Michel Foucault is among the most important and most applied postmodern theorists for understanding AI. His concepts of power-knowledge, discourse, and disciplinary society provide a framework for analysing AI systems that is both intellectually rigorous and immediately practical.
Power-knowledge. Foucault’s fundamental insight is that power and knowledge are not separate, power does not merely distort knowledge from outside, nor does knowledge exist independently of power. They are mutually constitutive. Power produces knowledge, and knowledge produces power. Every system of knowledge, every science, every discipline, every expertise, is simultaneously a system of power, defining who counts as an expert, what counts as evidence, whose experiences count as data.
Applied to AI: the technical expertise that produces AI systems is simultaneously a form of power. The engineers, data scientists, and executives who build AI systems are not merely applying neutral technical knowledge, they are exercising power, defining what counts as a relevant feature, what counts as an accurate prediction, what counts as a fair outcome. The authority of their technical expertise is itself a form of power, one that is rarely subject to democratic accountability.
Discourse. Foucault analysed discourse as the framework within which knowledge is produced, the rules that determine what can be said, what counts as a meaningful statement, who has the authority to speak, and what falls outside the boundaries of legitimate speech. Discourse is not merely language. It is the structured system of knowledge-production that determines what is thinkable and what is unthinkable.
In AI, dominant discourses determine what questions get asked, what data gets collected, what counts as a problem, and what counts as a solution. The discourse of AI safety focuses on existential risk from superintelligence, a framing that diverts attention and resources from the immediate, documented harms of current AI systems. The discourse of AI ethics tends to frame problems as technical, bias, accuracy, explainability, rather than political, which positions technical experts as the primary problem-solvers and leaves structural questions unasked.
Disciplinary society and the panopticon. Foucault’s analysis of modern disciplinary institutions, the prison, the hospital, the school, centred on Jeremy Bentham’s panopticon: a circular prison in which all cells are visible from a central tower, meaning that prisoners can be observed at any time without knowing when they are being observed. The effect, Foucault argued, is not the actual observation of prisoners but the internalisation of the observer, prisoners begin to discipline themselves because they know they might be watched.
AI surveillance is the digital panopticon. Workers who know their performance is monitored by AI begin to self-regulate in ways they might not if they knew they were unobserved. Social media users who know their content is subject to algorithmic moderation adapt their expression to anticipated algorithmic responses. Citizens in cities with AI surveillance infrastructure modify their behaviour in public spaces. The effect of AI surveillance is not merely the information it gathers. It is the disciplinary power it exercises through the mere possibility of observation.
Reflection question: Where in your professional or personal life do you modify your behaviour because you know or suspect you are being observed by an AI system? What has that changed?