AI algorithmic oversight: new frontiers in regulation

Abstract

The rise of artificial intelligence (AI) is transforming our societies, permeating an ever-expanding array of domains ranging from finance, to employment or healthcare. AI is prognosticated to be a key driver of economic growth in years to come (European Commission 2018). Importantly, the reliance on the technology is also increasingly spilling over from private into public domains. AI is becoming a "game changer" for public bureaucracies and our public institutions, with public power increasingly exercised algorithmically. Public regulators, among others, increasingly rely on machine learning techniques in the discharge of regulatory tasks (Yeung 2018; Yeung & Lodge 2019) -for instance, in prioritising regulatory targets for regulatory attention and/or enforcement, in areas as varied as environmental protection, securities or food safety. Algorithmic enforcement tools are being relied upon by public regulators, among others, to identify misconduct, to detect fraud in reporting, to prioritise sites or regulatees for inspection, in an effort to optimise the use of limited resources at their disposal (Engstrom et al. 2020;Coglianese & Lehr 2017). But the use of AI in government stretches well beyond the field of regulation, making significant inroads into the coercive and (re)distributive powers of the state itself (Pasquale 2011;Engstrom et al. 2020). Algorithmic outputs are used by police to predict crime in the context of predictive policing technologies; for teacher appraisals and to inform school reforms; and by public authorities to allocate grants, entitlements and benefits (