AI-Enabled Authoritarian Practices in Formal Democracies

Abstract

Western political scientists' treatment of digital authoritarianism has mostly been anchored in the concept of ``authoritarian regime.'' Constantly framing discussions in terms of regime type deflects attention from authoritarian phenomena in Western formal democracies. Moreover, despite accounting for some advances in machine learning, the literature often falls behind rapid technological innovation. This paper applies a ``practice''-based definition of authoritarianism to identify and analyze AI-enabled authoritarianism in democratic systems. We trace historical continuity in authoritarian practices in the national security apparatus of democratic states, while also describing new modes of authoritarianism made possible by deep learning. Characteristics of AI that surveillance and privacy scholars have argued occasion new kinds of social power are here shown to be specifically authoritarian affordances. These affordances can be used by power-wielders within a democratic state to disable the voice and information access not only of the public at large, but of intra-government actors tasked with holding those power-wielders to account.