Inhuman Surprises: Karl Friston’s Theory of (Normal) Life

Neuroscientist Karl Friston

Wired Magazine calls Karl Friston “the genius neuroscientist who might hold the key to true AI.” Friston, a psychiatrist and authority on neuroimaging, has written dozens of papers on his theory of everything related to life. The heart of the theory is the free energy principle, otherwise known as the principle of active inference, the idea being a generalization of Bayes’ Theorem. All organisms, says Friston, strive to maintain the health of their internal order by modeling the unobservable causes of their sensory states, so as to minimize “free energy” or surprise. This is done not just by making predictions and testing representational models, but by active inference, a type of embodied cognition whereby the organism selectively samples the environment and works to make the world less surprising by modifying it, thus providing evidence that the world isn’t so scary after all. The more energy is allowed to roam free, beyond the creature’s control, the more entropy wins out against the creature’s internal order. With this theory, Friston means to explain all aspects of life.

In a co-written paper, called The Markov blankets of life:autonomy, active inference and the free energy principle, Friston incorporates the machine learning concept of a Markov blanket. This “blanket” is that which “defines the boundaries of a system in a statistical sense,” the authors write. The states that make up the blanket can be “partitioned into active and sensory states,” meaning the states that occur spontaneously inside the organism, such as its interpretations or its voluntary bodily movements, and those states impressed upon the organism from the outer world, such as its sensations. Thus, the trick in life is to infer or control the unknown causes of the sensory states, by employing the active states. When this is done poorly, the organism is bound to be surprised by the world which makes for wear and tear, including ill-health and eventually death. We can control circumstances only for so long, of course, before the universe of unknowns nullifies our feeble schemes for holding them back or transforming them.

Here, though, is how Friston and his cowriters lay out some of the ideas:

Active inference, in its simplest formulation, describes the tendency of random dynamical systems to minimize (on average) their free energy, where free energy is an upper bound on (negative) marginal likelihood or evidence (i.e. the probability of finding the system in a particular state, given the system in question). This implies that the kind of self-organization of Markov blankets we consider results in processes that work entirely to optimize evidence, namely self-evidencing dynamics underlying the autonomous organization of life, as we know it. In Bayesian statistics, the evidence is known as ‘model’ evidence, where we can associate the internal states with a model of the external states.

The writers clarify that

any system that minimizes entropy by acting to minimize uncertainty about the hidden causes of its sensations must have a model of the kind of regularities it expects to encounter in its environment. This means that, over (phylogenetic and ontogenetic) time, an organism will become a model of its environment…In other words, it suggests that regularities in the environment of an organism become embodied in the organism—if the organism or species persists. Under the free energy principle, this implies that organisms are close to optimal models of their local surroundings, i.e. their niche. Organisms become close to optimal models by minimizing variational free energy, which bounds the evidence for each phenotype or individual model [25]. This does not imply that an agent must (somehow) construct an internal model (i.e. representation) of its outer environment. It simply means that an agent becomes a statistical model of its niche in the sense of coming to embody statistical regularities of its world in its physical and functional composition.

Applying these biological concepts to the evolution of culture and of people would amount to a Theory of Everything—for Normies. The goal in human life, too, would be to map and to control the unknown, and the complete elimination of surprise would be dystopian. Friston’s theory arises from the pretense of hyperrationality and so evinces the lunacy that’s commonly mistaken for neutral sanity.

Subversive Meta-Cognition

To see how that’s all so, consider the basis for Friston’s talk of optimality. If active inference doesn’t require a representational model of the environment, which is to say a higher-order mind with meaningful internal states, but only internal regularities which statistically are correlated with those transpiring beyond the organism’s boundaries, that mapping is as inherently meaningless as any other real pattern found throughout the lifeless cosmos. Whether the matching of internal and external regularities persists or succumbs to disorder isn’t objectively optimal or tragic. The mere “physical or functional” embodiment of external patterns, as in the evolution of a phenotype with traits adapted to exploit a niche, is as value-neutral as anything else that apparently happens for no reason in nature.

There are, then, only two sources of value that would warrant the ascription of optimality to the match between the regularities. First, of course, the organism can make that determination, assuming some such match is instrumental to the organism’s survival, in which case the optimality is subjective because it depends on the creature’s preference. Second, a theorist such as Friston can posit the optimality according to epistemic criteria for success in explaining phenomena. Even if many creatures have no explicit self-understanding or preferences or any other higher-order thoughts, the theorist who explains their behaviour can deem certain patterns relevant to certain scientific goals, in which case the source of that value-judgment is the appeal of useful theories.

Moreover, even if the probability involved in active inference or Bayesian reasoning is objective, in the case of people, at least, the data must still be understood and interpreted using concepts that always simplify the unobserved causes of sensory states. If the organism does employ representational models, the concepts indeed select relevant or interesting parts of the environment by way of staving off confusion, boredom, or insanity when faced with the fear that the totality of the unobserved cause, that is, the whole of natural reality is incomprehensible and overwhelming. If instead the organism only reacts instinctively to the environment, without thinking about its responses, the genes are nevertheless crucial to nature’s selection of the heuristics and the responses, which makes the organism’s species-centricity even more arbitrary since the genes and the environment do so mindlessly.

What this means is that Friston leaves out the possibility of horrific meta-cognition, of mystical contemplation of the futility and absurdity of normal, instrumental cognition. Suppose, for example, a philosopher ponders the business of being alive in Friston’s sense, and realizes that “surprise” would be a euphemism for “terror” or “awe.” To be surprised by the environment that doesn’t play along with our expectations isn’t just to be disappointed or annoyed, since the disparity entails the gamut of existentialist wisdom about how the world beyond the confines of our mind is alien in its indifference to our struggles. The organic enterprise of converting that absurdity to a manageable form, namely to one that reflects our self-familiarity may be commonplace, but that hardly implies optimality. This is because the normies have to contend with a rival type of cognition, with the type that sees through highfalutin displays of conservatism. You can presume that putting our spin on the environment and going as far as to replace the wilderness with an artificial refuge is optimal or proper, given established standards, but even this relativist judgment is easily defeated, as the environmentalist is quick to point out. The life-centricity of rampant instrumental reasoning would be self-defeating if our attempts to control nature were to result in our destruction, in which case the principle of active inference as formulated by Friston would be precisely as incoherent as parasitic logic (when the parasite kills itself by killing its host).

By contrast, the rival type of cognition takes for granted all creatures’ frailties, the barbarity of natural selection, the grotesqueness of what we call human progress, and the farcicality of our myriad self-deceits. What’s optimal, then, from that meta-viewpoint may only indirectly be the continuation of normal life and our automated pastimes. What’s paramount, though, would be the ecstasy of existential cognition, the courage and sadness involved in attempting to see the world as it is, which means as it would be without us. Just as the herd uses active inference to control the unobserved causes of its sensory states, by way of predicting or producing order and eliminating surprise (along with eliminating the higher-order ecstasy of dread/angst/awe/deranged mirth), the intellectual super-elites, that is, the omega men and women whose kingdom is God’s rather than Caesar’s, to express the point mythically, use the herd of normies, in turn. The herd’s success in fitting into its natural or artificial environment is fodder for comedy perceived only by the alienated watchers who gravitate not to any statistical matching of regularities, but to the opposite, to the abyss between life and lifelessness, to horror and “madness” rather than happiness and compliance.

Instrumental reason entails, then, a hierarchy rather than a duality. The divergence isn’t just between active and sensory states, life and death, roughly speaking. Living things branch off, in turn, between those who side blindly with life against death, without seeing the big picture, and those who stand nowhere, estranged from both nature and our artificial oases. The latter outsiders are torn, for example, between sympathizing with the masses, who suffer because the world’s course isn’t directed to our benefit, and being disgusted by our pretenses. Which is more contemptible, mindless nature which can’t help itself even as it creates trillions of living creatures only to ruin them, or those very creatures whose struggles are seldom heroic, marred as they are by self-deception? Should we side with life or with death? The mark of philosophical mastery seems to be the suffering from the doubt that the answer to that question isn’t obvious.

None of which is to say that Friston’s theory is false, as far as it goes. Exploitative reason is likely hardwired into all life by natural selection. I’ve extrapolated from our preference for artificiality, or from what you could think of as active inference, as the self-interested engagement with the world that ends in the intelligent reengineering of nature, and have speculated that the creation of artificial worlds represents a vindication of ancient animism. We fill nature with purpose and intention, creating the “spirits” (which we now call functions) that we longed to see everywhere to feel finally at home in what had been the haunted wilderness. The dubious leap is to suggest that this instrumental perspective could provide for a theory of everything, that there can be no counterexample or valid contrary way of life.