Neural Signatures of Moral Evaluation in Autonomous Driving: The Interplay of Agency, Agent Type, and Social Presence.

Engineering notes

Key topics: autonomous driving, autonomous vehicle, perception, control. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

The integration of autonomous systems into society, particularly in domains such as transportation, has intensified debates around machine morality and accountability. Autonomous vehicles (AVs) are increasingly facing ethical dilemmas that mirror classic philosophical problems, such as the trolley problem, raising questions about how these systems should behave and who is responsible for their decisions. Recent research in cognitive science and human-robot interaction suggests that people’s moral judgments are influenced not only by the nature of the dilemma but also by the perceived agency of the entities involved. Dennett (1978) proposed that agency is attributed when actions are seen as driven by internal states rather than external programming. This perception becomes especially relevant when robots are present in decision-making contexts, as studies show that anthropomorphic features and behavioural cues can lead humans to assign moral responsibility to machines (Banks, 2019; Trafton et. al., 2024). The transition toward autonomous vehicles shifts the human role in transportation from active agents to passive monitors, raising critical questions about how this loss of perceived responsibility affects moral evaluation when unavoidable damage situations appear. In sociological theories, the concept of the Panopticon (Foucault, 2023) describes how surveillance of human activities serves as a form of social control and discipline and thus regulates moral decision-making. The novelty in our current situation, with AI agents as social agents, is that these machines could also regulate our behavior. Therefore, our study investigates the neural dynamics of moral decision-making by manipulating agency (observing or making decisions) and social presence (being observed by a humanoid robot or a human being). The EEG analysis (Yun, et. al., 2019; Leuthold, et. al., 2015; Gantman, et. al., 2020) will be focused on N200 (conflict monitoring), N400/ LPP (emotional processing and expectancy violation), P300 (processing, decision closure, and cognitive effort), ErrPs (to be modulated by agreement with moral ratings), Frontal Theta (4-8 Hz, anxiety and monitoring), and Frontal Alpha Asymmetry (8-12 Hz, behavioral inhibition), alongside behavioral measures (reaction time, decision choice) and subjective appropriateness ratings. Moreover, a linear classifier (like Support Vector Machine - SVM) trained with time-frequency power features (Theta and Alpha) and ERP features can support the cognitive moral states classification (e.g., Moral vs. Non-moral).

Links and sources

Comments