Taking Risks on Behalf of Others: When to Defer, When to Decide

Abstract

This thesis examines what risk attitude decision-makers ought to adopt when making decisions on behalf of others and applies this normative framework to policymakers regulating emerging technologies. Specifically, I survey cases of therapeutic applications of artificial intelligence (AI) because this technology possesses great potential for both real benefits and harms. I argue that when patients’ risk attitudes are known, a deferential approach is the only normatively appropriate response. When patients’ risk attitudes are unknown, the appropriate approach depends on the number of people affected and the frequency of choice. For low-frequency decisions, I argue that a permissive approach may be taken by the decision-maker when acting on the behalf of few people, but avoiding risk is required when acting on the behalf of many. For high-frequency decisions, decision-makers must maximize expected utility. Drawing on decision theory and the philosophical literature on risk, I develop a context-sensitive framework that offers practical guidance for policymakers navigating these decisions. While regulating emerging technology is undeniably difficult, this thesis argues that policymakers need not consider these decisions as categorically different from the perspective of risk attitudes.