Karl J. Friston

In the 2012 paper with Ao, free energy is defined as a functional of sensory and internal states that upper bounds surprise and can be reduced by action¹. The authors describe it as a way to turn an intractable integration problem into an optimisation problem. The paper also simulates order and its loss. Sixteen coupled Lorenz oscillators settle onto a low-dimensional synchronisation manifold, because a mean field draws each oscillator towards the average². The same paper notes that such order is easy to destroy.

The 2015 paper with Levin, Sengupta and Pezzulo proposes that each cell carries an implicit model of its place in the final morphology³. Self-assembly is then movement to sample the local signals that the model predicts. The authors treat this as an optimality principle, the minimisation of variational free energy. A later paper by Ramstead, Constant, Badcock and Friston redraws the brain's partition of internal, sensory and active states as the interior, surface and cytoskeleton of a cell⁴.

Pio-Lopez and colleagues describe the 2015 work as simulating morphogenesis and dysmorphogenesis by changing extracellular signals while leaving the generative model fixed⁶. Levin cites it for the finding that embryos reach the same goal region of morphospace by new paths⁷. The 2019 paper also reports a critique. Bruineberg and Hesp, and Kirmayer, asked whether the Markov blanket formalism can individuate systems at scales beyond one organism, such as societies⁵.