Emotions as computations.

Area of Science:

• Computational neuroscience
• Affective science
• Cognitive psychology

Background:

• Emotions profoundly influence behavior, learning, and perception.
• The precise computational role of emotions remains a subject of debate.
• Existing models lack conceptual precision and neurobiological integration.

Purpose of the Study:

• To provide a computational framework for understanding emotions.
• To propose novel formulations using reinforcement learning.
• To map computational functions to specific emotions and moods.

Main Methods:

• Review of computational accounts of emotion.
• Application of reinforcement learning formalism.
• Integration of neurobiological and psychological evidence.

Main Results:

• Identified three classes of computations implemented by emotions: state evaluation, action evaluation, and prospect evaluation.
• Developed new reinforcement learning formulations for each computation.
• Proposed a one-to-one mapping between computations and specific emotions (e.g., pleasure/pain, happiness/sadness, fear/anxiety).

Conclusions:

• Emotions are integral to evaluating outcomes, learning predictions, guiding actions, and planning under uncertainty.
• The proposed computational framework offers a parsimonious explanation for the diverse roles of emotions.
• This work bridges computational theory with empirical findings in affective science.