Artificial Power

Origin

The concept crystallised in The Political Philosophy of AI (2022) but developed through Coeckelbergh’s earlier work on ethics and language. Once the relational framework established that AI systems are already embedded in relations that carry moral weight, and once the linguistic analysis established that the vocabulary of “intelligence” and “autonomy” performs specific political functions, the reframing to power became the natural next step. The question of whether algorithmic bias is a technical defect or a political question—and Coeckelbergh is unequivocal that “whether something is biased, and even if it is biased, whether that bias is problematic or not, it’s very much a political question”—is not answerable by better data science alone. It requires a theory of whose interests matter and how competing interests should be weighed. That theory is politics.

The concept gained particular urgency with his 2024 book Why AI Undermines Democracy and What To Do About It, where the power framing moves from diagnosis to political theory. AI is not merely powerful in a general sense; it is powerful in ways that specifically corrode the conditions—shared truth, protection against manipulation, rough distribution of power, the accountability of governing systems—that democracy requires to function. The threat is not incidental; it follows from the design incentives and economic logic of the systems as currently built and deployed.

Key Ideas

The politics is the substance. A hiring algorithm that disadvantages a protected group is not primarily an engineering problem with a technical solution. It is a political problem—a question about who bears the costs of automation, whose historical disadvantages are reproduced by systems trained on historical data, and whose definition of “qualified” the algorithm encodes. Technical fixes may improve the system’s performance by some metric, but the choice of metric is itself a political choice. The language of performance improvement conceals the political question rather than answering it.

Concentration and non-domination. The AI economy is characterised by extreme concentration: the data, the compute, and the models reside in a very small number of private firms. Drawing on the republican political tradition, Coeckelbergh argues that this concentration is an affront to freedom even when the concentrated power is currently exercised benignly. To live within systems that could shape your opportunities, information environment, and life chances—systems owned by actors you cannot hold to account—is to live in a condition of domination. Freedom as non-domination, the republican insight, condemns not only the exercise of arbitrary power but its mere existence as a structural condition.

Responsibility and the problem of many hands. When a powerful AI system causes harm, responsibility dissolves across the long chain of actors who designed the architecture, gathered the training data, optimised the objectives, deployed the system, and used it. Coeckelbergh calls this the “problem of many hands”—a condition in which the complexity of the system becomes a mechanism for laundering accountability. The political philosophy response is institutional: design accountability structures that assign responsibility clearly and enforce it reliably, so that the convenience of automation is never purchased at the price of the public’s ability to demand an answer.

Against ethics washing. Ethics boards, responsible AI principles, and corporate AI governance frameworks are, in Coeckelbergh’s assessment, necessary but radically insufficient as responses to AI as artificial power. Ethics that arrives after the architecture is poured can rearrange the furniture; it cannot move the walls. Ethics that operates at the pleasure of the firm—convened by management, dissolved when inconvenient—is not accountability. The response to artificial power must be external, binding, and democratic: the whole apparatus of law, regulation, public institutions, and democratic participation that societies have developed over centuries for constraining the exercise of power over people who did not consent to it.