The empirical status of predictive coding and active inference.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Theoretical Psychology

Background:

• Predictive processing models, including Predictive Coding (perception) and Active Inference (decision-making), offer broad explanatory frameworks.
• These theories have recently begun empirical evaluation, prompting a review of their current evidential support.
• Interconnectedness of these models highlights their potential for unifying understanding across cognitive functions.

Purpose of the Study:

• To critically review recent empirical studies evaluating Predictive Coding and Active Inference.
• To assess the degree of empirical support for these predictive processing algorithms.
• To identify research gaps and suggest future directions for validating these theories.

Main Methods:

• Systematic review of recent empirical research on Predictive Coding and Active Inference.
• Analysis of studies focusing on empirical evaluation and model fitting to behavioral data.
• Comparison of findings with alternative computational models.

Main Results:

• Empirical evidence provides modest support for Predictive Coding, though alternative feedforward models can also explain some findings.
• Active Inference models often fit behavioral data well, but validation typically involves explaining individual differences rather than direct theory testing.
• Formal comparisons with non-Bayesian or model-free reinforcement learning models are lacking for Active Inference.

Conclusions:

• While Predictive Coding and Active Inference are promising, their empirical adequacy requires further investigation.
• Specific research is needed to rigorously test the validity and explanatory power of these predictive processing algorithms.
• Future studies should prioritize formal model comparisons to establish the unique contributions of these theories.