The (false?) promise of human control over machines

Human oversight is advocated in AI policy as a solution to the problems of algorithmic decision making (ADM) for automating decision making. Human oversight promises control over algorithms that can be translated into procedural safeguards in ADM design. Yet I argue that human oversight builds on a false assumption of separability of human and machine action, which leads to misconceptions about its potential and limitations in producing control. A more suitable framing can be found in hybridisation of decision making in complex socio-technical systems, where human work is mediated through ADM systems. Focus on hybridization opens up questions on meaningful human-computer partnerships and interface design. Human oversight reflects broader societal fears about automation. Ultimately, human control reveals a connection between human agency, legitimacy of decision making and social expectations of fairness that enables new critical analysis of law’s anthropocentricity.