Markov blankets and Bayesian territories

Abstract

Bruineberg et al. argue that one ought not confuse the map (model) for the territory (reality) and delineate a distinction between innocuous Pearl blankets and metaphysically laden Friston blankets. I argue that all we have are models, all knowledge is conditional, and that if there is a Pearl/Friston distinction, it is a matter of the domain of application: latents or observations. This suggests that, if anything, Friston blankets may inherit philosophical significance previously assigned to observations.