This article argues that AI presents a two-pronged power challenge: AI introduces a different type of power relationship while simultaneously eroding the efficacy of existing procedures and institutions that are traditionally appealed to for resisting against power disparities.

The first prong of the AI power challenge is broken down into an analysis consisting of three levels of power (roughly mapping onto the radical view of power proposed by Steven Lukes), namely: (i) power exercised over the individual or groups in mundane spheres of activity where certain kinds of everyday decision-making may be displaced; (ii) power impacting upon the trajectories of societal development and hence impinging upon human rights, values, and aspirations, and their track-dependencies; and (iii) power involving existential threats to humanity. Each of these levels brings into play a series of regulatory challenges that do not sit comfortably together. At the first level, existing legal processes may be capable of coping with the challenges through creative interpretation and new legislation. The second level, however, is less easily contained because it involves structural questions of new rights and duties. Finally, with the prospect of AI posing existential threats to humanity, the vantage point offered by the power discourse may circumvent thorny hypothetical question marks over ‘intelligence’, but reveal other shortcomings of the power discourses and responses framed within this paradigm.

The second prong of the challenge is addressed with reference to the tendency of AI both to provoke a sense of human inferiority and to erode our means of checking power. This provides insight not necessarily into how AI may destabilise power relationships, but rather illustrates some of the shortcomings of our existing systems which have not been revealed because they have not been tested in such a manner.

Concluding, it is suggested that the focus upon responding to and regulating AI might be either overly specific or missing an important point. Rather, if the core challenges posed by AI are viewed as problems of power, this will not only unify hitherto divergent responses but also shield us from the technological dazzle that prevents us from seeing these problems clearly.