Making Moral Decisions With Artificial Agents As Advisors. An fNIRS Study

Abstract

Artificial Intelligence (AI) is on the verge of impacting every domain of our life. It is now being increasingly used as an advisor to help make (moral) decisions. The present study aimed at investigating the influence of moral arguments provided by AI-advisors (i.e., decision aid tool) on human moral decision-making and the associated neural correlates. Participants were presented with utilitarian and deontological sacrificial moral dilemmas and had to make moral decisions either by themselves (i.e., baseline run) or with AI-advisors that provided either utilitarian or deontological advice (i.e., AI-advised run), while their brain activity was measured using an fNIRS device. Overall, AI-advisors significantly influenced participants, who often modified their decisions according to AI-advisors arguments. Longer response times and a decrease in right dorsolateral prefrontal cortex activity were observed in response to deontological arguments than to utilitarian arguments. Besides triggering a conflict in participants, being provided with deontological arguments by machine appears to have led to a decreased appraisal of the affective response to the dilemmas. This resulted in a reduced level of utilitarianism, supposedly in an attempt to avoid behaving more like a machine than the machines themselves. Taken together, these results contradict Greene s moral dual-process theory and suggest that motivational power can led to a voluntary up- and down- regulation of affective reactions.