Download this Insight (PDF)
Abstract
This Regulatory Insight presents a practical, evidence-based governance framework for policy officials seeking to address artificial intelligence risks without stifling technological innovation and competitiveness. It offers four foundational concepts for effective AI policy, along with policy implications and challenges: 1) establishing direct human accountability and legal liability rather than treating autonomous models as rogue entities; 2) adopting an adaptive learning agenda that incorporates iterative reassessment and domain-tailored oversight; 3) preserving market competition by preventing regulatory capture and burdensome compliance regimes that entrench dominant incumbents; and 4) accounting for countervailing risks by recognizing that robust, accountable AI tools may be the best defense against AI-enabled threats. It recommends an agile, outcome-focused regulatory model that balances targeted deterrence with the operational flexibility needed to foster beneficial innovation.
Introduction
Recent cybersecurity events have brought the potential dangers of artificial intelligence (AI) into the spotlight. Legislators seeking to address these concerns face a challenge: can they design institutions to mitigate novel risks without stifling American innovation and competitiveness?
The George Washington University Regulatory Studies Center (RSC) has been conducting and supporting applied academic research aimed at addressing the challenge of effectively regulating rapidly emerging technology. We hosted an event at the end of September where a diverse group of experts shared their insights. Videos from that event and underlying research are available here.
This Regulatory Insight highlights practical, actionable, and evidence-based recommendations that emerge from that work. It attempts to identify some concepts for thinking about AI governance that are likely to be more constructive than either a precautionary focus on preventing speculative existential scenarios or a cavalier insistence on maintaining the status quo.
The remainder of this Insight lays out four essential concepts: 1) humans should be held accountable, 2) policies should incentivize learning and adaptation, 3) regulatory capture can entrench dominant incumbents, and 4) risk frameworks must account for countervailing risks and strategic adversaries (recognizing that accountable AI is often the most effective defense against AI threats). For each concept, it identifies possible policy actions and acknowledges challenges.
While these concepts and policy implications do not provide definitive prescriptions for devising AI policy, they may help provide a framework for thinking about next steps.