whereas production aims at an external product (Aristotle, Nicomachean Ethics VI, 1140a to 1140b). Practical wisdom deliberates about what is good and expedient for oneself and for human life in general, taking into account particulars and circumstances that no general rule can fully capture (Aristotle, Nicomachean Ethics VI, 1140b 5 to 6). Applied to the governance of technology, phronesis requires that designers, regulators, and users resist the temptation to treat transparency and opacity as technical toggles to be optimized. Instead they must judge, in each context, how much institutional visibility is needed to secure accountability and participation, how much personal opacity is needed to protect dignity and contextual integrity, and how these choices affect the capabilities and common goods described in the previous chapter.
Practical wisdom also demands that we accept the existence of tradeoffs and limits in the name of trust. A health system that refuses to aggregate certain kinds of intimate data across contexts may sacrifice some predictive power, but this sacrifice can be a phronetic decision that prioritizes the preservation of trust and the avoidance of subtle coercion over maximal informational advantage. A platform that limits microtargeting in order to preserve a shared public sphere may forego some advertising revenue, but this may be necessary for the maintenance of a common world. Aristotle’s distinction between action that is good in itself and production aimed at external products thus marks an important boundary. Technological production, however efficient, cannot substitute for wise judgment about the forms of life it supports or undermines.
We are now in a position to state an ethics of transparency and opacity for the post bubble era. Institutional transparency must be understood as a layered requirement that includes, at a minimum, public intelligibility about purposes, sources of funding, lines of accountability, and the basic logics by which decisions are made. It should be implemented in ways that support contextual integrity, so that people can know how information about them will flow within specific practices rather than being asked to accept blanket surveillance in exchange for convenience. It should be designed so that affected communities can monitor, evaluate, and modify systems in line with Ostrom’s principles of commons governance, rather than being reduced to passive subjects of distant authorities (Ostrom 88 to 102, 178 to 181).
Human opacity, in turn, must be reconceived as a positive right to zones of non disclosure, non profiling, and non inference, grounded not only in privacy but in the requirements of dignity, contextual integrity, and political freedom. People must have spaces in which they can experiment with identity, association, and speech without being continuously scored or predicted. They must be able to refuse certain forms of data collection and to withdraw from certain systems without exclusion from essential goods. Such opacity is not a withdrawal from social life but a precondition for trustworthy participation, because it assures that presence is not constantly leveraged as material for unknown purposes.
Trust emerges where these two arcs meet. It is not the naive confidence that systems will always behave as promised, but the reasonable assurance that they are structured so that deception and abuse are difficult, detectable, and corrigible, and that when harm occurs, those affected can confront institutions in forums that take their claims seriously. That assurance depends on transparency of institutions and opacity of persons, yet both must be interpreted through the lenses of contextual integrity, democratic rationalization, commons governance, appropriate scale, and practical wisdom. When those lenses are absent, transparency becomes a tool of exposure and opacity a mask for domination. When they are present, transparency and opacity become complementary conditions for a society in which people can entrust parts of their lives to technological infrastructures without surrendering their capacity to think, speak, and act as subjects.
Within the architecture of this book, Chapter Sixteen marks the point at which the ethical and political analyses of earlier sections crystallize into design principles. The question is no longer only who counts as a knower or whose dignity is affirmed in law, but how we shape the infrastructural conditions under which knowing, acting, and trusting are possible at all. The chapters that follow will move further into the concrete design of attentive systems and commons based infrastructures. Their viability depends on the grammar articulated here, that institutions must learn to live in the open and to accept being seen, while human beings must be granted spaces in which they can remain partially unseen and still fully present.
Chapter Nineteen
Vision: Toward a Post AI Ethical Society
The ruins of the artificial intelligence era confront us with a double revelation. First, they reveal the extent to which societies became habituated to infrastructures that demanded constant visibility, relentless productivity, and unceasing disclosure while offering only opaque mechanisms of evaluation and extraction in return. Second, they reveal how thoroughly many institutions abandoned the slower, riskier work of ethical and political life in favor of predictive systems that promised management without judgment and optimization without responsibility. The collapse was therefore not a technological failure alone. It was a moral, epistemic, ecological, and existential unravelling. The chapters that preceded this one traced the genealogies of that unraveling. This chapter turns toward the possibilities that remain. It attempts to articulate a vision in which ethical life becomes thinkable again, not as nostalgia for an imagined past but as the deliberate construction of institutions capable of sustaining attention, plurality, care, and planetary obligation in the aftermath of speculative excess.
The first voice needed here belongs to Simone Weil, because she understood attention as the deepest form of ethical orientation. In Waiting for God she writes that attention constitutes the “rarest and purest form of generosity,” since it involves the suspension of the will and the patient reception of the other without appropriation or control (Weil 105). For Weil, attention is not simply focus or cognitive concentration. It is the capacity to hold open one’s mind so that another person or another truth can disclose itself without being prematurely reduced to categories of use or advantage. Under the AI regime, this capacity was systematically eroded by infrastructures designed to monetize distraction and fragment interior life. These infrastructures habituated individuals to oscillate between micro engagements that were constantly interrupted, while simultaneously rewarding modes of thinking that were fast, instrumental, and deferential to algorithmic mediation. A post AI ethical society must therefore reorganize its institutions so that attention is not treated as a private luxury but as a shared civic virtue. This requires the construction of public, educational, medical, and technological environments where withholding intrusion is treated as a sign of respect, where interpretive patience is cultivated, and where people are permitted to maintain interiority without the continuous pressure to reveal, perform, or optimize themselves.
The second structural element of this vision derives from Hannah Arendt’s concept of natality. In The Human Condition Arendt describes natality as the fact that every birth introduces a new beginning into the world and that human action gains its meaning from this capacity to initiate what was not predetermined (Arendt 177). Arendt therefore locates freedom not in the sovereignty of the isolated will but in the shared, unpredictable space where people speak and act before one another. The AI era displaced this shared unpredictability by substituting prediction for judgment and by instituting systems that treated human plurality as a form of noise to be filtered rather than as the source of political life itself. In a post AI society, natality must be institutionalized so that new voices can enter public space without being algorithmically profiled, commercially targeted, or preemptively categorized. This means protecting the opacity of individuals, restoring the conditions for deliberation that have been weakened by automated sorting, and designing infrastructures that allow people to gather, to speak, and to initiate without being reduced to behaviorally inferred probabilities.
The third component of this vision engages the racial and colonial structures that the AI era magnified rather than resolved. Frantz Fanon’s The Wretched of the Earth demonstrates that colonial domination operates not only through the expropriation of land but through the internalization of inferiority and the psychic partitioning of the colonized self (Fanon 36 to 40). Achille Mbembe’s account of necropolitics in turn reveals how modern sovereignty organizes itself by deciding which lives may be exposed to violence, abandonment, or premature death (Mbembe 66 to 69). These frameworks expose the profound continuity between older colonial regimes and the new infrastructures of algorithmic governance. Predictive policing targeted communities already damaged by structural racism. Automated border technologies intensified surveillance of migrants without addressing the geopolitical forces that compelled migration. Data extraction from countries in the Global South replicated historical extraction of minerals, labor, and land. A post AI ethical society cannot simply repair technical inequities. It must center decolonial transformation by returning authority, decision making, and epistemic legitimacy to the communities that bore the greatest burdens of both colonial and digital exploitation. This involves not only data sovereignty agreements and community control over infrastructures, but also sustained material redistribution and long term reparative transformation.
The ecological horizon of this vision is articulated most sharply by Donna Haraway. In Staying with the Trouble Haraway insists that the most pressing task of our time is not to transcend ecological limits through technological substitution, but to acknowledge our entanglement with multispecies worlds and to cultivate modes of living that recognize responsibility across extended temporal and ecological scales (Haraway 1 to 3; 12 to 15). The AI era encouraged fantasies of immaterial computation, conveniently ignoring the mineral, energy, and labor infrastructures required to sustain global platforms. A post AI ethical society must instead ground itself in ecological realism. Every new system must be evaluated not only for its social impact but for its planetary cost. The extraction of rare earth minerals, the carbon emissions of training large models, and the geopolitical violence that attends the global supply chain cannot be justified by appeals to innovation. Haraway’s call to remain with the trouble urges the creation of technologies that acknowledge ecological finitude, that respect non human lives, and that refuse the instrumentalization of the planet for speculative objectives.
To integrate these insights into practice requires a further conceptual anchor, supplied by Sheila Jasanoff’s theory of co production. In The Ethics of Invention Jasanoff argues that societies and technological systems shape one another through a continual process of mutual reinforcement, such that new inventions produce new norms, new vulnerabilities, and new obligations that institutions must learn to govern (Jasanoff 12 to 15; 220 to 23). This insight means that a post AI society cannot isolate technical design from democratic deliberation. Choices about what to build, how to regulate, and when to withdraw must be made through processes that include publics who are capable of evaluating the risks and benefits. This requires new educational structures that teach technical literacy, historical understanding, decolonial awareness, and ecological responsibility. It also requires new legal frameworks that institutionalize participatory governance and ensure meaningful oversight over technological systems that impact collective life.
When these threads are woven together, a post AI ethical society begins to take shape as a form of life organized around several shared commitments. First, attention must become a public ethic. Institutions must be designed so that they recognize individuals as centers of meaning rather than as sources of data. Second, plurality must be protected as the condition for political freedom. Arendt’s account of natality demands that institutions allow for the unexpected arrival of new voices and possibilities (Arendt 175 to 77). Third, decolonization must be structural rather than symbolic. Fanon and Mbembe show that ethical legitimacy depends on dismantling the hierarchies of disposability built into the old order (Fanon 36 to 40; Mbembe 66 to 69). Fourth, ecological finitude must be at the center of technological evaluation. Haraway’s insistence on multispecies obligation requires that designs be measured by their contribution to planetary survival rather than by commercial innovation. Fifth, knowledge must be democratized. Jasanoff’s analysis reveals that publics must be positioned as co stewards of technological futures rather than as passive users or experimental populations.
These commitments imply new institutional forms. Public infrastructures must be open to scrutiny and collectively governed. Platforms must treat data as a shared resource subject to participatory regulation. Schools must teach the capabilities required for democratic technopolitics, including critical reasoning, ecological literacy, and historical repair. Judicial systems must guarantee the right to contest automated judgments and the right to explanation. Cultural institutions must cultivate the sensibilities necessary for recognition, empathy, and world building.
Finally, Judith Butler provides an essential reminder that any ethical society must accept the vulnerability that attends human life. In Precarious Life Butler argues that the bonds that tie us to one another expose us to grief and injury and that societies often disavow this shared precarity by refusing to recognize certain lives as grievable (Butler 19 to 20; 32 to 34). A post AI society cannot seek refuge in automated governance or predictive systems that promise order without exposure. It must embrace the fact that uncertainty, dependence, and mutual susceptibility are the conditions of ethical life. Technologies must be designed not to shield society from these vulnerabilities but to support the collective capacities that make them bearable: attention, care, judgment, plurality, and mutual recognition.
This chapter, therefore, does not present a solution. It presents a stance: careful, deliberate, planetary, plural, historically aware, and oriented toward the slow labor of rebuilding institutions that can sustain dignity. The conclusion that follows will translate this stance into evaluative criteria that can guide public action, education, and design. The wager remains that even after a speculative era that consumed attention, exploited bodies, and destabilized worlds, the capacity to begin again persists, and it is this capacity that an ethical society must be organized to protect.
Conclusion
The Work of Reorientation**
To write in the aftermath of a technological collapse is to confront both the fragility and the resilience of the human condition. The collapse of the artificial intelligence era disclosed not a technical miscalculation but a civilization that had slowly trained itself to outsource the difficult labor of judgment, care, and accountability to infrastructures that could not sustain the weight placed upon them. The bubble did not burst only within financial markets. It burst within institutions that had uncritically treated efficiency as a proxy for wisdom, prediction as a substitute for understanding, extraction as a mode of governance, and attention as a resource available for continuous harvest. What now remains is not the question of how to rebuild what failed but the more demanding question of how to become the kinds of beings capable of building differently.
Throughout this book, a set of intellectual companions has helped clarify both the stakes and the obligations that follow from the collapse. Simone Weil teaches that attention constitutes the foundational moral act because it requires that one receive the presence of another without grasping or control, and that one hold the mind open long enough for truth to appear on its own terms (Weil 105). Hannah Arendt reminds us that the capacity to begin anew, which she names natality, is the source of political life and the possibility of shared worlds that do not merely extend the past but interrupt it with new meaning (Arendt 175 to 77). Frantz Fanon and Achille Mbembe insist that ethical renewal cannot be imagined without reckoning with the racial and colonial structures that have rendered certain populations disposable, and that any institutional future worthy of the name must dismantle systems that authorize abandonment, constraint, and premature death (Fanon 36 to 40; Mbembe 66 to 69). Donna Haraway situates these struggles within the ecological finitude of a planet that cannot endure endless extraction and that demands a practice of making kin across species, systems, and timescales (Haraway 12 to 15). Sheila Jasanoff’s account of co production then reveals that technical and political orders shape one another, and that societies must intentionally design governance structures capable of anticipating harms, cultivating democratic oversight, and aligning innovation with planetary and social obligations (Jasanoff 12 to 15; 220 to 23). Judith Butler concludes this constellation by insisting that the recognition of shared precarity is not a sign of weakness but a foundation for political responsibility, because every effort to disavow vulnerability becomes an effort to declare some lives ungrievable (Butler 19 to 20; 32 to 34).
Taken together, these thinkers reveal that the future after the bubble is neither a matter of rebuilding nor a matter of abandoning technology. It is a matter of reorientation. Such reorientation is not principally technical. It is ethical, political, ecological, and metaphysical. It concerns what societies choose to notice and what they ignore, what they protect and what they expose, what they treat as valuable and what they allow to become disposable. If the prior era trained institutions to regard human beings primarily as sources of data, then the task before us is to reorganize institutions so that they regard human beings as sources of meaning, agency, and irreducible interior life. If the prior era treated communities in the Global South as reservoirs of extractable data and labor, then the task before us is to construct structures of sovereignty, reparation, and self governance. If the prior era treated the planet as an infinite resource for computational expansion, then the task before us is to align technological development with ecological viability and multispecies flourishing.
A post AI ethical society therefore requires transformations at every scale. At the level of the person, it requires recovering the capacity to attend, to think slowly, and to resist the pressures toward constant legibility. At the level of institutions, it requires constructing organizations that can hear, respond, and revise, rather than conceal, extract, and control. At the level of governance, it requires processes that welcome contestation, plural perspectives, and shared public reasoning. At the planetary level, it requires acknowledging the limits of growth and innovating within the ecological boundaries that make life possible.
This book has not offered a doctrine or a program, because any such program would replicate the conceit of the AI era, which promised salvation through systematization and optimization. Instead, the book has offered a set of dispositions and commitments that must guide the ongoing work of constructing ethical life. Attention must be protected and cultivated. Plurality must be welcomed as a political resource. Decolonization must be treated as a structural imperative rather than an aspirational metaphor. Planetary entanglement must be understood as the horizon of technological and social imagination. Democratic deliberation must be institutionalized as the form through which shared futures are chosen and contested.
The work ahead therefore resembles what Arendt calls the task of world building, the fragile and ongoing effort to sustain institutions and practices that make human plurality both possible and livable (Arendt 7 to 9; 175). It resembles what Weil calls the labor of attention, the disciplined formation of the mind toward reception rather than domination (Weil 62 to 65). It resembles what Fanon calls the struggle for a humanity that has not yet been realized, which can only emerge when the lives of the formerly colonized and racialized are no longer treated as expendable (Fanon 36 to 40). It resembles what Haraway calls staying with the trouble, the refusal to seek escapist futures that bypass ecological responsibility (Haraway 1 to 3). It resembles what Butler calls the recognition of shared precarity, the acceptance that ethical life is lived within mutual dependence and vulnerability (Butler 19 to 20).
The conclusion is therefore not a closing gesture but an opening one. It affirms that collapse reveals possibilities that were previously obscured, that crises can disclose the failures of prior orders, and that new forms of ethical life can be constructed if societies are willing to recognize their own entanglements, inheritances, and responsibilities. The question now is whether institutions, designers, policymakers, and communities will take up this work with the seriousness it demands. The wager of this book is that they can, and that the future remains open enough for the work of ethical reorientation to take root.
What remains is the slow, unglamorous labor of building worlds that can sustain attention, protect interiority, distribute responsibility, honor plurality, repair historical violence, and safeguard the planet that makes all life possible. Only within such worlds can intelligence, human or artificial, be judged meaningful. Only within such worlds can justice be imagined. Only within such worlds can the future become something other than a repetition of the past.
The final responsibility rests not with machines but with us.
Works Cited
Abebe, Rediet, et al. “Narratives and Counternarratives on Data Sharing in Africa.” arXiv, 2021, arxiv.org/abs/2103.01168.
“African Declaration on Internet Rights and Freedoms.” African Declaration, africaninternetrights.org.
Ahmed, Sara. On Being Included: Racism and Diversity in Institutional Life. Duke University Press, 2012.
Anderson, Elizabeth S. “What Is the Point of Equality.” Ethics, vol. 109, no. 2, 1999, pp. 287 to 337, University of Chicago Press, https://doi.org/10.1086/233897.
Arendt, Hannah. The Human Condition. Second edition, University of Chicago Press, 1998.
Arendt, Hannah. Crises of the Republic: Lying in Politics, Civil Disobedience, On Violence, Thoughts on Politics and Revolution. Houghton Mifflin Harcourt, 1972.
Aristotle. Nicomachean Ethics. Translated by Terence Irwin, second edition, Hackett, 1999.
Barlevy, Gadi. “Bubbles and Fools.” Economic Perspectives, vol. 39, no. 2, 2015, pp. 54 to 77, Federal Reserve Bank of Chicago.
Benjamin, Ruha. Race After Technology: Abolitionist Tools for the New Jim Code. Polity Press, 2019.
Benhabib, Seyla. Situating the Self: Gender, Community, and Postmodernism in Contemporary Ethics. Polity Press, 1992, DOI: 10.4324/9781003059516.
Berners Lee, Tim. Weaving the Web: The Original Design and Ultimate Destiny of the World Wide Web. HarperCollins, 1999.
Buchanan, Mark. “The Laws of Inflating the AI Bubble.” Nature Physics, vol. 20, no. 10, 2024, p. 1362, doi:10.1038/s41567 024 02627 5.
Butler, Judith. Precarious Life: The Powers of Mourning and Violence. Verso, 2004.
Cadwalladr, Carole, and Emma Graham Harrison. “Revealed: 50 Million Facebook Profiles Harvested for Cambridge Analytica in Major Data Breach.” The Guardian, 17 Mar. 2018, http://www.theguardian.com/news/2018/mar/17/cambridge-analytica-facebook influence us election.
Carel, Havi. Phenomenology of Illness. Oxford University Press, 2016.
Carroll, Stephanie Russo, et al. “The CARE Principles for Indigenous Data Governance.” Data Science Journal, vol. 19, 2020, pp. 1 to 12, doi:10.5334/dsj 2020 043.
Cavell, Stanley. The Claim of Reason: Wittgenstein, Skepticism, Morality, and Tragedy. Oxford University Press, 1979.
Coleman, Daniel. “Digital Colonialism: The Twenty First Century Scramble for Africa.” Michigan Journal of Race and Law, vol. 24, no. 2, 2019, pp. 349 to 398.
Collins, Patricia Hill. Black Feminist Thought: Knowledge, Consciousness, and the Politics of Empowerment. Second edition, Routledge, 2000, DOI: 10.4324/9780203907283.
Costanza Chock, Sasha. Design Justice: Community Led Practices to Build the Worlds We Need. MIT Press, 2020.
Couldry, Nick, and Ulises A. Mejias. “Data Colonialism: Rethinking Big Data’s Relation to the Contemporary Subject.” Television and New Media, vol. 20, no. 4, 2019, pp. 336 to 349, doi:10.1177/1527476418796632.
Couldry, Nick, and Ulises A. Mejias. The Costs of Connection: How Data Is Colonizing Human Life and Appropriating It for Capitalism. Stanford University Press, 2019.
“Democracy Shrinking, Violence Growing: Internet Shutdowns in 2023.” Access Now and the KeepItOn Coalition, 2024, accessnow.org/keepiton.
Donovan, Kevin P., and Emma Park. “Algorithmic Intimacy: The Data Economy of Predatory Inclusion in Kenya.” Social Anthropology, vol. 30, no. 2, 2022, pp. 120 to 139, doi:10.3167/saas.2022.300208.
Fanon, Frantz. Black Skin, White Masks. Translated by Richard Philcox, Grove Press, 2008.
Fanon, Frantz. The Wretched of the Earth. Translated by Richard Philcox, Grove Press, 2004.
Feenberg, Andrew. Transforming Technology: A Critical Theory Revisited. Oxford University Press, 2002.
Floridi, Luciano. “Why the AI Hype Is Another Tech Bubble.” Philosophy and Technology, vol. 37, no. 4, 2024, article 128, doi:10.1007/s13347 024 00817 w.
Foucault, Michel. Discipline and Punish: The Birth of the Prison. Translated by Alan Sheridan, Pantheon Books, 1977.
Foucault, Michel. The History of Sexuality. Volume 1: An Introduction. Translated by Robert Hurley, Vintage Books, 1990.
Foucault, Michel. Security, Territory, Population: Lectures at the College de France 1977 to 1978. Edited by Michel Senellart, translated by Graham Burchell, Palgrave Macmillan, 2007.
Foucault, Michel. The Birth of Biopolitics: Lectures at the College de France 1978 to 1979. Edited by Michel Senellart, translated by Graham Burchell, Palgrave Macmillan, 2008.
Fraser, Nancy. “Social Justice in the Age of Identity Politics: Redistribution, Recognition, and Participation.” The Tanner Lectures on Human Values, vol. 19, edited by Grethe B. Petersen, Utah State University Press, 1998, pp. 1 to 67.
Fricker, Miranda. Epistemic Injustice: Power and the Ethics of Knowing. Oxford University Press, 2007, DOI: 10.1093/acprof:oso/9780198237907.001.0001.
Graeber, David. The Utopia of Rules: On Technology, Stupidity, and the Secret Joys of Bureaucracy. Melville House, 2015.
Haggerty, Kevin D., and Richard V. Ericson. “The Surveillant Assemblage.” British Journal of Sociology, vol. 51, no. 4, 2000, pp. 605 to 622.
Hall, Stuart. “Encoding and Decoding.” Culture, Media, Language: Working Papers in Cultural Studies 1972 to 1979, edited by Stuart Hall, Dorothy Hobson, Andrew Lowe, and Paul Willis, Hutchinson, 1980, pp. 128 to 138.
Hardin, Garrett. “The Tragedy of the Commons.” Science, vol. 162, no. 3859, 1968, pp. 1243 to 1248.
Harding, Sandra. Whose Science? Whose Knowledge? Thinking from Womens Lives. Cornell University Press, 1991.
Haskel, Jonathan, and Stian Westlake. Capitalism without Capital: The Rise of the Intangible Economy. Princeton University Press, 2018.
hooks, bell. Feminist Theory: From Margin to Center. Routledge, 2015.
Honneth, Axel. The Struggle for Recognition: The Moral Grammar of Social Conflicts. Translated by Joel Anderson, Polity Press, 1995.
Jasanoff, Sheila. Science and Public Reason. Routledge, 2012, DOI: 10.4324/9780203113820.
Jasanoff, Sheila. The Ethics of Invention: Technology and the Human Future. W W Norton, 2016.
Jasanoff, Sheila, editor. States of Knowledge: The Co Production of Science and the Social Order. Routledge, 2004, DOI: 10.4324/9780203413845.
Kindleberger, Charles P., and Robert Z. Aliber. Manias, Panics, and Crashes: A History of Financial Crises. Seventh edition, Palgrave Macmillan, 2015, doi:10.1007/978 1 137 52574 1.
Kittay, Eva Feder. Love’s Labor: Essays on Women, Equality, and Dependency. Routledge, 1999.
Korsgaard, Christine M. The Sources of Normativity. Cambridge University Press, 1996, https://doi.org/10.1017/CBO9780511554476.
Kramer, Adam D. I., Jamie E. Guillory, and Jeffrey T. Hancock. “Experimental Evidence of Massive Scale Emotional Contagion through Social Networks.” Proceedings of the National Academy of Sciences, vol. 111, no. 24, 2014, pp. 8788 to 8790, doi:10.1073/pnas.1320040111.
Kukutai, Tahu, and John Taylor, editors. Indigenous Data Sovereignty: Toward an Agenda. ANU Press, 2016, doi:10.22459/CAEPR38.11.2016.
Latour, Bruno. “Why Has Critique Run out of Steam? From Matters of Fact to Matters of Concern.” Critical Inquiry, vol. 30, no. 2, 2004, pp. 225 to 248, doi:10.1086/421123.
Lippmann, Walter. Public Opinion. Harcourt, Brace and Company, 1922.
MacIntyre, Alasdair. After Virtue: A Study in Moral Theory. Third edition, University of Notre Dame Press, 2007.
Margalit, Avishai. The Decent Society. Translated by Naomi Goldblum, Harvard University Press, 1996.
Marx, Karl. Grundrisse: Foundations of the Critique of Political Economy (Rough Draft). Translated by Martin Nicolaus, Penguin Books in association with New Left Review, 1973.
Mansbridge, Jane. Beyond Adversary Democracy. Basic Books, 1980.
Mbembe, Achille. “Necropolitics.” Public Culture, vol. 15, no. 1, 2003, pp. 11 to 40, DOI: 10.1215/08992363 15 1 11.
Mbembe, Achille. Necropolitics. Translated by Steven Corcoran, Duke University Press, 2019, doi:10.1515/9781478007227.
Mignolo, Walter D. Local Histories Global Designs: Coloniality, Subaltern Knowledges, and Border Thinking. Princeton University Press, 2000.
Mills, C. Wright. The Power Elite. Oxford University Press, 1956.
Mills, Charles W. The Racial Contract. Cornell University Press, 1997.
Morozov, Evgeny. To Save Everything, Click Here: The Folly of Technological Solutionism. PublicAffairs, 2013.
Mouffe, Chantal. The Democratic Paradox. Verso, 2000.
Noble, Safiya Umoja. Algorithms of Oppression: How Search Engines Reinforce Racism. New York University Press, 2018.
Noddings, Nel. Caring: A Feminine Approach to Ethics and Moral Education. University of California Press, 1984.
Nothias, Toussaint. “Access Granted: Facebook’s Free Basics in Africa.” Media, Culture & Society, vol. 42, no. 3, 2020, pp. 329 to 348, doi:10.1177/0163443719890530.
Nussbaum, Martha C. Creating Capabilities: The Human Development Approach. Belknap Press of Harvard University Press, 2011, https://doi.org/10.4159/harvard.9780674061200.
Nussbaum, Martha C. Frontiers of Justice: Disability, Nationality, Species Membership. Belknap Press of Harvard University Press, 2006, https://doi.org/10.2307/j.ctv1c7zftw.
Nussbaum, Martha C. Not for Profit: Why Democracy Needs the Humanities. Princeton University Press, 2010, DOI: 10.2307/jj.15766989.
Nissenbaum, Helen. “Privacy as Contextual Integrity.” Washington Law Review, vol. 79, no. 1, 2004, pp. 119 to 157.
Nissenbaum, Helen. Privacy in Context: Technology, Policy, and the Integrity of Social Life. Stanford Law Books, 2010.
O Neil, Cathy. Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. Crown, 2016.
O Neill, Onora. Autonomy and Trust in Bioethics. Cambridge University Press, 2002.
Ostrom, Elinor. Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge University Press, 1990.
Ott, Tabea, and Peter Dabrock. “Transparent Human and Nontransparent Technology. The Janus Faced Call for Transparency in AI Based Health Care Technologies.” Frontiers in Genetics, vol. 13, 2022, article 902960, doi:10.3389/fgene.2022.902960.
Power, Michael. The Audit Society: Rituals of Verification. Oxford University Press, 1997.
Proctor, Robert N. “Agnotology: A Missing Term to Describe the Cultural Production of Ignorance and Its Study.” Agnotology: The Making and Unmaking of Ignorance, edited by Robert N. Proctor and Londa Schiebinger, Stanford University Press, 2008, pp. 1 to 33.
Proctor, Robert N., and Londa Schiebinger, editors. Agnotology: The Making and Unmaking of Ignorance. Stanford University Press, 2008.
Putnam, Hilary. The Collapse of the Fact Value Dichotomy and Other Essays. Harvard University Press, 2002.
Quijano, Aníbal. “Coloniality and Modernity Rationality.” Cultural Studies, vol. 21, no. 2 to 3, 2007, pp. 168 to 178, doi:10.1080/09502380601164353.
Rawls, John. A Theory of Justice. Revised edition, Belknap Press of Harvard University Press, 1999.
Sandel, Michael J. What Money Cannot Buy: The Moral Limits of Markets. Farrar, Straus and Giroux, 2012.
Schumacher, E. F. Small Is Beautiful: Economics as If People Mattered. Harper Perennial, 2010.
Sen, Amartya. The Idea of Justice. Belknap Press of Harvard University Press, 2009, https://doi.org/10.4159/9780674054578.
Shiller, Robert J. Irrational Exuberance. Princeton University Press, 2000.
Suarez Villa, Luis. Technocapitalism: A Critical Perspective on Technological Innovation and Corporatism. Temple University Press, 2009.
Taylor, Charles. Sources of the Self: The Making of the Modern Identity. Harvard University Press, 1989.
Thrift, Nigel. Knowing Capitalism. SAGE Publications, 2005.
UNESCO. Balancing Freedom of Expression and Addressing Disinformation in the Digital Age. UNESCO, 2020.
United Nations. Guiding Principles on Business and Human Rights: Implementing the United Nations Protect, Respect and Remedy Framework. United Nations Office of the High Commissioner for Human Rights, 2011.
Vosoughi, Soroush, Deb Roy, and Sinan Aral. “The Spread of True and False News Online.” Science, vol. 359, no. 6380, 2018, pp. 1146 to 1151, doi:10.1126/science.aap9559.
Walker, Margaret Urban. Moral Understandings: A Feminist Study in Ethics. Routledge, 1998.
West, Cornel. Race Matters. Beacon Press, 1993.
Weil, Simone. Correspondance. Editions l Age d Homme, 1982.
Weil, Simone. Waiting for God. Translated by Emma Craufurd, Harper and Row, 1951.
Weil, Simone. “Reflections on the Right Use of School Studies with a View to the Love of God.” Waiting for God, translated by Emma Craufurd, Harper and Row, 1951, pp. 105 to 116.
Young, Iris Marion. Inclusion and Democracy. Oxford University Press, 2000, DOI: 10.1093/0198297556.001.0001.
Zuboff, Shoshana. The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. PublicAffairs, 2019.
Jonathan Johnson-Swagel 
Leave a comment