Administrative X-Risk: Pinpointing the Flaws of an AI Regulatory Scheme Reliant on Administrative Action

Abstract

In light of the fact that no existing entity has all the resources and attributes required to exclusively or predominantly regulate AI, advocates should foster regulatory resiliency and innovation by empowering a litany of entities to play a part in mitigating AI risk. In other words, advocates should focus federal resources on seeding regulatory resilience and innovations while placing plans to build a centralized, potentially-fragile, and likely-inadequate AI Agency on the backburner until more is known about AI, its risks, and the best means of mitigating those risks.

A good first step would be the creation of an "AI Smithsonian" that offers two improvements on an AI Agency: first, it would be structured in a way that avoids administrative x-risks such as the Congressional Review Act; and, second, it would have a responsible regulatory footprint--rather than crowd-out regulatory resilience by luring talent, resources, and authority from other entities, it would provide those entities with a source of reliable information and provide them with opportunities to collaborate.

The polycentric regulatory system called for by this essay has emerged in other regulatory contexts, such as automobile safety testing and standard development. Given the complex and shifting nature of AI development, however, waiting for a similar system to “emerge” will introduce an unacceptable amount of delay into this endeavor. That’s why this essay urges a proactive polycentric approach that assigns the U.S. federal government a regulatory role that reflects its institutional capacity, current and forthcoming legal limitations, and dynamic social license to take on certain tasks.

Keywords: artificial intelligence, administrative law, major questions doctrine, separation of powers, institutional design, emerging technology