Friday, 24 December 2010

This is not an EM mind theory.... ITS STRUCTURALISM !

Central to this integrative neuroscience project the brain structures are proposed to be electromagnetic in form. And the proposal here is that intelligence has a definitive physics based form, so it might be thought this is the mother of all EM mind theories. But paradoxically this approach is not currently aiming for an EM mind theory. This post will try and explain how such a paradox could be.

Of course with something like a cortex dipole structure, it is important to be open minded to some types of magnetic brain model for various interesting concepts and developmental mechanisms, so I do mention various respected authors on this site. But for now its important just to define the physical system, and always ask if conventional concepts can fill in the rest. Overall I stick with traditional approaches and do not get into quantum mind theories. For the following reasons.

The concept of quantum mind is over-rated and takes us off track. It has been popular because theorists were fusing “cool” topics like Einstein and Darwin, and many academics wanted to be the first to claim this territory. And for others it has been the “soul” or escaping into the mysticism of quantum physics. But what structure of computation would a quantum mind even bring anyway ? Surprisingly many EM mind theories don’t even explore or formalize this basic aspect ! Now that we are actually building quantum computers its a a lot clearer what that would be.




What quantum computation really brings us, is dense matrix connectivity, something we already simulate with digital matrix systems. Which is why the Dwave had to scale up its system to prove its advantage over digital.


Findings of the entire set of Quantum properties in classical fluids that show us fluid dynamics can solve quantum weirdness in terms of pilot waves. So it appears like wave/particle could be unraveled and there may be no quantum/classical divide after all. You would think this would be interesting, but there is often a blanket denial of this information when raised in a discussion. Such denial serves to illustrate there is more of an attraction to the mysticism of the quantum ideas rather than a desire to get to the substance of solving the problem. i.e. What is wave particle duality really about ?




Yves couder in his fluid dynamics lab in CNRS Paris. His team have created the quantum phenomena by slow motion filming the motion of fluid droplets after they are oscillated. We uncover the same patterns for entanglement, and many other quantum phenomena. These accord with De Broglies Pilot wave concept. Put simply the waves guide the particle, so they are bound together.


Its important to point out what an EM mind theory is. For example the traditional neuron is defined by a combination of EM fluid equations, but the computation is still defined by traditional logic. Although even that is contested in neuroscience, however most neuroscientists do not subscribe to a full EM mind theory which would be a quantum mind theory, popularized to an extreme by Stuart hameroff. He proposed most neurons operate together and compute via EM fields orchestrated across the brain via the inner neuron structure (microtubules). The concept raised many interesting issues but did not survive scientific testing and was weakened by the arrival of more complete brain mapping projects. We now have new information such as that of dendritic computation, which places the logical process of neurons in the dendritic webs.




Where we now are sure computation emerges. From dendrites (1) to population burst patterns (2) to entire brain regions locked together in phases (3). We could represent these phase locks to have similar patterns as the quantum interference patterns, but we know they are classical patterns. So the same phenomena can emerge by different means.

What is important to bear in mind about the dipole cortex concept is that it does not seek to replace or do away with current neuroscience. Its more of an integrative approach to piece the brain together into a larger picture. For example although it proposes an EM structure for the brain, its not real time EM action like a quantum mind. A quantum mind theory would propose magnitudes faster neural processing than the millisecond of the action potential.




In biology there are many interacting levels and different types of physical principles. Physics in coded and controlled by biology, so a biophysics of brain structure needs to represent all these (and more) different levels within the same picture. In the brain structure the radial glia which produce the magnetic fields pulse these fields when migration is required. Then stop while other genetic or biophysical processes are in operation. So like conventional neuroscience, the solution will not be simple concepts such as a prevailing magnetic field model for the brain.


In neurons electromagnetic fluids are recruited and controlled by genetics, to the point that every aspect of how physical principles operate is slowed down in time, so they can work together within the limiting frameworks of how they are built, maintained and can reproduce their functions. All this becomes so complicated that the action potential cannot transfer into the speed of a pure electromagnetic field as it has far too many sub aspects inside to build, produce and maintain it. All the same applies to the proposal presented here for the brains EM structure, so it is entirely consistent with current neuroscience



This is just some of the complexity at the synapse in terms of protein signalling, and represents only a fraction of the aspects shown. For example if we were to fill in everything we know about the membrane it could not fit on this graphic. Same for the electrophysiology, genetics, neuroimmunity, Glial cells, Plasticity, Signal responses and much more.


More specific problems with quantum or even magnetic mind theory

I have to remain open minded, but so far there does not yet appear as if there is any "real time" EM role in actual adult processing. At least nothing that could threaten current models. So for computation this project develops with traditional neuron/structure approaches, and then considers as an adjunct any EM aspects mainstream (not fringe) neuroscience has found. For example it is already accepted in science that magnetic structures have a variety of computational properties so from that we can consider what could a magnetic field contribute to information processing when it assists in and is integrated into the development of complex computational structures.


Magnetic principles do not really scale up it appears. If they are coherent in development due to radial glia (which dissolves early in development) then this means loss of this means for a widescale magnetic mechanism in the adult brain.

The first problem is that we are not sure if there is any large scale field in the brain that arises from individual magnetic domains, because MEG readings are taken at the population level, and the equipment to measure individual axons while still active in the larger field is only just being developed.

The second problem is the magnetic mechanism (radial glia Ca2+ in connexins) mostly fades after cortical development and leaves the glial system. So for this approach proposed here, I don’t have a mechanism for the developed brain. At least not in the scale required to fit the concept of a magnetic multipole system.

The third problem is neurons are operating at sub-electromagnetic speed. They are evolved for the millisecond range. And why would brain networks even require such a field, when there is already all the connective machinery in place anyway ? Its already explained well enough. The outstanding questions in neuroscience (like how coding works) are not illuminated by current QM mind theories.

The biggest problem for EM mind ideas is the emergence of most ordered complexity at the microscale, and the way in which biology controls/encapsulates/integrates many difference principles of physics

And more ! see this for criticism of quantum mind concepts

If there is no EM mind theory, then why would there be an EM structure ?

The ideas presented here are more guiding concepts to understand organizing principles in neurodevelopment. Currently this project is not looking to extract from these concepts any computing with continuous EM coherence or magnetic organizing effects at the atomic scale. All we need to derive is the overall structure of computation that EM fields might leave in their imprint. But still remain partly open to the possibility of some EM aspects we don’t know off.





Some brief examples of how electromagnetism can assist in the development of good computing structures.  Magnetic fields can assist the coding of genetics to ensure microtubules fall in line with a soliton axon model, and these are all bundled in a neat linear manner suitable for good volume synchronized coherence of information transmission.  This property is called flux, or magnetic ropes. Magnetic clustering (right in diagram)  is actually used as an algorithmic method for sorting information in computer science. The magnetic ordering induces very good hierarchical structures. 


So in the actual cortical structure it is proposed that the structure provided by magnetic fields to the cortical asymmetry, would provide impart different types of tree structures that when filled with neurons have good structural properties for feature extraction. Magnetic fields are a very useful tool for nature to get cheap and controllable computing power which scales up well to increasing complexity. Neurons evolved first, then brain structures evolved to use magnetic fields to order increasing amounts of neurons with some very good structural computational properties.

However with our current knowledge it can also be proposed that even if EM fields stop in development, they will have imparted a computing structure which leaves an imprint of the essence of general computational properties. This “essence” evolved to integrate nicely with being filled by the brute force numbers of neuron sub-components. So nicely all that has to happen is the structure develops/crams as many neurons into the computing structure, then only has to prune these out or form some more later in limited numbers.


The process of how to build a structure can be itself tied into how the structure itself operates. The physical constraints of the building process if self evolving (rather than designed) would naturally ensure this takes place

What general computational properties or “essence” would these EM fields be ? This is another part of the project in itself, but it would basically be a simple dual process computational system which arises from fundamental physical properties. Most AGi theories (with good test results) are dual process theories and can be reduced to dual processes in physics. Googles Deepmind (the Atari beater) is basically dual processes of action-perception.


Googles Deepmind system is based on a simple dual process computing. They use Deep-reenforcement.  Dual process underlies even human cognition (see wiki summary here). If dual process underlies cognition it is reflected in brain structures (see wiki summary here). I have written an unfinished AGi paper (2012) which  used dipole neurology and dual process principles to predict what Deepmind would require, even before it was built !(see here). My later 2014 paper is actually a dual process AGi concept, but applied primarily to physics and neuroscience.

The basic concepts behind Googles Deepmind general learning can be expressed in a few elegant equations based on universal induction, and these can in turn be expressed in fundamental physics. But if you provide these equations life by dressing their operational structure with a larger brute force numbers of computational elements, they will in principle become complex general learners, even if the computational elements vary a lot. See wolframs principle of computational equivalence for more on this. Or simpler put, the structure of a mechanical machine can still operate when made out of many different types of elements.


Stephen wolfram filmed explain the principle of computational equivalence (see video here).  This principle is based on a large project which used computers to layout the landscape of proof systems (entire book here).  At this stage its an idea not to take his approach as proven for the landscape of all complexity, but it is interesting for the more well defined system of the brain.


SUMMARY

Because quantum computers do compute, It was always important to be open minded there could be EM effects. I have done so in this project and gathered the best I can find together. Those remain part of the project. And the concepts from this project, such as feature extraction resulting from converging/diverging networks derived from magnetic poles is an EM derived computation. But if there is no real time EM field, then its also a classical computation. It is also very important to ask what quantum computers do to information physically and can this occur by other means. Or is there even an EM/classical divide after all ? I think there are probably many ways to produce similar dual process types of computation at different levels of complexity. EM fields evolved as they enable a simple and cheap way for nature to get us a good dual process computing structure that we can shoehorn computational elements into via chemotaxis (neuron guidance).


So this is why this is not an EM mind theory as currently known. It is more suspected the actual highest level computing scheme to be derived for the brain will be an adjunct framework which is consistent with current neuron population models or biophysics concepts such as walter freemans mass action. The computing derivation is still a work in progress, and that is why I publish on AGi. To compare how universal computing matches with neuroscience. My 2014 paper started from basic thermodynamics (as a foundation). A recent post here outlines some broad themes in the meanwhile.