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

Abstract We investigated event-related potentials (ERPs) in the context of autonomous vehicles (AVs)—specifically in ambiguous, morally challenging traffic situations. In our study, participants (n = 34) observed a putative artificial intelligence (AI) making decisions in a dilemma situation involving an AV, expanding on the Moral Machine (MM) experiment. Additionally to the original MM experiment, we incorporated electroencephalography recordings. We were able to replicate most of the behavioral findings of the original MM: In case of an unavoidable traffic accident, participants consistently favored sparing pedestrians over passengers, more characters over fewer characters, and humans over pets. Beyond that, in the ERP we observed an increased P3 (322–422 ms), and late positive potential (LPP) (500–900 MS) amplitude in fronto-central regions when the putative AI’s decision on a moral dilemma was incongruent to the participants’ decision. As P3, and LPP are associated with the processing of stimulus significance, our findings suggest that these ERP components could potentially be used to identify critical, or unacceptable situations during human-AI interactions involving moral decision-making. This might be useful in brain computer interfaces research when, classifying single-trial ERP components, to dynamically adopt an AV’s behavior.