The Architecture of Manipulation: Dark Patterns, Game Theory, and the Theft of Cooperation

Image detail from ‘Child of War‘ (2023), Oil and Red Chalk on Linen, 40 x 30 cm by cornelia es said

This is the second of three essays in the series ‘Resistance as Material’ tracing a line from the handmade object to the blockchain protocol. Part 1 argued that digital materials — code, algorithms, blockchain protocols — possess agency: they push back, resist, and shape outcomes as much as the artist’s hand does. Part 2 asks: what happens when that material agency is weaponized? When the systems we interact with daily are designed not to collaborate with us but to extract from us — our attention, our data, our capacity to make informed decisions? Part 3 proposes a counter-architecture.

In the previous essay, Mi’kmaq artist Ursula Johnson described how the wood manipulates the maker into understanding what it can do.¹ The maker does not impose form on passive matter. The material pushes back, and from that friction, the work emerges.

Now consider the inverse. What if a material were designed to manipulate the maker — or the user — into not understanding what it does? What if the resistance were stripped out, the friction removed, and the surface made so smooth that the person interacting with it could no longer perceive the forces shaping their behavior?

That is the architecture of digital manipulation. And it is not a glitch: it is a business model.

I. Dark Patterns Are Part Of The Architecture

The term “dark pattern” was coined by UX designer Harry Brignull in 2010 to describe interface designs that trick users into unintended actions: hidden unsubscribe buttons, pre-checked consent boxes, countdown timers that manufacture urgency.² The term is useful. It names a real phenomenon. But it frames the problem as a collection of discrete tricks — bad decisions made by bad designers — when the reality is structural.

A more powerful framework comes from adapting Christopher Alexander’s pattern language, originally developed for architecture. Alexander argued that buildings are not isolated objects but emerge from the interplay between town, building, and construction — three scales that shape each other.³ Colin M. Gray extends this insight to digital systems in his CHI ’26 essay, introducing the Dark Patterns Knowledge Stack: a four-layered model that traces how high-level business imperatives produce interface-level harms.⁴

The layers work as follows:

• The Socio-Technical Landscape — Alexander’s “town” — is the high-level organizing concept: e-commerce, advertising, surveillance. This is the ecosystem in which everything else operates.
• The Goal or Intention layer — the “building” — is where the structural tension lives. The user opens Instagram to connect with friends. The shareholder needs the user to stay on-screen long enough to see more ads. These goals are in direct, systemic conflict. The entire architecture exists to manage this conflict — in the shareholder’s favor.
• The Interface layer — the “construction” — is what the user actually sees and touches: buttons, flows, scroll behaviors, notification badges. This is where dark patterns live. But they do not originate here. They are manifestations of the tension one level up.
• The User Limitations layer — the “inhabitant” — describes what the user brings to the encounter: cognitive biases (default bias, loss aversion, variable reward sensitivity), social vulnerabilities (identity performance, comparison), and limited time and attention. Dark patterns exploit these limitations with surgical precision.

The Knowledge Stack reframes the conversation. The hidden unsubscribe button is not a design failure. It is a design success — if the metric is shareholder value.

II. There Is No Innocent Code in a False System

This raises a question that Part 1 left open: if digital materials have agency, as Karen Barad’s agential realism suggests, where does the agency reside? In the feature? In the business model? In the programmer? In the user’s cognitive architecture?

The answer, in Barad’s framework, is: in none of these alone, and in all of them together. Agency is not a property of an isolated component. It emerges through intra-action — through the entanglement of all elements in the apparatus.⁵ The infinite scroll, taken in isolation, is a technical solution to a display problem: instead of paginated content, information flows continuously. As a standalone feature, it carries no more moral weight than a wheel.

But the infinite scroll does not exist in isolation. It exists inside an attention-economy business model that measures success in engagement time, a variable-reward structure borrowed from slot machine psychology, A/B tests that optimize for session duration, and a venture-capital logic that treats growth as an existential imperative. The feature was not invented in a vacuum and then misused. It was developed within this apparatus, shaped by its incentives, tested against its metrics, deployed in service of its goals.

Theodor Adorno wrote in 1951: “There is no right living in the wrong one.”⁶ The sentence applies to code as much as to life. If the entire system is configured for extraction — of attention, of data, of behavioral surplus — then no feature developed within that system is fully innocent. The incentive structure is present in the development environment: in the KPIs handed to programmers, in the A/B tests that define “success” as time-on-screen, in the quarterly earnings calls where engagement metrics are reported to investors. The code carries the handwriting of its context of origin.

At the same time — and this is where the differentiation matters — an infinite scroll is not the same thing as a facial recognition algorithm deployed for predictive policing. There are gradations. The question is not “guilty or innocent” but: how deep does the contamination run, and at what points is repurposing possible?

Three levels help to map this spectrum:

Level 1: The feature as technical possibility. A scrolling mechanism. A recommendation algorithm. A data storage protocol. Potential, unformed, not yet entangled with a specific apparatus. At this level, the technology is genuinely open to multiple uses.

Level 2: The feature inside the apparatus. The same scrolling mechanism, embedded in an attention-economy platform, paired with variable rewards, tested for addictiveness, deployed to maximize ad revenue. The development context is now inscribed in the code. Adorno’s false life penetrates the feature. Most of what we call “dark patterns” operate at this level: cookie consent flows designed to exhaust resistance, subscription cancellation labyrinths, notification badges calibrated to trigger anxiety.⁷

Level 3: The feature as instrument of power over life and death. The same tracking infrastructure that manipulates cookie consent is deployed for biometric surveillance, facial recognition, algorithmic border control, predictive policing. The line between commercial manipulation and state violence dissolves. This is Achille Mbembe’s terrain, and we will arrive there shortly.

The transitions between these levels are fluid. And that fluidity is itself the most dangerous design pattern of all. Because Level 1 looks harmless, the shift to Level 2 is normalized. Because Level 2 is perceived as “merely annoying” — cookie banners, dark patterns, consent fatigue — the shift to Level 3 becomes invisible. The normalization chain — from helpful feature through manipulative interface to necropolitical instrument — is a dark pattern that contains all other dark patterns.

Image detail from ‘Evolution I‘ (2025), Oil on Canvas, 70 x 50 cm by cornelia es said

III. Google: A Case Study in Structural Manipulation

The Knowledge Stack becomes concrete when applied to specific systems. Google’s location tracking is an instructive case, because it illustrates how all four layers interact.⁸

At the Landscape level, Google operates within the advertising economy. Its revenue depends on the precision of ad targeting, which depends on the granularity of user data, which depends on continuous surveillance.

At the Goal level, the tension is stark: the user believes she is using a smartphone for navigation, communication, and utility. Google’s objective is to capture location data granular enough to enrich advertising profiles. These goals are structurally opposed — and the system depends on this opposition remaining invisible. If users fully understood that their location data was being harvested for advertising profiles, many would opt out, and the data supply would shrink. The business model requires the gap between what the user believes is happening (utility) and what is actually happening (extraction). Dark patterns exist to maintain that gap.

At the Interface level, this tension manifests as “harmful defaults”: Web & App Activity is opted on by default. Privacy settings are buried behind progressive disclosure — multiple taps, nested menus, language designed to reassure rather than inform. The interface is confusing by design.

At the User Limitations level, default bias does the rest. Most people do not change default settings — not because they consent to tracking, but because changing defaults requires effort, knowledge, and time that the interface is designed to tax. And even when a user does opt out, proprietary technologies like IPGeo allow for continued location inference, subverting the user’s expressed preference through technical means.

The social media variant follows the same structure with different stakes. Teenage users seek social connection. Platform shareholders seek to maximize engagement time — which is the commodity sold to advertisers. The interface translates this tension into infinite scrolls, variable reward schedules, and forced actions (such as 30-day cooling-off periods before account deletion is finalized). These designs prey on perceptual limitations — the dopamine response to variable rewards, social vulnerabilities, the adolescent need for peer validation — to produce engagement metrics that translate directly into advertising revenue.⁹

The Knowledge Stack makes one thing visible that individual dark-pattern analysis obscures: the interface tricks are symptoms. The cause is the structural misalignment between user interests and shareholder interests — a misalignment that is the operating principle of surveillance capitalism.¹⁰

IV. Safiya Umoja Noble: The Algorithm Serves a Master

The Knowledge Stack reveals how manipulation works. But it has a blind spot: it frames the conflict as User versus Shareholder, as though all users were equally affected. They are not.

Safiya Umoja Noble, professor of Gender Studies and African American Studies at UCLA and recipient of a MacArthur Fellowship in 2021, demonstrated in Algorithms of Oppression (2018) that search algorithms do not merely manipulate — they discriminate. When Noble searched Google for “Black girls,” the first page returned pornography. When she searched for “white girls,” the results were radically different. The algorithm did not malfunction. It performed exactly as designed: reflecting the data it was trained on, the advertising incentives it was optimized for, and the cultural hierarchies embedded in both.¹¹

Although this bias might now seem less obvious (depending also on the user’s search history and personalized profile), Noble’s central argument still holds: search engines do not operate as neutral information brokers. Search algorithms structure discourse, shape representation, and reproduce social relations. They privilege whiteness, sexualize Black women, and render entire communities visible through the lens of stereotype — as a direct consequence of the business model that funds them. Private interests pay to promote certain sites. Monopoly status eliminates alternatives. The resulting algorithmic hierarchy is a (re)construction of reality — one that serves specific economic and racial interests while claiming objectivity.¹²

Noble’s work connects directly to Donna Haraway’s critique of the “god trick,” discussed in Part 1 of this series. Google performs exactly the operation Haraway described: it claims to see everything from nowhere — a neutral index of all human knowledge — while in fact occupying a very specific position: a for-profit corporation headquartered in Mountain View, California, funded by advertising revenue, built by a workforce whose technical and leadership ranks remain predominantly white or Asian and male — as of 2024, nearly 73% of Google’s U.S. tech employees were men, and Black and Latino employees together accounted for just over 13% of the U.S. workforce¹³ — and optimized for engagement metrics that reward sensationalism over substance. The algorithm is not objective. It is situated — in Haraway’s precise sense of the word — and its situatedness is rendered invisible by the interface’s claim to neutrality.¹⁴

Triptychon ‘Supergedenkjahr‘ (2014), Acrylic on Canvas, 100 x 220 cm by cornelia es said

V. Ruha Benjamin: The New Jim Code

If Noble exposes the discriminatory structure of search engines specifically, Ruha Benjamin, professor of African American Studies at Princeton, extends the analysis to technology as a system. In Race After Technology (2019), she introduces the concept of the “New Jim Code”: a regime in which technology hides, accelerates, and deepens discrimination while appearing neutral and even benevolent compared to the overt racism of previous eras.¹⁵

Benjamin identifies four mechanisms through which discriminatory design operates:

Engineered Inequity: Technologies that explicitly amplify racial hierarchies. The Beauty AI contest — an algorithm trained to judge beauty using a dataset dominated by white faces — consistently declared white women winners. The bias was in the training data, which encoded existing hierarchies as objective criteria.¹⁶

Default Discrimination: Technologies that ignore social context and thereby reproduce the default settings of race, class, and gender. A recidivism prediction algorithm used across U.S. states was found to wrongly flag Black defendants as future criminals at significantly higher rates than white defendants — not because the algorithm was explicitly racist, but because it was trained on historical policing data that already reflected decades of racial profiling. The numbers are stark: Black defendants who did not re-offend were misclassified as high risk at a rate of 44.9%, compared to 23.5% for white defendants. Conversely, white defendants who did re-offend were misclassified as low risk at 47.7%, compared to 28.0% for Black defendants.¹⁷ The algorithm did not need to be programmed with racial animus. It simply inherited the animus already encoded in the data.

Coded Exposure: The paradox of being simultaneously invisible to services and hypervisible to surveillance. Communities of color are underserved by algorithms that allocate resources (healthcare, credit, housing) while being over-targeted by algorithms that allocate suspicion (policing, border control, fraud detection).

Techno-Benevolence: Technologies that claim to address bias but deepen it. A healthcare algorithm used by major U.S. hospitals to allocate treatment resources systematically favored white patients over Black patients — not by using race as an input, but by using healthcare spending as a proxy for health needs. Because Black patients historically had less access to healthcare and therefore lower spending records, the algorithm interpreted their deprivation as lesser need. Correcting this single bias would have increased the percentage of Black patients receiving additional clinical attention from 17.7% to 46.5% — a gap that reveals how deeply structural racism is embedded in the data that algorithms treat as ground truth.¹⁸

Benjamin’s taxonomy maps onto the Knowledge Stack with precision. Where the Knowledge Stack asks “how does the apparatus produce manipulation?”, Benjamin asks “whom does the apparatus target, and whose interests does it serve?” The two frameworks are complementary lenses on the same structure. Together, they reveal that dark patterns are not equal-opportunity manipulations. They hit hardest where existing inequalities are deepest.

Benjamin makes one further move that resonates beyond the American context: she argues that race itself functions as a technology — a system designed to classify, stratify, and manage populations.¹⁹ If race is a technology, then the New Jim Code is not a misuse of technology. It is technology operating exactly as one of its oldest functions demands: the organization of human bodies into hierarchies of value.

VI. Achille Mbembe: When the Algorithm Decides Who Lives

Noble and Benjamin analyze discrimination within systems that nominally serve commercial purposes — search engines, healthcare allocation, criminal justice prediction. Cameroonian philosopher Achille Mbembe takes the analysis to its terminal point: what happens when these same systems serve the administration of death?

Mbembe’s concept of necropolitics, first articulated in 2003 and expanded into a book in 2016, extends Michel Foucault’s biopolitics — the governance of populations through the management of life — into its inverse: sovereignty exercised through the power to decide who lives and who must die.²⁰ In Mbembe’s analysis, the colony and the plantation are the paradigmatic sites of necropolitical power: spaces where human beings were reduced to objects of extraction, their lives rendered contingent and disposable. This is not historical. The afterlife of these structures persists in the present.

In the fourth chapter of Necropolitics, Mbembe introduces the digitalization of this sovereign power. Physical borders are digitized through technological surveillance and governance tools: digital databases, drones, sensors, biometric identification, facial recognition, location tracking. The mobility of populations — who may move, who is confined, who is expelled — falls under the control of systems that combine commercial tracking infrastructure with state power.²¹

This is where the three levels of feature contamination converge. The same tracking technology that powers Google’s ad targeting (Level 2) also powers the biometric border systems that Mbembe describes (Level 3). The same facial recognition algorithms that classify consumers for marketing purposes also classify populations for surveillance, profiling, and — in the most extreme cases — targeting. The same data infrastructure that enables dark patterns in cookie consent flows enables what Mbembe calls the creation of “death worlds”: conditions in which entire populations are subjected to an existence so precarious that it amounts to living death.²²

The bridge between these levels is not metaphorical. It is a legal construct: the data broker loophole. In the United States, the Fourth Amendment requires a warrant for state surveillance — but commercial data is exempt. State and federal agencies routinely purchase the same location data, device identifiers, and behavioral profiles that were harvested for advertising, bypassing constitutional protection by buying what they would otherwise need a court order to obtain. The Federal Trade Commission’s 2024 settlement with data broker X-Mode Social/Outlogic documented precisely this pipeline: location data harvested from ordinary apps, sold on to the Department of Defense, the Department of Homeland Security, Immigration and Customs Enforcement, and the Centers for Disease Control. Venntel and Babel Street — the named brokers in successive federal investigations — operate on the same model.²³ The proposed Fourth Amendment Is Not For Sale Act, which would close this loophole, passed the U.S. House in 2024 as part of the FISA Section 702 reauthorization package but remains stalled in the Senate. In the European Union, the Law Enforcement Directive creates a parallel structure that permits state access to commercially held data under conditions that civil society organizations continue to contest. The separation between commercial surveillance and state violence is not a fact of infrastructure. It is a fiction maintained by legal omissions.

Mbembe names racism as the primary mechanism through which necropolitics operates — both “hydraulic racism” (institutional, systemic, legal) and “nanoracism” (the daily, micro-level humiliations that accumulate into dehumanization).²⁴ Digital technologies amplify both: institutional racism is encoded in algorithms (as Noble and Benjamin document), while nanoracism is intensified through platforms that reward outrage, amplify stereotypes, and fragment solidarity.

The critical insight for our argument is this: the technologies of surveillance capitalism and the technologies of necropolitics are not separate systems. They share infrastructure, methodologies, and — increasingly — personnel. The data scientist who optimizes ad targeting and the contractor who builds predictive policing systems use the same tools, the same data pipelines, and often the same training data. The normalization chain from Level 1 to Level 3 is not a slippery slope argument. It is a description of existing supply chains.

And those supply chains are not only digital. The extraction that sustains the attention economy is not limited to data, attention, and behavioral surplus. It extends into the earth and into human bodies. The smartphone on which every dark pattern, every infinite scroll, every cookie consent flow operates is a physical object assembled from materials extracted under conditions that Mbembe’s framework describes with precision: cobalt mined by hand in the Democratic Republic of Congo — often by children, often under armed supervision — lithium extracted from salt flats in Bolivia and Chile at enormous ecological cost, rare earth elements refined in facilities with minimal labor protections, and electronic waste disassembled without safety equipment in Agbogbloshie, Ghana, and similar sites across the Global South.²⁵ The colony and the plantation, which Mbembe identifies as the paradigmatic sites of necropolitical power, have not disappeared. They have been reorganized as supply chains.

This is not a digression from the digital argument. It is the digital argument. The smooth surface of the interface — the frictionless swipe, the seamless scroll — is produced by friction elsewhere: in mines, in factories, in bodies. The “frictionless” experience that dark patterns optimize is only frictionless for the user in the Global North. For the worker in the mine, friction is the operating condition. Yuk Hui’s question from Part 1 — whose cosmotechnics are we participating in? — acquires a material dimension that goes beyond software architecture. The apparatus, in Barad’s sense, does not begin at the interface. It begins in the ground.

Critics have noted that Mbembe can be surprisingly optimistic about digital technologies in the African context, speaking of “Afro-computation” and a “techno-computational revolution.” Victoria Bernal, responding to Mbembe, asks the question that haunts this entire analysis: “What happens when necropolitics are digitized? What happens when commandment has the tools of algorithms and digital surveillance to bring to bear on relations of subjection?”²⁶ The question remains open — and urgent.

Image detail from ‘Evolution II‘ (2025), Oil on Linen, 70 x 50 cm by cornelia es said

VII. The New Game Theory: Cooperation Is Not Idealism — It Is Mathematics

After the darkness of necropolitics, a turn — not toward optimism, but toward evidence. The question is not whether cooperation is morally desirable. It is whether cooperation is structurally superior to the competitive, extractive models that dominate digital infrastructure. Recent developments in game theory suggest that it is.

The dominant model for decades has been the Nash equilibrium, named after mathematician John Nash. Its core claim: rational actors maximize individual self-interest, and the stable outcome of any strategic interaction is the point at which no player can improve their position by unilaterally changing strategy. This model has profoundly shaped law, economics, and technology design — always in the direction of competition, efficiency, and non-cooperation.²⁷

The problem is that the model does not describe how actual humans behave. In experimental settings, people consistently reject unfair offers — even when accepting would be individually rational — out of what researchers describe as moral disgust. In January 1950, Merrill Flood and Melvin Dresher, working at the RAND Corporation, conducted the first experimental study of what Albert Tucker would later name the Prisoner’s Dilemma. Over 100 rounds, their subjects — economist Armen Alchian and mathematician John Williams — chose cooperation far more often than Nash equilibrium would predict. Flood’s own report included Nash’s written comment: the observation that sophisticated subjects did not play the equilibrium strategy of unconditional defection in the repeated game did not imply that such behavior would be irrational in a one-shot version. The disagreement, in other words, was already visible at the origin of the field.²⁸

In 2012, mathematicians William Press and Freeman Dyson introduced Zero Determinant (ZD) strategies — probability-based approaches to repeated games like the Prisoner’s Dilemma that changed the mathematical landscape fundamentally.²⁹ ZD strategies come in two forms:

Extortionate ZD allows one player to unilaterally set the opponent’s score, essentially dominating the game. This strategy works — but only against opponents who lack awareness of the game’s structure.

Generous ZD, identified by Alexander Stewart and Joshua Plotkin in 2013, operates differently. Stewart and Plotkin proved that in evolving populations, extortion strategies are unstable; generous strategies — cooperative by default, forgiving after defection — replace them and dominate the population over time.³⁰ Under certain conditions in repeated games, mutual cooperation through generous ZD produces the maximum achievable payoff for both players.

This is not a moral argument. It is a mathematical proof: in repeated interactions, cooperation outperforms competition in evolving populations. The Nash equilibrium — mutual defection — is the worst outcome for both players. Generous ZD — mutual cooperation — is the best. The mathematics is unambiguous.

VIII. Why Cooperation Fails: The Theft of Theory of Mind

If cooperation is mathematically superior, why do the dominant digital architectures systematically prevent it? This is the question that Part 1 posed and that this essay exists to answer.

The answer lies in a specific condition that Press and Dyson themselves identified in the 2012 paper. They wrote that the iterated Prisoner’s Dilemma between two players with theory of mind — the ability to recognize the other as a strategic agent — becomes an ultimatum game: a player who detects extortion can refuse to play along, absorbing short-term losses to force the extortionist toward fairer terms.³¹ Strictly speaking, the mathematics of ZD does not require consciousness; it requires access to the game’s history and payoff matrix. But Press and Dyson reached for “theory of mind” because the structural condition — mutual recognition that the other is a strategic agent with their own interests — is what transforms unilateral domination into negotiated cooperation. The metaphor is load-bearing, and the authors chose it deliberately.

Dark patterns, analyzed through this lens, are instruments for the systematic destruction of theory of mind in the architectural sense Press and Dyson named. They prevent the user from understanding what the platform wants. Google’s location tracking apparatus does not present itself as a system that extracts behavioral data for advertising revenue. It presents itself as a helpful map, a convenient search engine, a free email service. The structural conflict between user goals and shareholder goals — the tension at the core of the Knowledge Stack — is concealed by an interface designed to project alignment where none exists.

The user cannot cooperate with Google because the user does not know what Google is doing. The conditions for Generous ZD are not met. What operates instead is Extortionate ZD: the platform unilaterally sets the terms, and the user — deprived of the information needed to strategize — complies. Not because she consents. Because the architecture has been designed to make non-compliance costly, confusing, or invisible.

Noble’s and Benjamin’s work sharpens this further. Algorithmic discrimination operates precisely by denying theory of mind to entire communities. When a search algorithm renders Black women visible only through pornographic stereotypes, it denies the humanity — the strategic subjectivity — of the people it represents. When a recidivism algorithm treats historical racial profiling as objective data, it denies the agency of the individuals it scores. When a healthcare algorithm interprets deprivation as lesser need, it denies the reality of the patients it claims to serve. In each case, the algorithm functions as an Extortionate ZD player: it sets the score unilaterally, and the affected population has no mechanism to negotiate, reciprocate, or resist within the system’s own terms.

Mbembe’s necropolitics represents the terminal form of this theft. In necropolitical regimes, there is no pretense of a game at all. The subject is not a player whose theory of mind has been impaired. The subject is not a player. She is the terrain on which the game is played — the resource extracted, the body managed, the life rendered expendable. The theft of theory of mind, at this level, is the theft of personhood itself.

This analysis reveals the central mechanism that connects cookie consent manipulation to border surveillance to algorithmic racism: the systematic prevention of mutual recognition. Dark patterns obscure the platform’s intentions. Discriminatory algorithms deny subjectivity to targeted populations. Necropolitical systems deny humanity altogether. The gradient is continuous, and the technologies are shared.

Image detail from ‘Evolution III’ (2026), Oil on Linen, 50 x 70 cm by cornelia es said

IX. Cooperation Exists — In Nature, In Community, Under Pressure

After the preceding analysis, a reasonable reader might conclude that cooperation is a mathematical curiosity with no real-world application. The evidence says otherwise.

Monarch butterflies, during their annual migration, cluster together in dense formations to survive cold temperatures. The behavior is not altruistic in any sentimental sense. It is structurally cooperative: benefiting the collective directly benefits individual survival. No butterfly sacrifices itself for the group. Each butterfly survives because of the group.³²

Vampire bats engage in reciprocal food sharing, regurgitating blood to roost-mates that failed to feed. In the wild, most sharing occurs between close kin, but Gerald Carter and Gerald Wilkinson showed in a 2013 controlled experiment that past help predicts donations 8.5 times better than relatedness. Bats remember who has fed them; bats who refuse to share are progressively excluded from future sharing networks. Cooperation is maintained not through punishment but through the withdrawal of reciprocity — a biological implementation of the “mild punishment” that game theory predicts is more effective than harsh sanctions.³³

Human communities demonstrate the same dynamics. Elinor Ostrom, who received the Nobel Memorial Prize in Economic Sciences in 2009 for her work on the commons, documented more than 800 cases of communities — small-scale fisheries, indigenous irrigation systems, Alpine pastures, forest commons — that have sustained shared resources for centuries. From these cases she derived eight design principles: clearly defined boundaries, rules matching local conditions, collective-choice arrangements, monitoring, graduated sanctions, accessible conflict-resolution, recognized rights to organize, and nested enterprises for larger systems.³⁴ The “tragedy of the commons” — the depletion of shared resources through individual self-interest — is not an iron law. It is the outcome of specific institutional failures, particularly the failure to maintain what game theory calls theory of mind: the mutual recognition of shared interests.

These models share a structural feature: transparency. In a butterfly cluster, every individual’s position is visible. In a bat roost, sharing and non-sharing are observable. In a small fishing community, reputation is public knowledge. Cooperation succeeds where all actors can perceive each other’s behavior and respond accordingly. It fails where perception is blocked — by distance, by scale, by deliberate obscuration.

The objection writes itself: monarchs do not prove anything about billion-user platforms. Ostrom’s principles emerge from communities that a single person could, in principle, walk through in a day. Translating reputation-based reciprocity to a global information system is not a scaling operation but a category shift. The objection is correct — and it is precisely the question Part 3 must answer. Nature gives the template: cooperation requires structural transparency and mutual recognition. The architectural question is whether those conditions can be engineered at platform scale, in systems where no participant can ever see more than a small fraction of the others. Whether a digital material can be configured to enable theory of mind rather than prevent it.

X. Institutional Implications — And What Remains to Be Built

The synthesis of the Knowledge Stack and the new game theory suggests consequences for institutional design that extend far beyond the tech industry.

In criminal justice, the move from efficiency-based punishment (harsh sentences calibrated to deter) toward restorative and transformative justice mirrors the move from Nash equilibrium (defection as optimal strategy) to Generous ZD (cooperation as structurally superior outcome). Research consistently shows that “mild punishment” — graduated sanctions, community accountability, de-escalation — maintains cooperation more effectively than harsh penalties, which trigger cycles of retaliation.³⁵

In property law and climate policy, the shift from ownership-based models to stewardship-based models reflects the same logic. The commons does not need to be privatized or regulated from above. It needs governance structures that make mutual interests visible and reciprocity possible — stewardship rather than extraction, consensus rather than enclosure.

In design ethics, the Knowledge Stack functions as a diagnostic tool. Practitioners can use it to trace how organizational imperatives (the “town” level) constrain user autonomy at the interface level — often unintentionally. The EU Digital Services Act (2022) prohibits manipulative interface design in Article 25. The AI Act (2024) prohibits subliminal and deceptive AI techniques in Article 5. The forthcoming Digital Fairness Act, scheduled for Commission proposal in Q4 2026 following a public consultation that closed in October 2025, is designed to close the gaps — dark patterns outside platform services, addictive design, unfair personalization, subscription traps — and will only reach application no earlier than 2028–2029 even under favorable legislative conditions.³⁶ And in the United States, the FTC’s 2025 settlement with Amazon over the Prime “Iliad Flow” cancellation labyrinth — four pages, six clicks, fifteen options designed to exhaust the user’s intent to cancel — produced a $2.5 billion penalty and marks the first billion-dollar precedent for dark pattern enforcement at scale.³⁷ Whether enforcement will match ambition globally remains to be seen. But the analytical framework, and the first real financial consequences, are now in place.

And in art? The question that drives this series to its conclusion: if the dominant digital architectures are designed to prevent cooperation, to destroy theory of mind, to obscure the structural tensions that shape our interactions — can an artwork function as a counter-architecture? Can it make visible what the interface hides? Can it build, within the material constraints of code and blockchain, a system that enables mutual recognition rather than preventing it?

The answer requires moving from diagnosis to practice. It requires a material — transparent by structure, permanent by design, decentralized by principle — and an artistic practice capable of working with that material against the logic of the systems it inhabits. That is the subject of Part 3.

XI. Questions That Remain Open

Several questions carry forward into the final essay of this series:

If transparency is the precondition for cooperation, what does transparency look like in a digital medium? Not “terms of service” transparency — not disclosure buried in legal language that no one reads. Structural transparency: a material that is readable, verifiable, and accountable by design. SVG — a visual format that is simultaneously readable code — is one candidate. The blockchain — a ledger that is public by architecture — is another. Part 3 examines what happens when these materials are combined.

If the same technology can serve liberation and oppression — if permanence protects political art from censorship but also permanently exposes its creator — how does an artist navigate the paradox? CC0 licensing, pseudonymous publication, collective wallets, and zero-knowledge proofs are technical strategies. But the tension between visibility and vulnerability is not a technical problem. It is a political one. And it has no clean resolution.

The conatus of code — the striving-to-persist that Part 1 identified in digital materials — has a specific political valence once the analysis of this essay is applied. Blockchain’s permanence is not a neutral technical property. It is the material condition under which a work can testify against the architecture of manipulation described here: a record that cannot be quietly deleted when political winds shift. Whether that permanence depends on an ecologically destructive consensus mechanism (Proof of Work) or on an energy-proportional one (Proof of Stake) is not a secondary technical detail but a constitutive question for any counter-architecture that wishes to remain consistent with the anti-extractive ethic this essay argues for. Part 3 addresses this directly.

Whose cosmotechnics do we participate in when we build on Western blockchain infrastructure? Yuk Hui’s question, raised in Part 1, returns with new force after Mbembe’s analysis. If the technologies of surveillance capitalism and the technologies of necropolitics share infrastructure, is building on that infrastructure always complicity? Or can the material be turned — its structural properties activated against its embedded logic, the way Johnson’s ash wood resists the maker’s assumptions and insists on its own form?

Part 3 does not promise answers. It proposes an experiment.

Notes

1. Ursula Johnson, quoted in Nicole Burisch, “From Objects to Encounters: Renegotiating Craft in the Contemporary Art World,” in Making Otherwise. See Part 1 of this series for the full discussion of material agency. ↩
2. Harry Brignull coined “dark patterns” in 2010. His taxonomy, including “Roach Motel,” “Confirmshaming,” and “Forced Action,” remains the standard reference. See deceptive.design (formerly darkpatterns.org). ↩
3. Christopher Alexander, Sara Ishikawa, and Murray Silverstein, A Pattern Language: Towns, Buildings, Construction (New York: Oxford University Press, 1977). Alexander’s core insight — that design problems exist at multiple nested scales that shape each other — has influenced software architecture (via the “design patterns” movement) and, more recently, dark patterns research. ↩
4. Colin M. Gray, “The Dark Patterns Knowledge Stack: Exploring New Ways to Negotiate Context, Law, and Design,” in Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI ’26), April 13–17, 2026, Barcelona, Spain (New York: ACM). The four-layer framework adapts Alexander’s architectural pattern language to digital systems. See also the related three-level ontology of dark pattern types in Colin M. Gray, Cristiana Santos, Nataliia Bielova, and Thomas Mildner, “An Ontology of Dark Patterns Knowledge: Foundations, Definitions, and a Pathway for Shared Knowledge-Building,” CHI ’24 (Honolulu, 2024). ↩
5. Karen Barad, Meeting the Universe Halfway: Quantum Physics and the Entanglement of Matter and Meaning (Durham: Duke University Press, 2007). On intra-action as the production of agency through entanglement rather than interaction between pre-existing entities, see Chapter 4. ↩
6. Theodor W. Adorno, Minima Moralia: Reflections from Damaged Life (1951), trans. E.F.N. Jephcott (London: Verso, 2005), §18 “Asyl für Obdachlose” (Asylum for Homeless). The German original — “Es gibt kein richtiges Leben im falschen” — has become one of the most quoted sentences in German-language philosophy. ↩
7. On the EU Digital Services Act (DSA), Regulation (EU) 2022/2065, Article 25 prohibits dark patterns on online platforms, with fines up to 6% of global annual turnover under the general DSA penalty regime. On the psychological mechanisms exploited by consent-fatigue and attention-capture designs, see the research summarized in the U.S. Surgeon General’s Advisory on Social Media and Youth Mental Health (2023). ↩
8. The Google location tracking case study draws on research documented in the Knowledge Stack literature. Google’s “harmful defaults” — Web & App Activity opted “on” by default — and the use of proprietary technologies (IPGeo) to infer location even after user opt-out are documented in multiple FTC investigations and state attorney general settlements. Arizona Mirror, September 9, 2022 , and the subsequent settlement announcement ↩
9. On the psychological mechanisms exploited by attention-capture designs, see the work of Tristan Harris (Center for Humane Technology) and the research summarized in the U.S. Surgeon General’s Advisory on Social Media and Youth Mental Health (2023). Variable reward schedules — borrowed directly from slot machine design — are the primary mechanism through which infinite scroll and notification systems maintain engagement. ↩
10. Shoshana Zuboff, The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power (New York: PublicAffairs, 2019). Zuboff defines surveillance capitalism as a new economic order that claims human experience as free raw material for hidden commercial practices of extraction, prediction, and sales. ↩
11. Safiya Umoja Noble, Algorithms of Oppression: How Search Engines Reinforce Racism (New York: NYU Press, 2018). Noble received the MacArthur Fellowship in 2021 for her work in critical information and algorithm studies. She is Professor of Gender Studies and African American Studies at UCLA and co-founder of the UCLA Center for Critical Internet Inquiry (C2i2). ↩
12. Noble, Algorithms of Oppression, Chapter 1. Noble demonstrates that Google’s search results are shaped by the intersection of private advertising interests, Search Engine Optimization (SEO) practices, and the monopoly status of a small number of search engines — producing a biased set of search algorithms that privilege whiteness and discriminate against people of color, specifically women of color. ↩
13. Google Diversity Annual Report 2024. U.S. gender data: 66.2% men, 33.8% women overall; in tech roles, 72.8% men, 27.2% women. U.S. ethnicity data: 5.7% Black, 7.5% Latino. In 2020, approximately 96% of U.S.-based leaders were white or Asian. In February 2025, Google ended its aspirational diversity hiring targets. See belonging.google/diversity-annual-report/2024. Google’s diversity progress is real but concentrated in non-technical roles; the teams that build the algorithms — and the leadership that sets strategic direction — remain substantially less diverse than the overall workforce. ↩
14. On Donna Haraway’s concept of the “god trick” — the claim to see everything from nowhere — see “Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective,” Feminist Studies 14, no. 3 (1988): 575–99. Discussed in detail in Part 1 of this series. ↩
15. Ruha Benjamin, Race After Technology: Abolitionist Tools for the New Jim Code (Cambridge: Polity, 2019). Benjamin is Professor of African American Studies at Princeton University and founder of the Ida B. Wells JUST Data Lab. ↩
16. Benjamin, Race After Technology, Chapter 1 (“Engineered Inequity”). The Beauty.AI contest (2016) is the case study Benjamin uses to demonstrate how training data encoding existing racial hierarchies produces outputs that amplify those hierarchies while claiming algorithmic objectivity. ↩
17. The COMPAS recidivism algorithm analysis was published by ProPublica in 2016: Julia Angwin, Jeff Larson, Surya Mattu, and Lauren Kirchner, “Machine Bias,” ProPublica, May 23, 2016. The investigation found that Black defendants who did not re-offend were misclassified as high risk at 44.9%, compared to 23.5% for white defendants; white defendants who did re-offend were misclassified as low risk at 47.7%, compared to 28.0% for Black defendants. Benjamin discusses the case in Chapter 2 (“Default Discrimination”). ↩
18. Ziad Obermeyer, Brian Powers, Christine Vogeli, and Sendhil Mullainathan, “Dissecting Racial Bias in an Algorithm Used to Manage the Health of Populations,” Science 366, no. 6464 (2019): 447–53. The study found that correcting the algorithm’s racial bias would increase the percentage of Black patients receiving additional clinical attention from 17.7% to 46.5%. The algorithm affected an estimated tens of millions of patients across major U.S. hospital systems. ↩
19. Benjamin, Race After Technology, Introduction. Benjamin makes a compelling case for race itself as a kind of technology, “designed to stratify and sanctify social injustice as part of the architecture of everyday life.” ↩
20. Achille Mbembe, “Necropolitics,” Public Culture 15, no. 1 (2003): 11–40, trans. Libby Meintjes. Expanded in Necropolitics (Durham: Duke University Press, 2019), trans. Steve Corcoran. Mbembe is Research Professor in History and Politics at the Wits Institute for Social and Economic Research, Johannesburg, and visiting professor at the Franklin Humanities Institute, Duke University. ↩
21. Mbembe, Necropolitics (2019), Chapter 4 (“Viscerality”). The digitalization of sovereignty operates through biometric identification, facial recognition, location tracking, and other intelligence technologies deployed to constrain movement and raise physical and digital barriers of enclosure. ↩
22. Mbembe, “Necropolitics” (2003), p. 40. “Death worlds” describes conditions in which vast populations are subjected to an existence that confers upon them the status of living dead. ↩
23. On the FTC’s January 2024 settlement with X-Mode Social/Outlogic for selling location data to government contractors, see Federal Trade Commission, “FTC Order Will Ban InMarket from Selling Precise Consumer Location Data” and the related X-Mode/Outlogic consent order (FTC File No. 212 3038, 2024). On Venntel’s contracts with Immigration and Customs Enforcement (ICE), Customs and Border Protection (CBP), the Department of Homeland Security, and the Department of Defense, see the U.S. House Judiciary Committee memo on “The Broker Loophole” and investigative reporting in Wall Street Journal and Motherboard (2020–2024). On Babel Street’s LocateX tool and its use by federal law enforcement, see 404 Media investigative reporting (October 2024). The Fourth Amendment Is Not For Sale Act (H.R. 4639) was included in the House-passed FISA Reform and Reauthorization Act of 2024 but has not advanced in the Senate. In the EU, the Law Enforcement Directive (Directive (EU) 2016/680) regulates state access to personal data processed by competent authorities, with contested boundaries regarding data sourced from commercial brokers. ↩
24. Mbembe, Necropolitics (2019), Chapter 2 (“The Society of Enmity”), pp. 57–58. “Hydraulic racism” describes institutional and systemic racism (state, law, administration); “nanoracism” describes the daily micro-aggressions designed to stigmatize, injure, and humiliate. ↩
25. On cobalt mining in the DRC and its human cost, see Siddharth Kara, Cobalt Red: How the Blood of the Congo Powers Our Lives (New York: St. Martin’s Press, 2023). Kara documents conditions including child labor, artisanal mining without safety equipment, and armed oversight. On lithium extraction in South America, see Thea Riofrancos’s research on the Atacama salt flats. On electronic waste processing, see the Basel Action Network’s documentation of Agbogbloshie, Ghana, and similar sites. The material supply chain of digital technology — from mine to device to landfill — follows routes that map closely onto colonial extraction patterns, a point developed by scholars including Jussi Parikka, A Geology of Media (Minneapolis: University of Minnesota Press, 2015), who argues that media technologies must be understood through their material substrates, from rare earth minerals to energy consumption. ↩
26. Victoria Bernal, “Digitality and Decolonization: A Response to Achille Mbembe,” African Studies Review (Cambridge University Press, 2021). Bernal’s critique addresses Mbembe’s surprisingly optimistic view of digital transformation in Africa, asking what happens when the tools of digital surveillance are placed in the hands of necropolitical power. ↩
27. On the influence of the Nash equilibrium on legal and economic thought, see the analysis in the game theory literature synthesized in Part I of the source document. Nash’s model prioritizes “efficiency” — the optimal allocation of resources given individual self-interest — over fairness or cooperative outcomes. ↩
28. On the 1950 RAND experiment by Flood and Dresher, and Tucker’s subsequent formalization that gave the Prisoner’s Dilemma its name, see Steven Kuhn, “Prisoner’s Dilemma,” Stanford Encyclopedia of Philosophy (substantive revision 2019), and William Poundstone, Prisoner’s Dilemma (New York: Doubleday, 1992). Nash’s ten-sentence comment on the draft of Flood’s 1952 RAND report is reproduced as a footnote in the final document (Merrill M. Flood, “Some Experimental Games,” RAND Research Memorandum RM-789, 1952). ↩
29. William H. Press and Freeman J. Dyson, “Iterated Prisoner’s Dilemma Contains Strategies That Dominate Any Evolutionary Opponent,” Proceedings of the National Academy of Sciences 109, no. 26 (2012): 10409–13. ↩
30. Alexander J. Stewart and Joshua B. Plotkin, “From Extortion to Generosity, Evolution in the Iterated Prisoner’s Dilemma,” Proceedings of the National Academy of Sciences 110, no. 38 (2013): 15348–53. Stewart and Plotkin show that in evolving populations of sufficient size, extortion strategies are evolutionarily unstable and are replaced by generous ZD strategies. See also Christian Hilbe, Martin A. Nowak, and Karl Sigmund, “Evolution of Extortion in Iterated Prisoner’s Dilemma Games,” PNAS 110, no. 17 (2013): 6913–18, for the parallel evolutionary analysis. ↩
31. Press and Dyson, “Iterated Prisoner’s Dilemma,” p. 10413. They write that when both players have “theory of mind” — the capacity to recognize the other as a strategic agent — the iterated game becomes “an ultimatum game,” in which the extorted player’s refusal to cooperate converts short-term losses into long-term fairness. The term is the authors’ own, not a later interpretive gloss; it captures the structural condition under which mutual recognition replaces unilateral domination. ↩
32. On monarch butterfly clustering behavior as a cooperative survival strategy, see Lincoln P. Brower, “Understanding and Misunderstanding the Migration of the Monarch Butterfly (Nymphalidae) in North America: 1857–1995,” Journal of the Lepidopterists’ Society 49, no. 4 (1995): 304–85. ↩
33. Gerald S. Wilkinson, “Reciprocal Food Sharing in the Vampire Bat,” Nature 308 (1984): 181–84, documented the original observation; Gerald G. Carter and Gerald S. Wilkinson, “Food Sharing in Vampire Bats: Reciprocal Help Predicts Donations More Than Relatedness or Harassment,” Proceedings of the Royal Society B 280, no. 1753 (2013): 20122573, established experimentally that past help predicts donations 8.5 times better than relatedness. The system is maintained through partner choice and the withdrawal of reciprocity from non-sharers — a mechanism consistent with game-theoretic predictions about graduated sanctions. ↩
34. Elinor Ostrom, Governing the Commons: The Evolution of Institutions for Collective Action (Cambridge: Cambridge University Press, 1990). Ostrom received the Nobel Memorial Prize in Economic Sciences in 2009 (shared with Oliver E. Williamson) — the first woman to receive the prize. The eight design principles for durable common-pool resource institutions are developed in Chapter 3. On the limits of the framework and the debate around scale, see also Elinor Ostrom, Understanding Institutional Diversity (Princeton: Princeton University Press, 2005). ↩
35. On graduated sanctions and de-escalation as more effective cooperative-maintenance strategies than harsh punishment, see Ostrom, Governing the Commons, Chapter 3. The game-theoretic basis is consistent with Generous ZD: harsh punishment triggers defection cycles, while mild, graduated responses maintain the conditions for reciprocity. ↩
36. EU Digital Services Act (DSA), Regulation (EU) 2022/2065, Article 25: prohibition of dark patterns in online platform interfaces. AI Act (Regulation (EU) 2024/1689), Articles 5(1)(a) and 5(1)(b): prohibition of subliminal, manipulative, or deceptive AI techniques. On the Digital Fairness Act: the European Commission’s Digital Fairness Fitness Check (SWD(2024) 230) was published in October 2024; the public consultation on the forthcoming Digital Fairness Act ran from 17 July to 24 October 2025; the 2030 Consumer Agenda (19 November 2025) confirmed the DFA as a headline legislative initiative; the Commission work programme for 2026 schedules the proposal for Q4 2026. Under the ordinary legislative procedure, adoption is not expected before late 2027 / early 2028 and application no earlier than 2028–2029. See European Parliament Legislative Train Schedule, “Digital Fairness Act” (retrieved April 2026). ↩
37. Federal Trade Commission v. Amazon.com, Inc. (W.D. Wash., filed June 2023, settled September 2025). The settlement imposed a $2.5 billion payment and injunctive relief requiring Amazon to simplify Prime cancellation and to provide a clear, conspicuous option to decline enrollment. The FTC complaint described the cancellation flow — internally code-named the “Iliad Flow” — as a four-page, six-click, fifteen-option sequence “seeking to entice consumers not to cancel.” By Amazon’s own estimation, the practice resulted in more than 35 million nonconsensual Prime enrollments over a seven-year period. See FTC press release, “FTC Takes Action Against Amazon for Enrolling Consumers in Amazon Prime Without Consent and Sabotaging Their Attempts to Cancel” (June 2023), and subsequent trial reporting (September 2025). ↩