Video-Based experiments better unveil societal biases towards ethical decisions of autonomous vehicles

Autonomous vehicles (AVs) encounter moral dilemmas when determining whom to sacrifice in unavoidable crashes. To increase the trustworthiness of AVs, policymakers need to understand public judgment on how AVs should act in such ethically complex situations. Previous studies have evaluated public perception about these ethical matters using picture-based surveys and reported societal biases, i.e., systematic variations in ethical decisions based on the socioeconomic characteristics (e.g., gender) of the individuals involved. For instance, females may prioritise saving a female pedestrian in AV-pedestrian incidents. We investigate if these biases stem from personal beliefs or emerge during experiment engagement and if the presentation format affects bias manifestation. Analysing neural responses in moral experiments measured using electroencephalography (EEG) and behaviour model parameters, we find that video-based scenes better unveil societal biases than picture-based scenes. These biases emerge when the subject interacts with experimental information rather than being solely dictated by initial preferences. The findings support the use of realistic video-based scenes in moral experiments. These insights can inform data collection standards to shape socially acceptable ethical AI policies.