Causal blankets: Theory and algorithmic framework

TL;DR

This paper introduces a new theoretical framework called causal blankets for identifying perception-action loops directly from data, applicable to complex dynamical systems without requiring steady-state or Markovian assumptions.

Contribution

It develops a novel concept of causal blankets based on computational mechanics, enabling the construction of perception-action loops from data in broad settings.

Findings

Every bipartite stochastic process has a causal blanket.

The effectiveness of PALO formulation depends on the integrated information of the bipartition.

The framework does not require steady-state or Markovian dynamics.

Abstract

We introduce a novel framework to identify perception-action loops (PALOs) directly from data based on the principles of computational mechanics. Our approach is based on the notion of causal blanket, which captures sensory and active variables as dynamical sufficient statistics -- i.e. as the "differences that make a difference." Moreover, our theory provides a broadly applicable procedure to construct PALOs that requires neither a steady-state nor Markovian dynamics. Using our theory, we show that every bipartite stochastic process has a causal blanket, but the extent to which this leads to an effective PALO formulation varies depending on the integrated information of the bipartition.