Saturday, July 3, 2010

Butterflys and Brains

Can you believe that your brain is intrinsically unreliable? I can't, I implicitly trust it. I have no choice. In this news release they address one of the great mysteries of nervous function: is "noise" a byproduct of nervous function or does it serve some fundamental purpose?

I used to become very annoyed at people who likened brains to complicated computers. The analogy is so stupid it is hardly worthy of serious consideration yet Artificial Intelligence bods adopted it as a working assumption. They should stick to designing computer games because I'm bored with ones I've got.

As an "information processing" device, and I have great difficulty with that phrase but I won't go there, the brain represents a formidable challenge in understanding its function. This is because......

Unlike our computational devices brains must function in a constantly changing milieu. Not only is the organism continuously confronted with a changing environment but much more importantly the internal milieu in which brains must function are constantly changing. Circadian rhythms are constantly shifting the concentrations of various physiological agents yet the brain remains remarkably stable and reliable. By way of analogy, try changing the voltage on your computer and see what happens.

Our computational devices are all of a one, everything can be reduced to "0" and "1". Brains cannot function in this way because they are so messy. There cannot be, for the very greater part, any precision in neural connections. Synapses come and go all the time, volumes in varying regions of the brain are dynamic, constantly shifting in demand to both internal and external environmental contingencies. Then there are concepts like "volume transmission" which refers to the manner in which various neurotransmitters are released across relatively wide areas, rather than, as the typical "Bookworld analysis" suggests, just jumping across synapses.

It is completely incorrect that behavior is mediated by the brain. Stuff and nonsense, behavior is mediated by the whole body in interaction with the environment. We are inclined to believe that simply understanding how the brain works will enable us to understand behavior. Nah, not going to happen, in fact I believe it is impossible to create a comprehensive theory of behavior, that there will always be a large indeterminacy in predicting behavior(too bad for Asimov's "Foundation Series", great novels predicated on atavistic notions about human behavior), and this study does highlight why such an indeterminacy will always bedevil us, no matter what the likes of Dr. Phil may proclaim.

For now, excluding the external environmental issue, we must recognise that an understanding of nervous function needs to incorporate not just neurotransmitters, but quite literally an overwhelming array of physiological agents that have time and again being demonstrated to play a cardinal role in mediating behavior. This is, unequivocally, a computationally intractable problem. Again, this very obvious fact is lost on Artificial Intelligence researchers, who operate under the delusion that all one needs to do is understand how neurotransmitters affect nervous function. Go back to writing computer games.

This matter of "noise"

"Noise" is probably a poor appellation. I suspect this noise plays a fundamental role in regulating nervous function. My suspicion is that "noise" is highly variant across time and cerebral regions. It is not a nuisance but a fundamental requirement that drives novelty and creativity. At present we call it "noise" because that is the best we can do but perhaps I'll live long enough to see the day when some clever bod manages to distinguish various "noises" and so demonstrate that it is not a byproduct of neural function but a fundamental and critical product of nervous activity.


Steve Edney said...

in fact I believe it is impossible to create a comprehensive theory of behavior, that there will always be a large indeterminacy in predicting behavior(too bad for Asimov's "Foundation Series", great novels predicated on atavistic notions about human behavior)

Well yes. I remember though that in the novels he tries to get around that by saying that masses of people are predictable and at the level of the universe extremely so. More or less an appeal to a behaviour version of the central limit theory. Of course it is also rubbish (not the CLT itself, but its application to social science). For it to work individual behaviour would have to be random and constrained.

Unfortunately individual behaviours are neither random, constrained nor predictable.

John said...

The ideas Asimov explored with plausible and relevant in his time. There was MK Ultra, that doomed CIA experiment to imitate brainwashing techniques used by the Chinese. Fail.

The group perspective though does still show some promise. Something strange and fascinating going on there. I recall a book "Criticality" by a physicist, wherein he outlines how many complex physical processes can be mathematically modelled. Then there is this, a real ray of hope:

Steve Edney said...

I've read criticality many years ago. In fact my PHD was partly on this topic.

The issue generally isn't that we can't know anything about the system, we can model the system to get out statistical properties out but generally can't predict in meaningful detail.

For example there is some sugesstion that both earthquakes and solar flares are caused by analogous phenomena which is why both have similar stasitical distributions of size vs frequency. The basic idea was that in an earth quake you have strain building in various points of the earth's crust, periodically a small section gives, releasing strain locally and then putting it onto surrounding regions. Depending how strain is distributed this may result in more regions giving, while others take a bit more strain. Whether there is a mass reaction or a small one then is a function of the local strains. you can build models of this type of thing which reproduce similar statistical distributions of outcomes. The downside of this is that it means that while we will know the likelyhood of Earthquake of size x occuring in the next time t we will never be able to predict where and when such a one will arrive as it depends entirely on the entire micro structure of strains in the earths crust.

A similar idea is proposed for solar flares where it is the corronal magnetic field which twists and then reconnects releasing energy.

the idea in all these is that the systems are kept at the critical point naturally by energy being driven into the system, and being released balancing on average.

Anyway this type of model may be very useful, if all you want are statistical outcomes, if what you want however is the specific details, they are telling you more or less that you will never actually get any good knowledge of the details.

John said...

Thanks Steve,

One of the problems with modelling behavior is the presumption it is caused by the brain. Yet there are multiple studies indicating that the whole physiology is involved in some surprising ways.

One striking example is on a study about Tax gondi, a parasite that is very common in humans and is contracted through cats. The parasite resides in the brain and in mice at least has a pacifying effect, making them unafraid of predators. The paper I have in mind argued that given the very high infections rates in some parts of Brazil(60%+), the cumulative effect probably does have cultural implications.

Re earthquakes, this came up a few weeks ago.

John said...

I forgot to add:

I loved the old Hard SciFi, which seems to have largely disappeared from view but a friend of mine reckons the mathematical explorations of Greg Egan, an Aus scifi writer, are a treat.

If it's a page-turner you're looking for, Egan's Permutation City beats Connes' Noncommutative Geometry hands down.