There are many theories in philosophy and cognitive science that I don’t understand. There are some theories that I have tried to understand but couldn’t. The book “A Drive to Survive: the Free Energy Principle and the Meaning of Life” by Kathryn Nave covers two theories that I was unable to comprehend: enactivism and Karl Friston’s active-inference-predictive-coding-free-energy-principle fusion. Now, I comprehend them a little bit better, and good enough to make me even more skeptical than I was before. Let me briefly describe both theories. 

Enactivism

The origin of enactivism can be attributed to the 1991 book "The Embodied Mind" by Varela, Thompson, and Rosch. In this book, they proposed that cognition (human, animal, etc.) arises from the interaction of the agent with its environment. Therefore, cognition can only be coupled with action, and cognition has no underlying representational essence. It is just what we do. Later, enactivism as a theory of cognition morphed into enactivism as the theory of life. Living things, according to enactivists, are autonomous and sense-making. They are autonomous because they can engage with the environment. They are sense-making because they create meaning from these interactions. To take it even further, bioenactivists (like enactivists but with addition) agree with autonomy and meaning of living creatures but additionally root for mind-life continuity and life as self-production view. Mind-life continuity implies that the mind is not an evolutionary addition to life but rather an intrinsic aspect of life from its very beginning. Life as self-production refers to the idea that living things are self-creating and self-maintaining; that they regenerate their own components to maintain integrity and identity. 

Active inference/predictive coding/free energy principle

These three concepts are all parts of Karl Friston’s theory, which is (as some people, including Friston, claim) the theory of life. The active inference part assumes that agent can act in the environment and change its own sensory input. For instance, a squirrel can move around the tree to see what it looks like from behind. The agent does it by interacting with the environment not blindly or randomly but with a predictive model in ‘mind’. Perhaps the squirrel thinks that if the tree is round, the behind part is similar to the front part. Generating predictive models is a statistical task that involves aggregating information from the environments and tuning an internal set of simple recognition models to fit the data. Hence, if the squirrel finds out that the tree is round, its predictions about other trees being round will be strengthened.

Recognition models have parameters that can be tuned using gradient decent over two properties: empirical adequacy and complexity. Empirical adequacy estimates how likely the model is given the data. Complexity measures the extent of changes in the parameters of the recognition model, that are needed to fit the data. When empirical adequacy is maximized and complexity is minimized, free energy is reduced. The free energy principle (FEP) thus postulates that the agent will survive if it will remain within states that it had found themselves before, rather than in new states. To sum up, being autonomous involves exercising active inference via predictive coding, which presupposes prediction error minimization that comes in the form of a decrease in free energy. 

Both bioenactivists and Karl Friston dare to try to explain life. However, the author disagrees with Karl Friston. She points out that for many organisms, their survival doesn’t depend so much on the stability of their parts but often on material turnover and fundamental shifts in organization. What makes an organism, according to the author, is the continuous relations within a sequence of evolving process-cycles. In the last chapter, she proposes another theory of life. 

Theory of life as hierarchy of constraint 

Constraints are factors that shape outcomes by limiting possibilities. A typical example of a constraint is a catalyst substance in chemical reaction; it accelerates the reaction without changing itself. Every component of living organism acts both as a constraint on its metabolic network and as a product of that network, thus making a living organism a hierarchy of constraints. This hierarchy of constraints operates at different levels and timescales, some of which set the boundary conditions for the closure of other processes. For an organism to maintain its integrity, these constraints must be continuously active, as they enable the regenerative processes that prevent the degradation of the organism's structure. 

It was an extremely difficult book to read and to comprehend. I was familiar with enactivism on the one hand and Karl Friston’s theory on the other, but I didn’t know that they were somehow related. The author could have spent more time on the premises, revealing why she discusses enactivism and active inference/predictive coding/free energy principle together in one book. Additionally, the proposed theory of life as a hierarchy of constraints doesn’t follow (for me) from the long discussion of inconveniences of Karl Friston’s theory. Similarly, the proposed theory seems to be somehow a version of bioenactivism, but it wasn’t clear for me how they are related. Lastly, I was not and I am not able to intuitively grasp neither bioenactivism nor active-inference-predictive-coding-free-energy-principle fusion nor proposed theory. If life is a hierarchy of constraints, how did life emerge or how did constraints emerge? How was hierarchy formed, and why is it a hierarchy in the first place? If we would want to artificially create life, how would we set up constraints? 

I would suggest reading the book if you find yourself having advanced knowledge at least in one of the theories of life or if you don’t like active inference and want to have arguments against it (there are plenty).

Favourite quote

“What lies beneath the FEP's twisted thicket of mathematical terminology is just a formulation of survival as stability in the face of perturbation and the insight that such stability can be formally redescribed in inferential terms”

August, 2024