Regularization, Action, and Attractors in the Dynamical “Bayesian” Brain

The idea that the brain is a probabilistic (Bayesian) inference machine, continuously trying to figure out the hidden causes of its inputs, has become very influential in cognitive (neuro)science over recent decades. Here, I present a relatively straightforward generalization of this idea: The primary computational task that the brain is faced with is to track the probabilistic structure of observations themselves, without recourse to hidden states. Tracking this structure in the face of noise requires regularization, and prior experience is the best source of such regularization. Regularization and, by extension, prior expectations can be thought of as abstract “pulling” forces in the space of observations. The same is true for behavioral goals: Organisms strive toward (observing) goal states, so these states similarly exercise an attractive force. Prior expectations, regularization, and action induction can thus fruitfully be seen as attractors in the dynamical system constituted by the brain. This perspective refines thought within the “Bayesian brain” framework, avoids some previous counterintuitive conclusions, and may inspire new empirical and theoretical work by alerting researchers to parallels they were hitherto unaware of.