Source-linked AI summary
Generative models as parsimonious descriptions of sensorimotor loops
Manuel Baltieri, Christopher L. Buckley
TL;DR
The paper questions whether generative models for cognition must accurately represent the world. It proposes parsimonious sensorimotor models that describe action–percept relationships relevant to behaviour, while remaining consistent with generative-model mathematics and control theory.
Problem
Generative models in predictive-processing accounts are commonly treated as accurate representations of the environment, potentially narrowing their use in explaining cognition and behaviour.
Method
The paper advocates action-oriented generative models that parsimoniously describe sensorimotor contingencies and an agent’s perspective rather than veridical world structure.
Results
The paper argues that such models remain mathematically consistent with generative-model definitions and provide a framework for modelling sensorimotor loops through established relationships with optimal control theory.
Takeaways & Limitations
Generative models can represent valuable information, desires, and action–percept relationships for agents without serving as accurate maps of the world.
Abstract
from arXiv · showhide
The Bayesian brain hypothesis, predictive processing and variational free energy minimisation are typically used to describe perceptual processes based on accurate generative models of the world. However, generative models need not be veridical representations of the environment. We suggest that they can (and should) be used to describe sensorimotor relationships relevant for behaviour rather than precise accounts of the world.