Virtuous Vehicles: Identifying the Values Profiles of Human Drivers as a Basis for Programming Virtuous Decision-Making in Self-driving Cars

Is AI capable of virtue? We propose that autonomous systems can, at least, be programmed to respond in ways that align with the behaviors of virtuous humans. Our test case is programming autonomous vehicles (AVs). Regardless of AVs’ competence, crashes are inevitable. However, it is impossible to program rules for failing gracefully in every crash scenario; nor is the ideal crash outcome always clear. We contend that AV crash decisions must be flexible, reflecting human drivers’ diverse values, driving styles, and crash responses. To that end, we aim to understand both how and why human drivers behave the way they do in crash scenarios. We conducted three studies using surveys and driving simulations and found that the core value of benevolence is associated with a prosocial driving style and self-sacrifice. In contrast, the value of power is associated with a competitive driving style and self-preservation. There may be roles for both benevolent and power AVs. We are currently conducting proof-of-concept studies on programming small-scale robot AVs with driving behaviors that reflect these human values.