Autonomous AI systems are not politically neutral technical achievements. They embed a specific ideology — one that privileges optimisation, competition, and measurable performance — that is historically linked to neoliberal market capitalism and can be traced from the mathematical foundations of reinforcement learning through to system deployment and governance.
Introduction
Recent debates on artificial intelligence increasingly challenge the idea that autonomous AI is a politically neutral technical achievement. Across technical AI research, intelligence is commonly operationalised as the capacity of an agent to optimise behaviour relative to goals, rewards, or tasks. Dan McQuillan's political-economy framing captures this crisply when he defines AI as the construction of autonomous systems that maximise some notion of reward.
This definition reveals a deep affinity between dominant AI paradigms and broader social logics that privilege optimisation, competition, efficiency, and measurable performance. The central question of this review is whether this architecture of intelligence is simply a convenient engineering abstraction or whether it reflects a historically specific ideology linked to neoliberal market capitalism.
Technical Foundations
The technical literature on autonomous agents provides the first basis for this argument. Multi-agent reinforcement learning frequently defines success through competitive or quasi-market environments in which agents adapt to scarce information, strategic opponents, or shifting incentive structures. Classic work on multi-agent actor-critic methods explicitly models "mixed cooperative-competitive environments."
OpenAI's hide-and-seek research is especially revealing: increasingly sophisticated tool use emerges not from moral reflection or deliberation but from recursive strategic pressure generated by competition and counterstrategy. Surveys of LLM-based autonomous agents similarly organise the field around construction, capability enhancement, application, and evaluation — reinforcing a view of agents as optimisable task performers.
The mathematical grammar of reinforcement learning — reward maximisation, policy optimisation, competitive equilibrium — did not arise in a vacuum. It was shaped by and continues to reinforce a specific conception of intelligence as strategic self-interest. This is not inevitable; it is a design choice with ideological consequences.
Political Economy & Critical Theory
This design grammar becomes more politically charged when read through critical political economy. Verdegem argues that AI capitalism is characterised by commodification of data, concentration of talent and compute, and enclosure under a winner-take-all principle. That argument matters for autonomous AI because it links the technical architecture of agents — optimisation, self-interest, competition — to the economic architecture of platforms, markets, and capital accumulation.
Foucauldian analysis adds a complementary angle. Where political economy emphasises structural constraint, Foucault's concept of biopolitics illuminates how AI governance increasingly operationalises the measurement, classification, and management of human conduct. Risk scoring, predictive policing, and hiring algorithms all constitute governance through quantification — a form of power that is both productive and normalising.
Science & Technology Studies
Science and Technology Studies (STS) approaches treat AI not as a neutral artefact but as a sociotechnical assemblage whose meanings, uses, and effects are co-produced through interaction between technical systems and social contexts. Wendy Chun's analysis of habitual new media illuminates how algorithmic systems generate subjects — users who come to understand themselves through the categories and interfaces that AI produces.
Lucy Suchman's work on human-machine reconfigurations is particularly relevant: she argues that AI systems are not simply automating human tasks but reorganising the social and material arrangements through which work, knowledge, and value are produced. The apparent neutrality of technical systems conceals ongoing processes of political negotiation, value inscription, and power allocation.
Alternatives & Counter-Arguments
The literature does not support a simple or totalising ideological critique. Several important counter-arguments exist. First, cooperative AI research — including work on mechanism design, social welfare functions, and multi-stakeholder alignment — demonstrates that the competitive paradigm is not technically mandatory. Second, open-source AI development, though contested, represents a genuine challenge to the enclosure thesis. Third, recent work in AI ethics and value alignment explicitly attempts to inscribe non-market values — fairness, dignity, care — into system design.
The most sophisticated position in the literature acknowledges that AI systems carry ideological tendencies without being ideologically determined. The design space is contested, and outcomes depend on political choices about governance, ownership, and the values that are embedded in both technical architectures and deployment contexts.
The ideological critique of autonomous AI is substantiated by the literature but requires nuance. The critical task is not simply to denounce AI as ideological but to identify the specific mechanisms through which particular values become embedded, to make those mechanisms visible, and to open them to democratic contestation.