Rethinking Social Cognition Beyond Mindreading

Social cognition is often centered on empathy—the capacity to feel or understand others’ emotions. But in many goal-driven interactions—like negotiation, interviews, or sales—empathy offers limited utility. Even if you pick up fragments of the other person’s internal state, that information is often partial, ambiguous, or unreliable—and rarely enough to influence the outcome.

ToM assumes we can access others’ mental states, but in reality, these states are unobservable and unstable. People often don’t know what they believe, change their minds without noticing, or deliberately hide their motives. This makes social inference less about accurate understanding, and more about inference based on memory and loosely related information.

To build a more realistic model of social cognition, we must shift from internal simulation to external structure: not “what is she thinking?”, but “who adjusts to whom?”, “who can afford to act?”, and “who holds leverage?”.

Theory of Mind (ToM)—the capacity to infer others’ mental states—has long been treated as a cornerstone of social cognition. Yet its conceptual boundaries and mechanistic foundations remain vague, and its empirical robustness is less secure than often assumed.

This tension is especially striking because ToM originated in comparative work on chimpanzees, even though chimpanzee social life is best known for power-related behavior. The same problem appears in humans: lie detection—arguably the most iconic real-world false-belief task—reveals an accuracy of only about 54\%, barely above chance. Across humans and other social species, mental-state inference is inconsistent, context-dependent, and error-prone, indicating that “mindreading” is better understood as pattern-based social inference rather than a dedicated cognitive module. By contrast, status recognition and power-relevant behavior are robust, contextually stable, and consistently observed across species.

We therefore propose reframing social cognition as the tracking of status and the navigation of power dynamics rather than hidden mental-state inference. This reframing provides an empirically grounded account of social cognition that better reflects social interaction in real-world settings.

For decades, we've built our theories of social intelligence on the idea of "mindreading"—that humans (and maybe machines) infer others’ beliefs to cooperate.

But the evidence doesn’t add up: animals fail ToM tests, humans misread each other all the time, and power dynamics often predicts behavior better than belief inference.

What if social cognition isn't about beliefs at all—but about power? Not internal beliefs, but external structure—who leads, who yields, who shapes the outcome.

Rethinking ToM around power shifts our focus from imagined minds to real, observable social dynamics. It changes how we study cognition, how we model agents, and how we explain the fabric of social life.