Shaozhe Cheng^1,2 (), Kaelin Main², Tamar Kushnir², Tao Gao¹; ¹UCLA, ²Duke University

Humans can perceive rich mental information, such as intentions and goals, from the mere visual movement of simple shapes. While much research has focused on third-party inferences of mental states, little is known about how people use these inferences to make first-person, agentive decisions—especially in social interaction contexts involving conflicting goals. Here, we explore how spontaneous intention inference influences one’s own decision-making processes. Participants played a computer game where they freely chose between two equally desirable goals while another agent, controlled by a reinforcement learning (RL) model optimizing its own expected utilities, played the same game on the same visual display. Participants were explicitly informed that they could pursue any goal, regardless of the other agent’s actions. We aimed to test whether the mere presence of another agent would prompt participants to infer its intentions and adjust their decisions accordingly. The results revealed that: (a) Both adults and children (ages 5–7) spontaneously used theory-of-mind to infer the intentions of the other agent, as evidenced by their tendency to choose goals opposite to those of the other player, demonstrating "conflict avoidance." (b) This tendency was not universal but depended on the temporal dynamics of intention presentation. Participants avoided conflicting goals when the other agent revealed its intention first or simultaneously but did not yield when they revealed their own intention first. This conditional social avoidance, based on active intention inference, suggests a sense of intention-based ownership. (c) Humans consistently took longer to make decisions compared to the RL agent, highlighting the deliberative nature of human decision-making in social contexts. Together, these findings suggest that visually grounded theory-of-mind significantly impacts individual decision-making in social contexts. Humans spontaneously infer others’ intentions, deliberate more, and tend to avoid conflicting with others’ goals, revealing the dynamic interplay of vision, intention inference, and social coordination.