The Life & Mind Seminar Network

A unified brain theory?

Posted in Seminars by Marieke on January 21, 2010

This article may be of interest:

The free-energy principle: a unified brain theory? by Karl Friston. Nature Reviews Neuroscience 11, 127-138 (February 2010) | doi:10.1038/nrn2787:

Friston, who is a key player in neuroscience/imaging, pairs up lingo, ideas and results that share grounds with dynamical and enactive approaches, but presents them in a reductionist interpretation.

He talks about the free energy principle, which broadly translates to sensorimotor entropy minimization, about how brain function self-organises according to homeostatic principles, etc. – ideas that you would not necessarily expect to read about in Nature. He argues how many major theories of brain function can be seen as a special case of this principle (Bayesian brain, theory of neuronal group selection, optimal control theory, …). An interesting review of cognitive neuroscience with sharp observations and bold claims.

The problem I see is that the issues enactivists typically raise are not addressed. In his view, agents are heteronomous: value is defined as the opposite of surprise (entropy minimization), rather than autonomously generated in interaction with the environment. Friston talks about organisms as self-organising systems, but he never addresses the problem of the  distinction between the system and its environment and boundary construction. Similarly, in his conclusion he proposes that the brain is a “generative model of the world it inhabits”, keeping up a Cartesian divide between an observer-independent reality and its recration in the the brain, home to the mind.

Any thoughts?


8 Responses

Subscribe to comments with RSS.

  1. Nathaniel Virgo said, on January 21, 2010 at 2:23 pm

    I’ve been looking into Friston’s work a bit recently. It’s so full of holes it’s hard to know where to start. My main problem with it is that even if you ignore all the problems you mention and look at it from a purely GOFAI point of view — let’s try to build a robot that reasons about the world by applying information-theoretic principles to its internal model — the concept of *minimising* sensori-motor entropy is, in all the cases I can think of, the exact opposite of what’s desirable. Our robot could minimise its exposure to surprise by turning off all its sensors and remaining motionless until its battery runs out. If it were instead to *maximise* its exposure to surprise it would obtain the most information possible about the world, which seems a far more sensible thing to do.

    Friston justifies his notion of surprise minimisation with examples like a fish needing to avoid surprising situations like being out of water. But this is just confusing “surprising” with “bad”: if the fish were to deliberately leap out onto the riverbank, being out of water would be an entirely unsurprising consequence. Conversely, the fish could unexpectedly happen upon a tasty snack, which would be both surprising and beneficial.

    • Marieke said, on January 21, 2010 at 8:09 pm

      I was puzzled by the idea that certainty should be value, but then the idea relates to principles such as the homeokinetic principle (Der, Martius et al.) which appears to account for exploratory behaviour in the face of uncertainty. However, it would not explain switching to a different behaviour from some kind of internal drive towards novelty…

      But I am not sure entropy maximization is the answer either. An observation that has been recently made in several places is that life is characterised by alternating periods of openness to influences (behavioural switching, change of preference, play, curiosity) and of closure (pursuing a certain course of action to its end, falling into a pattern). Some work on this: Iizuka & Di Paolo on preference; Barandiaran, Di Paolo & Rohde 2009; Aucouturier & Ikegami, 2009; some other of E. Di Paolo’s works when talking about Heidegger, too). Trying to get a formal grip on this idea of alternating periods of openness and of closedness would probably mean to find oscillations in entropy – a bit like Aucouturier and Ikegami’s (2009) criterion of oscillations in information circulation.

      • marekmcgann said, on January 21, 2010 at 8:32 pm

        I haven’t dug into the Friston stuff yet (the Nature article is behind a paywall for me, but I got hold of some of his other papers that appear to be same thing), but there is some work in Psychology that I think is relevant to the issue of surprise/expectation.

        The concept of “flow” (Csikszentmihalyi, 2008) is a state of consciousness strongly associated with feeling happy and satisfied, and typically occurs when the challenges a person is facing well matches their level of skill. Specifically, we get bored by things that are too simple or too difficult (shall we translate to information/computation-speak as “too well predicted” versus “too surprised”?) and get drawn into things that are pitched just right for our capabilities. The switching between modes may be a failure of whatever mechanisms balance things rather than a specific or goal-directed change in behaviour, or perhaps a the switching to a mode appropriate to whichever side of the balance things have fallen.

        An idea that I’ve been toying with recently on back of that is that skill-development is like a damper (an entropy reduction process – am I getting control theory all wrong here?) between two processes in tension, the tug of war on an agent’s behaviour between the intrinsic dynamics of autonomy and the perturbations from the environment. If I can get it making sense I’ll probably come over and try to present it at a Life & Mind.

      • Nathaniel Virgo said, on January 21, 2010 at 11:56 pm

        To be clear, I didn’t mean to propose sensory-entropy maximisation as the answer to everything – I was just using the fact that it’s sometimes desirable as a counterexample to Friston’s claim (as I understand it) that entropy minimisation is the answer to everything.

        Entropy maximisation is only useful if your only goal is to learn as much as possible about the world. Organisms have all sorts of goals that can conflict with that, such as gathering food from known locations, conserving energy, avoiding the risks associated with exploration and generally sticking to behaviours that seem to be working out okay. I can see many situations where you wouldn’t want to maximise uncertainty, but equally I don’t think that minimising uncertainty (as opposed to risk) can be a useful goal in itself.

  2. Tom Froese said, on January 25, 2010 at 1:15 pm

    Just a general comment: Some of the problems in Friston’s paper can be found in the enactive approach as well. As I see it, the main difficulty is to shift our focus away from the stasis in homeo-stasis. Surely, permanence cannot be the main value we are after?

    As Hans Jonas remarked long ago: if stability is the defining goal of biological existence, then life should not have started out in the first place!

    I think an autonomy/constraint dialectic founded in the theory of complex systems might be a more promising approach.

  3. Ezequiel said, on January 25, 2010 at 2:06 pm

    I’m afraid the comment I’m about to make might be too narrow, but it may connect with the running themes so far. I have NOT looked into Friston’s ideas too closely. Apart for not having lots of time, there is one reason that has discouraged me from having a closer look. I must confess I stopped reading his TICS paper when I came upon this sentence on the first page:

    “Under ergodic assumptions, this entropy is:”

    Now, if there is something that brains are definitely NOT is ergodic. They’re are among the most non-ergodic systems in the universe. To clarify: this is a term from statistical mechanics. It refers to a class of system where the long term expected behaviour is the same as the average between the long term states of an ensemble of similar system. In other words, initial conditions are forgotten in ergodic systems. History doesn’t matter.

    The whole idea of a habit or a skill would be meaningless if a brain was an ergodic system. (Might this be connected with the other points raised in the comments, about surprise, homeostasis, etc?)

    Since my reading stopped at that point in his paper, I admit that maybe this assumption may play a weak role in his theory (some of you might be able to clarify this for me). In that case, maybe I will bother to finish the paper!

    • Ricardo Sanz said, on March 20, 2011 at 12:14 pm

      It is not the same to say that “initial conditions are forgotten in ergodic systems” and that “history doesn’t matter”. In fact the idea of “generative model of the world it inhabits” implies that the brain adapts to its environment by a history of interactions. History is hence critical for this capability. Starting point is not.

  4. Anete Kuree said, on December 2, 2010 at 10:45 pm

    Cool content. Waiting for you to continue the topic.

    Anete Kuree
    escort warsaw

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: