Economic Analysis of Ethical Principles for AI Algorithms: An Autonomous Driving Case

Abstract This paper develops an economic framework to evaluate ethical principles for autonomous driving vehicle (ADV) algorithms in trolley dilemma–type accidents. Specifically, we compare three social rules (the utilitarian rule, the Rawlsian rule, and the passenger priority rule) by embedding them into a heterogeneous agents model in which people alternate between being ADV passengers and pedestrians. We provide algorithms that implement the three rules and show that the utilitarian rule is weakly superior in terms of social welfare, though under specific parameter conditions the three rules yield equivalent outcomes. Based on the theoretical results, we derive the following policy implications: First, governments should consider economic factors such as ADV adoption rates and harm discrepancies to choose an optimal social rule; second, complementary policy instruments such as ADV purchase subsidies can increase the adoption incentives without compromising social welfare. The policy discussion also emphasizes that cultural characteristics can affect enforcement costs of optimal policies.