AI Uses Second-Order Theory of Mind to Detect Human Misbeliefs

A new AI framework enables agents to model and correct human misconceptions about the agent's knowledge using second-order Theory of Mind (ToM-2). Built on the I-POMDP model, the system tracks how a person's erroneous beliefs evolve and identifies cognitive biases and heuristics (CBH) behind them. During interactions, the agent can detect when CBH may influence a person's actions and generate adaptive feedback to address those biases. An in-person user study demonstrated that a ToM-2 learner significantly improved the informativeness of teacher actions compared to baseline methods. Subjective results indicated that participants found the ToM-2 learner's feedback more useful. The research was published on arXiv (ID: 2605.12745) under the Human-Computer Interaction category.