Moral Judgements on the Actions of Self-driving Cars and Human Drivers in Dilemma Situations from Different Perspectives

Abstract

Self-driving cars have the potential to greatly improve public safety. However, their introduction onto public roads must overcome both ethical and technical challenges. To further understand the ethical issues of introducing self-driving cars, we conducted two moral judgement studies investigating potential differences in the moral norms applied to human drivers and self-driving cars. In the experiments, participants made judgements on a series of dilemma situations involving human drivers or self-driving cars. We manipulated which perspective situations were presented from in order to ascertain the effect of perspective on moral judgements. Two main findings were apparent from the results of the experiments. First, human drivers and self-driving cars were largely judged similarly. However, there was a stronger tendency to prefer self-driving cars to act in ways to minimize harm, compared to human drivers. Second, there was an indication that perspective influences judgements in some situations. Specifically, when considering situations from the perspective of a pedestrian, people preferred actions that would endanger car occupants instead of themselves. However, they did not show such a self-preservation tendency when the alternative was to endanger other pedestrians to save themselves. This effect was more prevalent for judgements on human drivers than self-driving cars. Overall, the results extend and agree with previous research, again contradicting existing ethical guidelines for self-driving car decision making and highlighting the difficulties with adapting public opinion to decision making algorithms.

Hybrid Cognitive-Affective Strategies for AI Safety

“…A similar trend of sparing the life of lawfully behaving patients was reported in the moral machine questionnaire for eastern societies [38]. Finally, concerning the VR experiencer, for instance the specific view (fully immersed or more detached) made a difference in the confidence of moral judgements [255] and moral action varied fundamentally given psychopathic personality traits [250] but also given time pressure [282,399] or the beliefs that the experiencer had about the nature of the agent [434]. Finally, cultural differences might lead to differences in moral judgements within AV dilemmas [38] which should be further considered in future VR studies.…”

“…In order to identify what society wants in an AV context, different studies have used especially VR settings implementing either different ethical dilemmas inspired by classical trolley problems [255,399,434] or conceptualized for more practical scenarios [238,282]. In our view, both types of studies appear valuable due to the fact that while ethical dilemma scenarios have been criticised to be unrealistic because AV scenarios are more complex [322], one could argue that if these simplicistic scenarios can already not be resolved, it seems a rather pessimistic outlook for real-world AV cases implying decisionmaking under uncertainty.…”

“…Other VR studies in the context of decision-making in AVs led to diverging conclusions. For instance, Kallioinen et al [255] analyzed moral judgements in VR from various perspectives (passenger, pedestrian and observer) with human or artificial drivers and came to the conclusion that humans and AVs were largely judged similarly. Thereby, the only important difference was related to the stronger expectation for harm minimization in the AV case.…”

“…When analyzing the results obtained in these studies it becomes apparant that potential conflicts between widely hold ethical conceptions and the state-of-the-art legal frameworks for AVs might arise. An example for this are the German guidelines binding for AVs stating that "personal features, such as age, should not be taken into consideration in unavoidable accident situations" [255]. Against this backdrop, it appears vital to jointly analyze both ethical and legal dimensions within immersive experiences for a more differentiated assessment in order to identify future integrated societal strategies for the deployment of AVs and more generally ethical intelligent systems.…”

“…In Table 7.1, we provide a simplified compilation of the different relevant decision-making focuses that have been revealed in the mentioned predominantly VR studies on traffic dilemmas with some exemplary non-comprehensive set of parameters that made a difference. For instance, the nature of the agent (artificial or human) seemed to matter for some people expecting a different behavior from AVs in comparison to human drivers [434], while for others the expectations were relatively similar [255]. Moreover, the accident liability of the agent deduced from legal settings was relevant in the presented Chinese ethical and legal dilemma study [282].…”