NEUROCOGNITIVE CORRELATES OF TRUST IN ARTIFICIAL AGENTS WITH DIFFERENT MORAL REASONING STYLES

Despite the increasing presence of artificial systems in people’s everyday lives, trust in AI remains a contested topic. Previous studies suggest that people trust artificial agents more when they are able to justify their actions using contractualist reasoning – a type of moral framework focused on agreements with humans. We investigated the neurocognitive processes underlying the evaluation of agents and how this process influences decisions about whether to trust artificial agents as a function of the type of moral reasoning artificial agents use to justify their actions. We found greater neural activity in brain areas related to mental state attribution, such as the temporoparietal junction (TPJ), when people judged whether to trust contractualist agents compared to agents that justified their actions based on utility or following strict rules. In contrast, areas typically involved in value-based computations were more active when judging utilitarian agents compared to agents that offered justifications related to agreements or rules. These neural activity patterns were stronger for human agents than artificial agents that offered justifications, and were influenced by previous physical interaction with a social robot. Our findings reveal distinct neural mechanisms underlying decisions on whether to trust artificial agents as a function of the moral framework they use to justify their actions, suggesting that humans engage comparable neurocognitive pathways when judging human and robot agents.