Human decision-making biases in the moral dilemmas of autonomous vehicles

Frank, Darius‐Aurel; Chrysochou, Polymeros; Mitkidis, Panagiotis; Ariely, Dan · 2019 · OpenAlex-citations

DOI: 10.1038/s41598-019-49411-7

archive: archived pipeline: cataloged verified

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

Summary

This paper investigates how human decision-making biases affect moral judgments in the context of autonomous vehicle (AV) dilemmas, addressing the challenge of programming a universal moral code for AI. The authors argue that relying on human preferences to train AVs risks incorporating inherent biases, such as those driven by cognitive load and emotional engagement. Specifically, the research examines whether personal perspective (passenger vs. pedestrian) and decision-making mode (intuitive vs. deliberate) systematically alter moral choices, potentially leading to inconsistent or self-serving ethical guidelines. The study comprises seven experiments involving over 12,000 participants from the US and Denmark, replicating scenarios from the Moral Machine experiment. In Study 1, participants were assigned to different perspectives (passenger, pedestrian, or observer) and decision-making modes, manipulated via time pressure (five seconds for intuitive, 30 seconds for deliberate). Subsequent studies varied situational factors, including the number of victims, the presence of children, and the lawfulness of pedestrian behavior, to test for action bias, status quo bias, and underlying moral doctrines (utilitarian vs. deontological). The results demonstrate that moral decisions are significantly biased by both perspective and decision-making mode. Under intuitive conditions, participants favored deontological decisions, predominantly choosing to sacrifice the passenger rather than the pedestrian. In contrast, deliberate decision-making shifted preferences toward utilitarian outcomes, with participants more likely to sacrifice the pedestrian to minimize total casualties. Personal perspective also induced self-preservation biases; participants viewing the scenario from the pedestrian’s perspective were less likely to sacrifice the pedestrian, while those in the passenger perspective favored sparing the passenger. These effects persisted across various situational manipulations, and the authors found no evidence for action or status quo biases as primary drivers. Additionally, female and older participants were significantly less likely to sacrifice pedestrians. The findings imply that human moral judgments are not stable or universal but are heavily influenced by cognitive processing modes and self-interest. This variability poses a significant social challenge for designing a consistent moral code for autonomous vehicles, as programming machines based on aggregated human data may embed conflicting biases. The authors conclude that future research must account for these psychological constraints to develop ethical AI systems that align with societal expectations rather than individual predispositions.

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 2026-06-26; verification: verified.

Topics

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

• decision making risk perception