Human Mentalizing as Rational Probabilistic Inference.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Psychology

Background:

• Humans infer mental states (mentalizing) from actions.
• Computational models assume rational strategy selection for behavior.
• Limited understanding exists on how uncertainty is handled in mentalizing.

Purpose of the Study:

• To theorize how Bayesian inference accounts for uncertainty in mental state estimation.
• To investigate how humans adapt inferences to new environments for predicting future behavior.
• To explore how multiple observations mitigate uncertainty in mentalizing.

Main Methods:

• Utilized Bayesian inference to model mentalizing under uncertainty.
• Conducted three experiments: joint inference, future behavior prediction, and uncertainty mitigation via multiple observations.
• Incorporated flexibility tests to validate computational rationality.

Main Results:

• Developed a probabilistic model of mental state inference incorporating uncertainty.
• Demonstrated human ability to predict future behavior based on observed actions and adapt to new environments.
• Showed that integrating multiple observations reduces inferential uncertainty.

Conclusions:

• The study provides a computational account of mentalizing that explicitly incorporates uncertainty.
• Findings support the rational agent assumption in mentalizing models.
• The research advances understanding of how uncertainty influences social cognition and prediction.