Multidimensional Preferences for Regulating Self-Driving Cars. Evidence from a Conjoint Experiment conducted in the US, Japan, and Germany

Effective regulation of self-driving cars requires broad public support. Although policy-makers and practitioners agree upon the growing need to regulate the development of self-driving cars and the importance of regulation that is consistent with citizens' moral believes and societies' legal standards, there is little systematic evidence about which type of regulation citizens prefer and whether the public is sensitive to specific features of possible regulation regimes. In a conjoint experiment, we asked respondents to compare two multidimensional hypothetical regimes regulating self-driving cars and to decide which regime they prefer. The regime profiles varied with respect to three substantive dimensions: (1) Safety (Regulatory body for self-driving cars and safety standards compared to conventional cars), (2) legal framework (liability for accidents caused by the autopilot and ethical prioritization) and (3) autonomy vs. human agency (data protection and supervision of autopilot by the driver). The pre-registered conjoint experiment has been conducted on representative online samples for the USA (N=1,188), Japan (N=1,135), and Germany (N=1,174). Besides the automotive industry being a major industry in those nations, the country selection also reflects anticipated cultural differences regarding the subjective evaluation of AI and autonomous vehicles. However, across all samples, we find that citizens strongly prefer regulation that requires permanent human supervision of self-driving cars and stricter safety standards. Cross-country differences emerge on the safety dimension, as respondents from Japan and Germany prefer public authorities overseeing the approval of self-driving cars, while American respondents show the strongest preference for an independent expert body. Furthermore, in-depth sub-group analysis reveals that preferences towards self-driving cars' regulation are weakly affected by respondents' attitudes towards technology (technophobia), while partisan orientation has only a minor effect on regulatory preferences.