From reaction functions to the assumption layer
What do economists or social scientists actually do? Ultimately in our attempt to understand society and human behavior more generally, one key goal is to learn “reaction functions”. That is, we make structured attempts to know or learn and predict how individuals, firms, voters, bureaucrats, etc respond when incentives change, when prices move, when narratives shift, when constraints bind, or when risks materialize.
As social scientists we infer patterned responses from observed behaviour and describe “mechanisms” ultimately as distinct species or profiles of changes in behavior. As I documented in the work on causal claims in economics, over the past decades, the way social science research is carried out has seen a marked shift towards empirical analysis and (attempts to make) sharp causal claims.
The advent of AI drastically increases the scale, speed, and intimacy of this pattern detection and matching exercise. What was once slow, academic, and approximate becomes fast, automated, personalized, and operational. AI does not merely help us study reaction functions, but it may enable governments, firms, platforms, and intelligence systems to estimate and shape them in near real time. And once that happens, the decisive question is no longer simply what people do, but which assumptions about human motivation are built into the systems that predict and steer them.
This is where the “assumption layer” becomes central. At the frontier of knowledge, evidence does not interpret itself. Data always enters through priors: assumptions about intent, motivation, trustworthiness, and likely behaviour. The decisive variable is not only what AI models can do, but what people believe others will do with those capabilities.
The assumption layer can be interpreted as a baseline or prior that provides the default position in absence of the ability to have knowledge about a specific situation or context. As AI pushes the boundaries of what is technically possible, and in its process, produces knowledge, the natural question is whether this knowledge will be used to heal, coordinate, and enlarge freedom, or to profile, corner, and manage people into compliance?
It also raises the possibility that knowledge simply will not be understood by human agents anymore. As a result, humans have to either accept the boundaries of their own (ability to) understand or, and that is the more concerning possibility, they may fall back to a set of assumptions that they hold – or respectively, their “base programming”, to fill the cognitive void that seeks explanation.
It naturally does not help that the human brain is a threat prediction engine and as such, we are naturally skewed in some directions for our attention mechanism.
When inference becomes a technology of power
We can call this ideology. In “inference meets retrieval”, I have highlighted my distinct concern about this possibility: action on information or knowledge derived through low dimensional retrieval is incredibly dangerous.
This is to me where the precautionary principle as a guiding principle to how societies manage risks always dominates the burden of proof approach that is more deeply institutionally anchored in the US.
In this situation, evidence stops functioning as a stabilizer since every claim becomes a move in a game, every measurement a possible weapon, every model a targeting system, every institution a suspected vehicle of capture.
That matters because AI radically expands the ability to align people along different reasoning chains, logics, topologies, or narratives. If humans reason through stories, analogies, fears, identities, status concerns, and perceived threats, then AI gives political and commercial actors a new capacity to map those reasoning structures and intervene inside them.
This is not just persuasion in the old sense, but rather, it is the possibility of topological steering: identifying the path through which a person or group arrives at a conclusion, then feeding them narratives that make a desired conclusion feel internally generated. This will create a perception of co-ownership of an idea. The target is then not merely opinion, but rather, the target is the reasoning chain itself. It enables or may facilitate a tacit and indirect reprogramming of the collective mind, in particular, if paired with (digital) ID that is being rolled out and the all-too-pervasive backdoors making such interference more than just an abstract possibility.
From explanation to intervention
This is the bridge between social science and power. Economists and social scientists estimate reaction functions to understand the world. Platforms and AI systems estimate reaction functions to act upon the world. The former seeks explanation, while the latter enables intervention. Once reaction-function estimation becomes cheap, scalable, and embedded in digital infrastructure, whoever controls the data gains a new form of authority: the authority to say what is likely, what is rational, what is risky, what is credible, what is extreme, what is normal, and what futures should be shaped.
This creates an ultimate demand for ontological authority. If the world becomes too complex for individuals, firms, or governments to process directly, they increasingly outsource judgment to AI systems: summarize this, rank this, predict this, evaluate this, recommend this, govern this. But to delegate analysis is also to delegate parts of reality-construction and ultimately human agency.
And in the process of retrieval, AI does not merely answer questions, rather, it may frame the problem, selects relevant variables, compresses uncertainty, ranks explanations, and implicitly privileges one model of human behaviour over another. If decisions are increasingly mediated by AI, then authority migrates from democratic deliberation, institutional judgment, and human interpretation toward the patterns that are baked in the training data, along with the implicit rules systems that define the operative ontology.
The self-fulfilling logic of extraction
That is why the assumption layer matters so much. If the dominant assumption built into AI-mediated governance is that human beings are fundamentally self-interested, status-seeking, opportunistic, and manipulable, then the systems built on top of that assumption will reproduce exactly such a world.
They will recommend surveillance, and hence, distrust, because defection is expected. The design of incentives will be recommended over trust because intrinsic motivation is discounted. Behavioural nudging will be recommended, because actual deliberation is treated as inefficient. Private intermediation will be advantaged, because public institutions are presumed slow or captured. And most fundamentally: they may recommend competitive positioning because cooperation is interpreted as vulnerability.
In this sense, US-led technology capitalism is not merely exporting platforms, cloud infrastructure, AI models, chips, or payment rails. It is exporting a model of the human being. It embeds an anthropology into infrastructure: the human as a bundle of preferences, fears, insecurities, status anxieties, and predictable responses. The individual becomes a marketizable reaction function. Knowledge becomes power not because it enlightens, but because it allows more precise intervention into the conditions under which people choose.
The danger is that this anthropology becomes self-fulfilling. If institutions are designed on the assumption that everyone is extracting, people begin to defend themselves against extraction. If every actor assumes others are gaming the system, gaming becomes rational. If every state assumes others will weaponize AI, defensive weaponization becomes unavoidable. Distrust is no longer merely a mood, it becomes part of infrastructure: once extraction becomes the default expectation, cooperation starts to look like being the only fool without armour.
This is where capitalism becomes more than an economic system: capitalism, in this deeper sense, is not only private ownership, markets, or profit. It is the generalized belief that self-interest is the most realistic description of human motivation. If this belief is embedded in AI systems, platforms, financial architectures, and geopolitical strategy, it becomes difficult to think or act outside it. Cooperation appears naïve, public purpose appears rhetorical and the ability for societies to solve collective action problems may degenerate.
More crucially, shared institutions that require solidarity, appear fragile. The assumed extractive behaviors of others becomes what is widely accepted as “real”.
The concern is that the US is weaponizing the latent assumption layer of self-interest, making actors believe that others are fundamentally guided by private advantage, thereby eroding performative trust and increasing dependence on platform-based intermediation. The point is not that every American actor consciously intends this. The point is structural: US technology capitalism benefits from a world in which trust is thin, public institutions are weak, and private platforms become the systems through which risk, identity, credibility, and coordination are managed.
Europe, China, and the geopolitical trap
This has a geopolitical signature. The EU is caught because it epitomizes a cooperative alternative. It is a political formation built around pooled sovereignty, mutual recognition, law, standards, regulatory trust, and the idea that interdependence can be governed rather than dominated. But the EU’s model depends on a functioning trust architecture. If it relies on US-controlled cloud, AI, data, and platform systems, it risks importing an infrastructure whose logic is private intermediation, extraction, and behavioural steering. If it asserts digital sovereignty, it is framed as bureaucratic, protectionist, anti-innovation, or strategically weak. The cooperative system is thus judged by the standards of the very techno-capitalist model that corrodes the conditions under which cooperation can thrive.
China is caught in a still deeper and more existential version of this trap. The CCP’s legitimacy still rests on a communist ideological inheritance that ultimately rejects the land-based rentier system, private extraction, inherited privilege, and the organization of society around self-interested accumulation. The Party can tolerate markets as instruments of development. It can tolerate entrepreneurs, firms, and private wealth if these remain subordinate to national rejuvenation and collective purpose. But it cannot easily tolerate the deeper conclusion that capitalist subjectivity has won: that people ultimately orient themselves around private advantage, property, status, and individual advancement rather than collective purpose.
This is why the threat is not merely external containment. It is internal ontological erosion. China’s development model has already generated many capitalist social realities: property accumulation, local-government land finance, household wealth strategies, education races, entrepreneurial ambition, consumer aspiration, and status competition. If US-led technology capitalism succeeds in making individualism and self-interest appear universal and unavoidable — even inside China — then the CCP is not merely geopolitically pressured. Its ideological foundation is hollowed out from within.
That is China’s impossible situation. If China opens itself fully to US-dominated digital infrastructure, it exposes itself to systems whose operating logic is extraction, profiling, behavioural prediction, and geopolitical leverage. But if it resists through data localization, domestic platforms, industrial policy, ideological control, or digital sovereignty, it appears to confirm the Western narrative that it is closed, authoritarian, paranoid, and untrustworthy. Cooperation becomes almost impossible because every move is interpreted through the assumption that all actors are ultimately self-interested and strategic.
The CCP’s dilemma is sharper still: it needs markets for innovation and growth, but it must prevent market society from producing a fully capitalist subject. It needs platforms for productivity and governance, but it cannot allow platform capitalism to become a rival sovereign. It needs AI, but it cannot allow AI-driven attention systems to produce atomized, status-hungry, individually optimized citizens whose primary horizon is private advancement. It needs openness, but it cannot allow openness to dissolve the ideological basis of Party rule.
The anthropology beneath the technology
The deepest struggle, then, is not democracy versus authoritarianism, or capitalism versus communism in the old Cold War sense. It is a struggle over the anthropology embedded in AI-mediated society. Are humans relational beings capable of trust, stewardship, and collective life? Or are they self-interested agents whose behaviour must be predicted, nudged, priced, and controlled? The answer matters because AI systems will not merely reflect that anthropology, rather they will help stabilize it. They will train institutions, governments, and individuals to act as if that model of the human being were true.
A different information topology
The proposed alternative points toward a different information topology: less raw data sharing, more validated knowledge sharing, stronger privacy, selective disclosure, auditability, zero-knowledge proofs, and institutions that reward validation rather than mere narrative production. This is the institutional logic behind Europe’s federated democratic learning state: a European architecture in which public institutions can compute more while seeing less, turning protected administrative and economic data into auditable public knowledge without constructing a centralized surveillance system. The point is not to reject AI, it is to prevent AI from becoming an extraction machine operating on human reasoning traces. The alternative is a technology stack that supports auditable cooperation rather than personalized distrust.
So the full argument is this: social scientists have always tried to infer reaction functions, but AI turns reaction-function estimation into a scalable technology of power. As more decisions and interpretations are delegated to AI, authority shifts toward whoever controls the architecture of reality compression (media!). This creates a demand for ontological authority: systems that tell us what kind of world we are in and what kind of beings humans are. If US-led technology capitalism controls that layer, it can entrench self-interest as the default model of human behaviour. Once that happens, cooperation becomes unbelievable, collectivist systems become internally unstable, and the EU’s cooperative legal-institutional model and China’s communist ideological project are both forced to operate defensively inside an infrastructure built around extraction. The battle is therefore not only over technology. It is over the maintained assumption layer of civilization itself.