A tale of two Regulatory Markets: the role of institutional incentives in supporting sustainable Regulatory Markets for future AI systems
Paolo Bova, Alessandro Di Stefano, The Anh Han · 2023
In the near and long term, the deployment of powerful AI capabilities raises concerns of accidents, misuse, and systemic risk (Brundage et al., 2018; Shevlane and Dafoe, 2019; Zwetsloot and Dafoe, 2019; Hernandez-Orallo et al. ´ , 2019). These capabilities also require new techniques to audit and certify (Cihon et al., 2021a; Gursoy and Kakadiaris, 2022). Regulatory Markets could help AI governance to be more adaptive (Clark and Hadfield, 2019). Governments set targets and mandate that companies employ the services of private regulators to demonstrate compliance with those targets. Private regulators must compete with each other to regulate AI companies. This competition may lead to innovations in methods to detect unsafe behaviour and better understand what safe development practises look like. While the size of these benefits is uncertain, regulators must be incentivised to invest in better methods in the first place. One can ask what role governments can play in providing incentives for higher quality regulators to join the regulatory market. To this end, this extended abstract highlights findings from a recent evolutionary game analysis. The paper explores how different institutional incentives influence the evolutionary dynamics of interactions between AI companies and regulators (Bova et al., 2023). Namely, the paper considers two types of incentives governments might consider, showing that only one of these types, dubbed ”Vigilant Incentives”, can support regulators in evaluating cuttingedge AI systems.