Autopoiesis and the Markov blanket

Friston and Ao (2012) asked what it means for an agent's states to be distinct from those of its environment. Their answer was that internal and external states are conditionally independent and are separated by a Markov blanket.¹ Friston, Levin, Sengupta and Pezzulo (2015) described the term as one that began in work on Bayesian networks. They gave the cell surface as an example of a blanket between intracellular and extracellular states.⁴ They added that the external states can only be seen indirectly by the internal states, through the blanket. They also split the blanket into sensory and active states.⁴

Friston (2013) offered a heuristic proof, with simulations of a primordial soup. It argued that life, understood as biological self-organization, is an emergent property of any ergodic random dynamical system that has a Markov blanket.² Internal states then appear to minimize free energy, the quantity optimized in Bayesian inference. On this account they appear to model and act on their world, which gives homoeostasis and a simple form of autopoiesis.² In the simulations, lesioning internal states or blanket states produced decay and dispersion. The intact system kept an ordered arrangement.³

Allen and Friston (2018) tied this to the older autopoietic idea. They wrote that a cell persists by creating and maintaining a boundary, and they cited Varela and colleagues (1974) for the point.⁵ In their account the existence of a system depends on conserving its blanket. Otherwise the system would dissipate into its environment.⁵ They described the free energy principle as recasting this as a self-fulfilling prophecy, in which the organism embodies a belief that it will prevail.⁵ Ramstead, Badcock and Friston (2018) extended the argument to nested systems, such as cells within organs, which they described as Markov blankets of Markov blankets.⁶

Rorot (2022) noted continuing tension between computationalist and anti-computationalist theories of mind. Villalobos and Dewhurst and Korbak, he said, had proposed a middle view called computational enactivism, in which autopoietic requirements are compatible with, or depend on, computation in living organisms.⁷ Fábregas-Tejeda and Sims (2025) grouped defenders of autopoiesis, enactivism and the free energy principle as panextensionalists, who hold a strong life-mind continuity thesis. They contrasted these with oligoextensionalists, who restrict cognition to a few clades.⁸