Observing the observer (I): meta-bayesian models of learning and decision-making.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Decision Theory

Background:

• Optimal decision-making under uncertainty is a core problem in cognitive science.
• Existing models often struggle to disentangle perceptual and response processes.
• Bayesian approaches offer a powerful framework for modeling belief formation and decision-making.

Purpose of the Study:

• To present a generic meta-Bayesian approach for inferring how subjects make optimal decisions under uncertainty.
• To distinguish between perceptual and response models in decision-making.
• To enable the observation of an individual's (context- or subject-dependent) prior beliefs and utility functions.

Main Methods:

• Developed a theoretical framework distinguishing perceptual and response models.
• Utilized Bayesian decision theory, incorporating prior beliefs, posterior beliefs, and utility (loss) functions.
• Proposed evaluating the likelihood of observed behavior based on updated beliefs and utility functions.

Main Results:

• The meta-Bayesian approach allows for the inference of hidden states, prior beliefs, and utility functions from observed behavior.
• This framework provides a method to "observe the observer" by analyzing psychophysical or neurophysiological measures.
• The approach is theoretically described, with a companion paper detailing its implementation and application to reaction time data.

Conclusions:

• The proposed meta-Bayesian framework offers a unified approach to understanding optimal decision-making under uncertainty.
• It provides a powerful tool for reverse-engineering cognitive processes, including belief formation and goal-directed behavior.
• This methodology has broad applicability in cognitive science, neuroscience, and artificial intelligence.