Emotion prediction as computation over a generative theory of mind

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Social Psychology

Background:

• Human observers can predict emotions from limited event information.
• Understanding emotion prediction is crucial for artificial intelligence and social interaction modeling.

Purpose of the Study:

• To propose a formal computational model for predicting emotions in social dilemmas.
• To investigate the role of inferred beliefs and social preferences in emotion prediction.

Main Methods:

• Utilized inverse planning to infer beliefs and preferences (equity, reputation).
• Computed 'appraisals' by comparing events with expectations and preferences.
• Learned functions mapping appraisals to 20 emotion labels.

Main Results:

• The model accurately predicted human observers' quantitative emotion predictions.
• Inferred social preferences, not just monetary ones, were critical for nearly all emotion predictions.
• Both the model and humans used minimal individualizing information to adjust predictions.

Conclusions:

• The framework integrates inverse planning, appraisals, and emotion concepts.
• It provides a computational approach to understanding intuitive theories of emotion.
• Social preferences play a significant role in predicting emotional responses.