Trusting in the competence of humans and artificially intelligent agents varying in generosity

Around the world and with increasing frequency, humans now interact with artificial intelligence (AI) in various domains and collaborative settings. Because of this rapid and growing integration, investigating human-AI interactions can identify effective ways for humans and AI to work together. In two preregistered experiments, we examined how competency (high or low), benevolence reputation (malevolent, neutral, or generous reciprocity), and recent feedback (meeting expectations for generous reciprocity) impact trust towards human and AI partners during iterative Trust Games. In Experiment 1, participants played with human or AI partners; in Experiment 2, participants played with both. Positive behavior from partners (generosity or reciprocity) was consistently rewarded with trust, regardless of context. However, the impact of agent type and competence were context dependent. In a comparative context, people initially trusted humans more, but as they learned about their partner from their behavior (reputation and recent actions (i.e., feedback)), differentiation between humans and AI generally decreased. However, when interacting with only a human or AI, they were treated similarly, and trust depended on the partner's behavior (reputation and feedback). Competency had a more nuanced, context-dependent effect: individuals initially trusted highly competent partners more when interacting with humans or AI, but competency shaped trust towards AI more than towards humans in a comparative context. Overall, these findings indicate that human-human and human-AI trust are context-dependent. Importantly, trust in AI can match trust in humans when AI exhibits benevolence.