Researchers at North Carolina State University have developed a new method to study how people make moral decisions while driving, with the aim of using this data to train artificial intelligence in autonomous vehicles. The technique was validated through experiments involving philosophers.
“Very few people set out to cause an accident or hurt other people on the road,” said Veljko Dubljević, corresponding author of the study and professor at NC State. He explained that accidents often arise from minor decisions, such as speeding slightly or not fully stopping at a stop sign. Understanding these decisions is crucial for training AI to make moral choices.
The research uses the Agent Deed Consequence model, which considers three factors in moral judgment: the agent’s intent, the deed itself, and its consequences. Test subjects were presented with traffic scenarios and asked questions about their moral acceptability.
In this validation study, 274 philosophers participated by evaluating driving scenarios and sharing their views on driver decisions’ morality. The researchers also assessed participants’ ethical frameworks.
“Different philosophers subscribe to different schools of thought regarding what constitutes moral decision-making,” Dubljević noted. However, he found it noteworthy that despite differing philosophical backgrounds—utilitarianism versus deontology—their conclusions on driving morality were consistent.
This consistency suggests that the findings can be generalized and used for AI training. “This is a significant step forward,” Dubljević stated. Future steps involve scaling up testing across broader populations and languages to explore generalizability beyond Western culture.
The paper titled “Morality on the road: the ADC model in low-stakes traffic vignettes” has been published in Frontiers in Psychology. Michael Pflanzer, a Ph.D. student at NC State, is the first author, with contributions from Dario Cecchini and Sam Cacace.
This research received support from the National Science Foundation under grant 2043612.
