Predictable Identities: 22 – The Entropic Brain

Brain researcher Robert Carhart-Harris and physicist Karl Friston suggest that psychedelics will save humanity. They are, of course, far from the first people to do so. But they’re the first to explain how psychedelics will save humanity using the theory of predictive processing, which was developed in large part by Friston himself. And so with my interest sufficiently piqued, allow me to condense a decade of research on the Entropic Brain Hypothesis into a few paragraphs.

Our brains evolved to model the environment and minimize surprise and uncertainty. Since our environment is complex and dynamic, so are more evolved brains. fMRI allows us to measure brain entropy, how unpredictable is one’s brain state in the future based on its current state. It’s a proxy for a brain’s flexibility and complexity. 

Human brains are more entropic than those of our animal relatives, which in turn have more entropic brains than phylogenetically distant species. But humans have also developed a brain structure that suppresses entropy: the default mode network (DMN). According to Friston, optimal prediction is achieved when the brain is finely poised between order and disorder, rigidity and entropy. The DMN is less developed in children, and is suppressed in REM sleep, the onset of psychosis, and by psychedelics — all the above states are characterized by wandering thoughts, creativity, and magical thinking (hello, Ribbonfarm!) that isn’t strongly constrained by reason and prior experiences.

The DMN is also overactive in people with depression. This manifests in two features of depression: “depressive realism” (a capacity to judge reality more accurately) and rigidity of thought (the mind is stuck in a negative bias that doesn’t respond to changes in the environment). In fact, many other mental disorders can be thought of as disorders of mental rigidity. For example: addiction (the brain is stuck in loops of craving and indulgence), autism, PTSD, and schizophrenia. Psychedelics increase brain entropy and “shake it out” of its rigid habits, allowing it to settle into more salubrious patterns (especially if guided by a good therapist).

There’s another effect that psychedelics cause by inhibiting the DMN — the dissolution of ego and the sense of self. Should we think of self-identity too as a disorder of mental rigidity? Stay tuned.

Series Navigation<< Predictable Identities: 21 – EnlightenmentPredictable Identities: 23 – The Self >>

Get Ribbonfarm in your inbox

Get new post updates by email

New post updates are sent out once a week

About Jacob Falkovich

Jacob is so proud of his blog,, that it's on his online dating profiles. He also tweets @yashkaf.


  1. Interesting, until recently, (as Carhart-Harris and co mention briefly in that paper) the default mode network itself was characterised by “wandering thought”, though primarily by being on the outside of the task oriented thought generally being studied. It could well be that we will increasingly burrow our functional models into the apparently meaningless patterns of activity of the brain, slowly peeling them away from that structure of “everything else”.

    But in the meantime, I am cautious about associating the action of the default mode network in its entirety with entropy reduction, because that seems to be a little like saying that the purpose of central rooms in a house is to be warm; they are warm, or tend to be, because they are less immediately connected to external sinks for heat, but that is caused by their functional organisation rather than the purpose of it.

    In the same way, I would personally expect in daily life (if the brain does have endogenous processes that ubiquitously try to reduce entropy distributed through its structure), that externally coupled task oriented cognitive processes would act at a higher average entropy than more internal processes, simply through the network structure (that sounds arrogantly overselling my knowledge of this topic, but “topology” sounds worse, adjacency maybe?), indeed, some very old approaches to cognition attempted to track connectivity through the brain via a combination of correlation on efferent and afferent connections, and with increased entropy. (Correlated behaviour corresponding to successful control processes, and higher entropy to the corrections required to maintain them)

    That they say that it makes sense that the brain should be “more ordered than that which it seeks to control”, without any reference to the old theories about transducer entropy seems striking to me; it should actually be surprising and notable that the brain manages to compress environmental entropy effectively such that its control process can still act on it from a lower entropy position, with the base assumption being that it should directly reflect the entropy of that which it is coupled to, or more specifically, those elements of the brain engaged in immediate control processes that require continuous circular information flows should have an entropy equal to those subset of environment variations it is seeking to compensate for.

    Also not super convinced about this paper’s assertions that maximising entropy corresponds to formlessness, since for any given temperature, a highly ordered crystal can correspond to the maximum entropy state (a natural physical assumption we rely on daily), even if it has that entropy inherently because of how it conceals from external investigation the highly complex vibrational states of its interior, or equivalently, map them to each other in processes that sustain the overall recognisable structure unchanged; it has entropy not through being disordered, but by being plural.

    It’s not a great look to be basing your model on a concept of entropy, and jumping from a very light touch examination of the theory of criticality to 19th century thermodynamics without anything in between, especially given that those high entropy states that develop by creating highly complex but externally decoupled internal dynamics should I would think be a primary touch point in considering a “high entropy” brain. We can see in many cases that a crystal has higher entropy than a generalised gloop by the fact that given a chance, all kinds of subcritical fluids will want crystalise. Leaving honey around too long should be a primary physical mnemonic for this process.

    But back to the idea of externally coupled processes having higher entropy, I also would expect a contrasting situation to occur occasionally, where people are in a very relaxed and creative mode, where we might find that their brain does generate more internal entropy, in the process of some internal musing, than is present in their environment, not because their brain is actually more noisy, but because following highly self-coupled non-linear processes can often appear to have a higher entropy if you as an observer aren’t sure how to encode it properly (the question of the entropy of an unknown computer process etc. or in fact the question of characterising environmental entropies earlier).

    Looking at increased default mode network activation primarily in the context of depression, and using the obvious behavioural and cognitive fixity of the depressed as a proxy for the behaviour of the default mode network, strikes me as an investigative and conceptual approach dangerously likely to assume a certain set of conclusions by it’s choice of starting place, considering that people without depression can also exhibit increased default mode network activation, and it seems a little like confusing the CPU with the OS; if the default mode network and other deep brain structures correspond to certain non-task and “personality” or “ego” oriented functions, then connecting the function of such systems in a fundamental way with certain personality specific cognitive behaviours strikes me as like saying the problem with a virus strewn computer is that the processor is running too hot, it will be, but we can’t jump from that to assume that the causal relationship should flow in that direction.

    On the other hand, I do think the general organising principle of a sub-critical and critical brain seems plausible, especially if as part of its control processes, the brain shifts over thresholds from highly organised and ordered thought to disordered and transformative thought. I think it would also be interesting to track down those times I suspect exist, where elements of the DMN is actually operating “hotter” (in terms of entropy, but possibly literally) than the environmentally coupled or task specialised functions of the brain, yet in the context of discursive and pleasurable cognitive processes rather than depression.

    • Interesting thoughts in this post and this reply.

      If you’re making analogy to a physical process like crystallization of honey, it’s not just entropy that’s important. A process will be spontaneous if it has a negative Gibbs free energy; if you look up the equation you see it involves both enthalpy and entropy. Systems want to give off energy (to carry less inherent energy), and they want to become more disordered. An example where both are satisfied is burning a piece of wood. However, a strong reduction in energy can drive a reduction in entropy (crystallization of honey, which is exothermic), and a strong increase in entropy can drive an increase in inherent energy (dissolving sugar or salt in water, which is endothermic). It depends on the magnitudes (and thus balance) of enthalpy and entropy for a given system.

      I have a hunch there’s a useful analogy in comparing my mind to a spontaneous process that gives off energy to drive a decrease in entropy (like crystallization of honey). I need to have mental energy available to spend in compressing/reducing/converting the outside world’s entropy to the point that my DMN finds it legible. Depression arises when I don’t have the energy to keep doing this. Maybe because each conversion is too costly (rigid thinking, ADD, or simply a naturally independent thinking personality), I’m asked to do too many of these conversions (burnout at work due to an unconquerable workload), and/or both.

Leave a Comment


This site uses Akismet to reduce spam. Learn how your comment data is processed.