Aligning Artificial Intelligence to the Law
This piece was originally published in the Villanova Law Review.
Abstract
As artificial intelligence (AI) continues its breathless advance into modern life, figuring out how best to align its behavior with human objectives and values has become a matter of profound societal importance. This Article advances a law-first approach to solving what is popularly known as the AI “alignment problem.”
Prior commentators have envisioned law as a top-down constraint on AI. On this view, law is an umpire, calling foul when an AI system has been inadequately “aligned.” These accounts have largely neglected law’s potential role as a coach: a model of alignment solutions for AI to emulate.
At bottom, the concerns underlying AI alignment are closely related, and at times identical, to those the law has grappled with for centuries in governing relationships among humans, as well as human relationships with prior forms of technology. Because law already contains well-developed alignment solutions—memorialized in authoritative and machine-learnable sources like statutes and caselaw—it can “coach” AI systems to become better aligned from the bottom up.
The Article concludes by canvassing the considerable implications of legal alignment for other areas of scholarship on AI, including risk regulation, legal constructivism, legal personhood, tort liability, computational law, and soft law.