paraphrasing, "destroying the village to save the village". — TheMadFool

The theory here is sort of like that. You need to destroy surprise in order to be surprised. You have to create a baseline where the world is made as predictable and unsurprising as possible. That then allows you to experience the counterfactuality of events which are actually surprising - events that have personalised meaning or information because they must force you to revise your beliefs about the world.

Surprise can’t exist in the usual sense if everything that happens counts as something out of the blue. That becomes just randomness.

The brain desires salience. It has to discover the signal by first eliminating the noise.

He is giving neuroscience its own proper physicalist foundation - Bayesian mechanics - to wean it off the Universal Turing Machine formalisms that want to treat the brain as a representing and simulating computer. — apokrisis

Any kind of science needs to be driven by evidence, not a desire for totalising unity. We do not accept theory on the grounds that it's neat, but on its accord with observation.

That, and a theory that insists we minimise surprise by maximising surprise is just incoherent.

When he talks of surprise, it is as a technical term within a new mathematical structure. — apokrisis

He defined surprise in the article; I quoted it above. Yes, a human is only one example of a biological system, but you only need one counterexample to falsify a law.

The idea that this is what _evolution_, rather than a biological system, does is quite interesting. Generally, I tend to see nature working in the guise of optimisation procedures, and free energy minimisation is an optimisation procedure. The problem again is that there are already optimisations that better fit the facts. Essentially equating surprise minimisation with fitness might be neat, but it's not really explanatory. Evolution isn't really trying to minimise surprise; it's trying to maximise fitness.

The theory here is sort of like that. You need to destroy surprise in order to be surprised. You have to create a baseline where the world is made as predictable and unsurprising as possible. That then allows you to experience the counterfactuality of events which are actually surprising - events that have personalised meaning or information because they must force you to revise your beliefs about the world.

Surprise can’t exist in the usual sense if everything that happens counts as something out of the blue. That becomes just randomness.

The brain desires salience. It has to discover the signal by first eliminating the noise. — apokrisis

Someone mentioned children - fresh minds, tabula rasa. To younglings, the world is full of surprises because they haven't had time or are in the process of constructing a faithful model of the world, one which would help them to anticipate events, prepare for them, etc. and then they mature into adults. As adults, growing older is just another name for accumulation of empirical data to refine said model which is an asymptote for actual/true reality. My guesstimate is the model any person develops should be a good map of the territory by the 25th year of life if it's to be of any use at all.

Friston's particular contribution is in bringing the Helmholtz free-energy approach to bear on the problem, and then trying to extend it beyond cognitive science to living systems in general. — SophistiCat

:up: Well said.

As a model of neurophysiology, it goes back to the first efforts at mind science in the 1800s. What is new is to cash out the story as actual maths and physics.

Psychology went down the wrong path of trying to found itself on computer science in the 1950s. Now it can join biology in rooting itself in the maths and physics of dissipative structure theory.

The Hard Problem gets sorted at root if mind, like life, is shown to be a physical principle from the getgo.

Friston positions Bayesian mechanics as the new fourth branch of physics - following on from classical mechanics, quantum mechanics, and statistical mechanics.

Any kind of science needs to be driven by evidence, not a desire for totalising unity. We do not accept theory on the grounds that it's neat, but on its accord with observation. — Kenosha Kid

So Kuhn was wrong about paradigms? Whatever.

(Do you see how you just employed the data processing paradigm that Bayesian mechanics replaces to try to argue against Bayesian mechanics? Oh the irony.j

Yes, a human is only one example of a biological system, but you only need one counterexample to falsify a law. — Kenosha Kid

You are creating your own confusion by talking about surprise as if it were just a “feeling” here and not an information theoretic metric.

Check out Friston’s recent Markov Blanket and Unified Theory of Unified Theories presentations if you want to get beyond the verbiage and kick the tyres of the actual maths.

Someone mentioned children - fresh minds, tabula rasa. To younglings, the world is full of surprises because they haven't had time or are in the process of constructing a faithful model of the world, one which would help them to anticipate events, prepare for them, etc. and then they mature into adults. As adults, growing older is just another name for accumulation of empirical data to refine said model which is an asymptote for actual/true reality. My guesstimate is the model any person develops should be a good map of the territory by the 25th year of life if it's to be of any use at all. — TheMadFool

Agreed. But what is also key is that the map of the territory is one that is a map of the territory with oneself in it as well. So it isn’t a map with the whole world represented, it is a map of the route you want to take to complete your self-defining life mission. It is a map of yourself as much as a map of the world you must inhabit.

This is the difference between a Cartesian representational model of what the brain does - the computer science model - and an enactive or embodied view of cognition. Our neural models of the world are maps which embody a personal point of view.

As I recall, I was quite content in my darkened room until I was expelled from it by a nightmare squeezing that left me beached on a bloody sheet gasping for breath. Breath was the second surprise. Darkened rooms are unavailable for a longer lease than about 9 months. Thereafter, minimising surprise involves seeking out surprise, aka novelty, in order to familiarise oneself with it. I think this is known as "learning|". — unenlightened

Yep. That's the argument.

So Kuhn was wrong about paradigms? — apokrisis

That has absolutely nothing to do with it.

Do you see how you just employed the data processing paradigm that Bayesian mechanics replaces to try to argue against Bayesian mechanics? — apokrisis

I'm not arguing against Bayesian mechanics. Not liking turkey doesn't make me a vegetarian.

You are creating your own confusion by talking about surprise as if it were just a “feeling” here and not an information theoretic metric. — apokrisis

Again, the definition is quoted above. I know your initial MO was to claim that, by disagreeing, I must be employing a different definition, but it's a tad late for that now. I'm speaking of it in the same terms as the paper, both as a metric to optimise and, as per the paper, how it ought to be used scientifically, i.e. in predicting outcomes. And, as I said, I'm quite sympathetic to viewing nature in terms of parameter optimisation, that's my go-to. But, again, just because I like duck, doesn't mean I like turkey, and this is a turkey.

That has absolutely nothing to do with it. — Kenosha Kid

:yawn:

Again, the definition is quoted above. I know your initial MO was to claim that, by disagreeing, I must be employing a different definition, — Kenosha Kid

And yet where you correctly quote surprise as it is defined by Friston as....

mathematically, surprise is also the negative log-evidence for the model entailed by the agent. This means minimizing surprise maximizes the evidence for the agent (model).

You then go on to discuss it in terms of some pop-neuropsychology bullshit...

When it finds something surprising, i.e. that the model could not predict, it rewards itself with a hit of dopamine. — Kenosha Kid

A "hit" of dopamine in the nucleus accumbens triggers a switch from acting on habit to responding with focused attention. The prefrontal cortex is also "hit" with dopamine so as to underwrite the fixing of the moment in working memory.

It is not about subserving a feeling - even if it might feel like something to be alerted, focused, engaged. It is about a certain kind of surprise or prediction error that leads to a positive orientation response. A global decision to approach closer and explore, gather more information.

Then surprise or prediction error can trigger the opposite form of orientation response if that is judged more appropriate. We can flinch, pull away, prepare for fight or flight (or even simply freeze). The brain's hierarchy gets "hit" by other neuromodulators like noradrenaline - a transmitter with information processing consequences like jacking up the signal/noise ratio of our neuronal responses. When facing radical uncertainty - like a lurking tiger - the brain tilts towards a willingness to jump to rapid conclusion based on limited data and prior experience.

So going off on some riff that surprise = dopamine = reward is just telling me you haven't studied the role that neuromodulators play in the complex information economy of the brain.

There is a hell of a lot more to it all. Which is why it is so valuable that Friston might boil all the complexity down to its simplest possible mathematical expression.

Evolution isn't really trying to minimise surprise; it's trying to maximise fitness. — Kenosha Kid

Evolution doesn't try anything. Evolution is just a human invention, replacing a notion of god and his creation of life. There is no such thing as evolution standing above life, trying to create maximally fit life. It is life itself that evolves, not life guided by evolution. And life evolves in ways that are fit for the habitats they evolve in.
Likewise, it's nonsense to say that life evolves under the strict influence of selfish genes which shape the vessels they are in in order to procreate themselves, or that memes guide human behavior in order to spread them. Of course, gene prolongation and memes hopping from mind to mind is what happens (like evolution) but life is not ordered by them (nor evolution).

There are parameter maximization parameters indeed. If there is only the expected around, then the free energy will be tuned to the minimum (or entropy to a maximum) for which life will stay alive. Living in a dark room, how much of an unexpected event it might be, temporarily creating increase of free Gibbs energy, will become a state one gets used to, and a state of minimal free energy will arise.

Is it really though? No, although the theory here seems to imply that (that life, once there, is a drive towards minimal free Gibbs). The dark room state will create rest at first, but then the excitement kicks in. Like fish kept in dark water develop big eyes. There is a need to encounter surprises or challenges. To exist in a world with surprises. Staying in a dark room will drive one mad. It might be a surprise to end up in it, but that's not the surprise one needs. Though occasionally it's nice. Like going to sleep temporarily reduces free energy. Maybe there are even people who want to stay asleep their whole lives, I dunno (didn't I hear it sung: "I am tired, I am weary. I could sleep for a thousand years. A thousand dreams, that would awake me. Different colors, made of tears").
Living in a dark room might prolong life, but only for people not expecting anything.

That's much clearer than you first posts, which left me with the impression that you simply weren't addressing the article.

There are other, better, more factual reasons why we fear dark caves. — Kenosha Kid

The difficulty with the surprise avoidance theory is that, even here, it provides an answer: the cave is to be avoided precisely because it holds surprises. You are right to draw attention tot he methodological issue, which seems to me to be the same as that addressed by Watkins, and used as the basis for a previous thread: namely that if any action on the part of an organism can be explained as avoiding surprise, then the explanatory value of the theory is zero. Those who see it as true in some a priori fashion (Apokrisis?) will always be able to explain any given observation in terms of the theory, but at the cost of introducing ad hoc hypotheses to make it fit.

The problem is in trying to model all human behaviour according to one general rule when in fact it is an interplay between many physical processes evolved at different times in different environments, some overriding. Our fear of lurking tigers _is_ quite different from our innate curiosity for the novel, and should be treated as such. — Kenosha Kid

I would not be at all surprised to see in a few years a crusading pedagogue explaining how the only way to teach kids is to minimise the surprises to which they are exposed. And there will be schools that follow that advice, with mixed success.

But then, it is fascinating to see ideas from different areas coming together in a more unified theory. The explanatory power of surprise avoidance will take years, and much subtle empirical evidence, to evaluate.

No mice were harmed in the preparation of that article.

Andy Clark (of The Extended Mind) — SophistiCat

Did we do a thread on that article? Perhaps its time for another look. Clark is a co-author on the Free-energy minimization and the dark-room problem article of the OP.

No, although the theory here seems to imply that (that life, once there, is a drive towards minimal free Gibbs). — Cartuna

But the theory actually states that life expresses the drive to avoid becoming randomised by its environment.

It gets confusing because life and mind must also be able to do work. The free energy principle is being treated here as a statement that the goal of life and mind is to do no work - retreat into a dark room and do nothing. But what it actually is saying is that the goal of minimising free energy is how a baseline for personal being gets established. The first requirement is to be able to have a structure of belief about the world - a reality model - that is not immediately being entropified away into uncertainty. That baseline is then what sets the scene for the second thing of actually getting to work on the world.

In constructing some state of certainty, some meaningful or autopoietic distinction is made between the self and the world. And that epistemic cut between the world and the self is what allows us to deal with entropy or information uncertainty to the degree it has some practical interest for us.

So the model is of course a structure with high negentropy - and island of wisdom afloat in a world of dissipation or entropy increase. There is a historical investment in some structure of learnt habit. And now the free energy principle is about maintaining that investment by minimising the need to make any changes to it.

But that just then limits the need to make change and promotes enduring stability. The environment is of course going to continue to demand adaptive change to the neural model. However, that change is now focused just like a Brownian ratchet. Encountering uncertainty leads to adaptation or learning by the model, rather than randomising it - eroding our historically constructed structures of belief.

The free energy principle imposes a thermodynamic direction on the modelling relation the organism has with its world. By pre-filtering noise, what comes through is by definition salience - a meaningful signal. So what should ordinarily act to scramble a structure of certainty - increase a model's entropy - instead only is allowed to increase its negentropy.

If you can wedge the forwards steps, you can milk a randomising environment for its energy in Brownian ratchet fashion.

Coincidentally, this same principle is also the biggest paradigm shift to hit theoretical biology in the last decade. It is how life operates down at the quasi-classical nanoscale, as with kinesins and other molecular machines.

The explanatory power of surprise avoidance will take years, and much subtle empirical evidence, to evaluate. — Banno

Probably won't ever achieve it though.

It would be a cowardly act and there is something fundamentally valuable about courage?

Agreed. But what is also key is that the map of the territory is one that is a map of the territory with oneself in it as well. So it isn’t a map with the whole world represented, it is a map of the route you want to take to complete your self-defining life mission. It is a map of yourself as much as a map of the world you must inhabit.

This is the difference between a Cartesian representational model of what the brain does - the computer science model - and an enactive or embodied view of cognition. Our neural models of the world are maps which embody a personal point of view. — apokrisis

:up:

It is not about subserving a feeling - even if it might feel like something to be alerted, focused, engaged. It is about a certain kind of surprise or prediction error that leads to a positive orientation response. A global decision to approach closer and explore, gather more information. — apokrisis

Why do you think that dopamine response evolved? Why do you think children are naturally drawn to novelty? Just for the luls? It's there to maximise information for building models about our environments.

You then go on to discuss it in terms of some pop-neuropsychology bullshit... — apokrisis

I'm dismissing it in terms of its fidelity to actual biology. That will have different degrees of importance to different people, of course.

The difficulty with the surprise avoidance theory is that, even here, it provides an answer: the cave is to be avoided precisely because it holds surprises. — Banno

Tbf I've never argued that this was even a good question. There really _are_ optimisation procedures for ensuring that biological models of environments (not just mental models, but in the physical makeup of the organism) are a good fit for the environment you find them in. Nonetheless, I like night-diving. It's excellent! And the only time I'm apt to have to deal with that environment is... when I'm night-diving.

So okay we can say "Ah but yeah no the algorithm doesn't know which environments we have to adapt to in future, so it's best to sample them all." So then which is it? If the phrase "minimising surprise" must mean not living in dark caves because we don't have a model for that, and exploring things not in the model to prepare us for future surprise-minimisation, i.e. both X and !X, it's not a good definition. The article reminds me of bad psychologists and sociologists who pick the statistical tests to give closest to the desired answer, except it fudges its meaning over scales instead.

Those who see it as true in some a priori fashion (Apokrisis?) will always be able to explain any given observation in terms of the theory, but at the cost of introducing ad hoc hypotheses to make it fit. — Banno

This might be above your pay grade, but listen to Friston and he tells you he understands that he offers only a generalised theory of everything. To then use that as the basis of models of particular functions or systems is another matter. Of course that becomes a more ad hoc exercise - even if just for the reason that the beautiful theory must be spoilt by random historical accidents. :razz:

Australia got the kangaroos, Europe got the cows, New Zealand got the moa. The general logic of ecology demands large vegetation munchers to balance the entropic equation, recycle the material negentropy and keep it within the system. And while kangaroos, cows and moa are all equally good solutions at that level of explanation, the reasons why they are all such different solutions becomes a matter of historical accident as far as we can tell.

Again, what matters is that Friston offers a mathematics - Bayesian mechanics - which captures the essence of the semiotic modelling relation in a way that the mathematics of Turing computation doesn't.

It is counterfactually an improvement on the kind of logicist paradise that has pervaded the study of life and mind up until now.

Neuroscientists and biologists always knew that organisms aren't simply machines. Friston provides a general mathematics that supports that.

I would not be at all surprised to see in a few years a crusading pedagogue explaining how the only way to teach kids is to minimise the surprises to which they are exposed. And there will be schools that follow that advice, with mixed success. — Banno

If folk are too dumb to understand what is actually being said, then sure.

The explanatory power of surprise avoidance will take years, and much subtle empirical evidence, to evaluate. — Banno

There is a vast amount of empirical evidence already. Helmholtz put psychology on a scientific footing in the 1800s by showing that prediction was the basis of cognition and sensation.

But then along came computer science and its logical atomism. The machine model of reality. Psychology lost its way for a good while, especially in the Anglo world.

But what it actually is saying is that the goal of minimising free energy is how a baseline for personal being gets established. The first requirement is to be able to have a structure of belief about the world - a reality model - that is not immediately being entropified away into uncertainty. That baseline is then what sets the scene for the second thing of actually getting to work on the world. — apokrisis

That baseline is achieved indeed by maintaining a certain free energy that's needed to live a life. In order to maintain that minimal required energy (a maximum possible entropy, above which the system collapses to a lower free energy, like a mixture of gas molecules reducing their free chemical potential energy if not kept below a certain temperature), there have to occur as less surprises as possible. The status quo can be maintained. If unexpected things happen, the free energy has to increase. Excitement occurs. Information increases, depending on the new situation. Free energy and information temporarily rearrange, and a new status quo will be achieved. After which the situation is familiar and free energy will diminish again. One can arrange life to meet as little surprises as possible, like seek sanctuary in a dark room, but surprises are needed in life.

Sure, all that. What I said still stands: the notion described in the article is interesting, but not decided.

I was hungry, for real. I had 0 burgers, 0 hotdogs, 0 eggs, and 0 liters of milk. I'm, for some strange and unfathomable reason, still hungry.

So I understand that you think the dark room objection carries some considerable force against the surprise avoidance hypothesis because behaviour is more complex than can be explained by that hypothesis. Is that right?

What I said still stands: the notion described in the article is interesting, but not decided. — Banno

This is a reasonable conclusion.

Why do you think that dopamine response evolved? Why do you think children are naturally drawn to novelty? Just for the luls? It's there to maximise information for building models about our environments. — Kenosha Kid

Do you read what I say? Of course once you can prevent the environment from increasing your belief uncertainty, you then lock in the possibility of ratcheting belief in the direction of ever-broader uncertainty.

It is monotonic to claim "children are naturally drawn to novelty". They have a natural need to learn and construct a stable neural model of the world. They have a natural need to take the random, over-connected, mess of synaptic connection which they are born with and prune the thickets back to give them useful shape. They have a natural need to act on the world and so discover the good and bad consequences that might follow.

Your framing of the issues is too one note.

I'm dismissing it in terms of its fidelity to actual biology. — Kenosha Kid

And you are simply wrong on that score. You don't appear to understand the theory, so no surprise you don't understand what could count as legitimate criticisms of it (and I believe there are criticisms, such as that Friston is so physicalist that he doesn't give sufficient place to the role that semiotic code principles play in actually being the ratcheting mechanism at the heart of the Markov Blanket).

the cave is to be avoided precisely because it holds surprises. — Banno

A cave is no dark room. I know exactly what to expect in the dark sleeping room. A dark cave may hold surprises. But why not enter it? Because it would increase free energy? That would be the result. Trying to be a minimum would not be the cause of not entering though.

I was hungry, for real. I had 0 burgers, 0 hotdogs, 0 eggs, and 0 liters of milk. I'm, for some strange unfathomable reason, still hungry. — TheMadFool

I'm sure there is some dark room in which burgers are smellingly inviting you.

I'm sure there is some dark room in which burgers are smellingly inviting you. — Cartuna

Why do you care?

Do you read what I say? Of course once you can prevent the environment from increasing your belief uncertainty, you then lock in the possibility of ratcheting belief in the direction of ever-broader uncertainty. — apokrisis

You seem hell bent on missing the point. Fill yer boots obvs.

And you are simply wrong on that score. — apokrisis

:rofl: Okay, strong argument, I got no counter. Peace out.

Well, the discussion is about the avoidance of dark rooms because there might be surprises in there, and minimal free energy considerations would keep you away from it. Still you wanna increase your free energy to find out what's in the dark room, maybe the fridge.