Event-related potentials reveal incongruent behavior of autonomous vehicles in the moral machine dilemma

Bertheau, Maren A. K.; Boetzel, Cindy; Herrmann, Christoph S. · 2025 · DOAJ

DOI: 10.1038/s41598-025-99531-6

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study investigates the neural correlates of human acceptance regarding autonomous vehicle (AV) decisions in morally ambiguous traffic situations. Motivated by the need for AVs to demonstrate comprehensible and acceptable behavior to facilitate widespread adoption, the authors sought to identify physiological markers that could unobtrusively assess user perceptions of AI decision-making. Specifically, the research addresses a gap in existing literature by examining event-related potentials (ERPs) when participants observe an artificial agent’s moral choices, rather than making decisions themselves. The study utilizes the Moral Machine dilemma framework, which presents unavoidable accident scenarios requiring trade-offs between different groups of characters. The experimental design involved 34 participants who viewed static image vignettes depicting moral dilemmas. For each scenario, participants first indicated their preferred outcome. Subsequently, they were shown the decision of a putative AI, which was either congruent with or incongruent to their own choice. Researchers recorded electroencephalography (EEG) data to analyze ERP components, focusing on the P3 (322–422 ms) and late positive potential (LPP) (500–900 ms), which are associated with stimulus significance and higher-order cognitive processing. Behavioral acceptability ratings were also collected to validate the experimental manipulation. Behavioral results largely replicated findings from the original Moral Machine study: participants significantly favored sparing pedestrians over passengers, saving more characters over fewer, and prioritizing humans over pets. However, unlike the original large-scale study, the preference for inaction (keeping the AV on its lane) was not statistically significant, likely due to the smaller sample size. Crucially, participants rated AI decisions as significantly more acceptable when they were congruent with their own choices. Neurophysiologically, incongruent trials elicited significantly increased amplitudes in both the P3 and LPP components compared to congruent trials. The P3 effect was localized primarily in fronto-central regions, suggesting the AI’s incongruent decisions were perceived as surprising. The LPP effect, indicative of processing stimulus significance, was distributed across broader scalp regions, including parietal sites, and reached its maximum magnitude during the 500–900 ms window. The findings suggest that P3 and LPP amplitudes serve as reliable neural indicators of user dissatisfaction or perceived incongruence in human-AI interactions involving moral decision-making. The authors conclude that these ERP components could be utilized in brain-computer interface research to dynamically adapt AV behavior or identify critical, unacceptable situations in real-time. This approach offers a potential pathway for designing inclusive autonomous systems that respond to user preferences without requiring obtrusive active input, such as questionnaires, thereby enhancing the transparency and acceptability of autonomous driving technologies.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-25; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• ehmi external hmi