Should AI be allowed to break the law? Public acceptance and tolerance of discretionary actions performed by autonomous vehicles in response to situations varying by degree of acuteness
通过三个实验探究公众对自动驾驶汽车在紧急情况下采取合法或违法酌情行为的态度，发现自动驾驶汽车比人类司机因相同违法行为受到更多指责和更少信任，但合法行为下的差异取决于情况紧急程度。
Many countries are on the verge of introducing highly autonomous vehicles (AVs) into current traffic networks dominated by human-driven vehicles. The complex and dynamic nature of road traffic situations requires AVs to exhibit human-like, discretionary behaviours that may fall outside the scope of formal Rules of the Road (e.g., straddling solid lines to let an emergency-services vehicle pass). It is important to understand public attitudes towards these behaviours especially when they may lead to negative outcomes. The current paper presents three experiments in which participants’ judgements of blame and trust were probed after being presented with traffic scenarios where AVs or human drivers chose to perform (Experiment 1 & 2) or not perform (Experiment 3) legal or illegal discretionary actions (DAs) to avoid a danger or optimize traffic flow, with various consequences. The results reveal that AVs were blamed more and trusted less than human drivers for performing the same illegal DAs. But with legal DAs, this difference was contingent on the acuteness of traffic situations, hinting toward a shift of judgemental focus from the justifiability of an action to the quality of their execution. Additionally, witnessing AVs performing (or not performing) DAs could potentially improve or worsen their general acceptance depending on the outcomes of the DAs. Our findings paint a promising picture of allowing adaptive behaviours of AVs yet highlight the need to establish formal protocols for designing, regulating, and appraising DAs of AVs as well as the necessity of improving the transparency of their decision-making processes for users.