Selective observation following betrayal shapes the social inference landscape

Despite limited access to others' actions and outcomes, humans excel at inferring hidden intentions. Given only partial access, how do they decide what to observe, and how does selective observation shape inference? Here, we examined how choosing what to observe can bias the inference about others' intentions. Participants played a game where they pursued a fleeing target while a computerized opponent acted competitively or cooperatively. Participants overestimated the opponent's competitiveness after the opponent acted more competitively than expected, whereas no such bias occurred when the opponent was more cooperative than expected. This asymmetry depended on the sequence of events, resembling hysteresis, a form of path dependence observed in physical systems. We found that these biases became stronger when participants chose to observe the opponent instead of their own avatar, and this choice came at the cost of losing precise control over their avatar. Our findings highlight the trade-off in selecting what to observe, as the resulting inference biases propagate differently depending on the interaction history. We often think that everything necessary for understanding others is already visible. However, in reality, we see only a small part of what others do. Focusing on one specific cue while ignoring others—like a gesture versus facial expression— can lead to misunderstandings about their intentions. How do we decide what to observe and how does that decision shape our understanding of others' intentions? These questions have been a long-standing challenge in social inference and, more broadly, theory of mind. To address this, we question the common assumption that all social cues are freely accessible. Instead, we propose that observing others is limited by available resources and, thus, comes with a cost— focusing on one cue means paying less attention to others. This idea of partial observability helps explain how a specific sequence of negative past experiences can shape lasting biases in inferring others' intentions. Our study builds on previous models by relaxing the assumption that we see everything others do, and offers a more realistic view that people understand others based on what they choose to observe.