AI Insight
This study tests a key prediction of Friston's free energy principle by examining whether self-replicating spots in a Gray-Scott reaction-diffusion model possess Markov blankets that statistically separate their interior from exterior states. The researchers found that flux-defined boundaries fail to screen interior from exterior states and may actually increase dependence when conditioned upon, challenging the principle's claim that such blankets naturally arise in short-range coupled systems. While a band of surrounding field roughly one organism radius thick does provide some screening, this appears to be a trivial consequence of short-range interactions rather than evidence of Bayesian inference.
Why it matters
The free energy principle has been proposed as a unifying framework for understanding biological self-organization, cognition, and life itself. This computational analysis provides empirical evidence questioning whether the principle's mathematical conditions hold in a concrete physical system, which has implications for its applicability to real biological organisms and cognitive systems.
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Abstract: Friston’s free energy principle holds that a system separated from its environment by a Markov blanket (a boundary whose states statistically screen its interior from its exterior) can be treated as performing approximate Bayesian inference, and that such blankets arise wherever coupling is short-range. We test this blanket condition in a concrete nonlinear pattern-forming system: self-replicating spots in a reaction-diffusion model that persist, compete, reproduce, and die. Directed boundary exchange and timescale separation hold up reasonably well. A flux-defined blanket does not screen interior from exterior. Flux removes only part of the linear dependence, and a nonlinear test shows that conditioning on flux increases the dependence. We argue that this is because flux is a statistical collider. Conditioning on the boundary’s own concentration state screens about a fifth of the nonlinear dependence. The dependence is concentrated within about one organism radius of the boundary. A band of the surrounding field of that thickness does screen the interior from the exterior beyond it, as any short-range field would; this is not evidence of inference.