Guidance Mechanism of Social Prediction Errors in Learning

Determine the computational structure and target of social prediction errors within cerebellar-based models of social cognition, specifying the direction of change they induce and how such errors guide learning updates to internal forward models that anticipate other agents’ behavior.

Background

The paper evaluates whether principles from the Marr-Albus-Ito framework and forward models in motor control extend to cognition, including social cognition. In motor domains, prediction errors have a clear vector that guides learning through gradient descent. When mapping this logic to social contexts, the authors note that the definition of a social prediction error lacks a clear structure and goal, raising uncertainty about how such errors would drive learning.

This unresolved issue is central to establishing whether and how a cerebellar forward model can operate in social cognition. Clarifying the representation (e.g., vector, target) and update rules for social prediction errors is critical to formulating a mechanistic, cerebellar-dependent account of learning about others’ mental states and behaviors.