Friday, 8 March 2013

Blue Sky Research and the Black Hole in Brain Research

Following my posting "How the Human Brain works – concept cells, memodes and CODIL" I got a message suggesting I look at a recent article, in The Times of 2nd March, outlining the Human Brain Project and the work of Professor Henry Markram. In reply I drafted the following note which explains why I feel that research, such as the Human Brain Project, and the American  Brain Activity Map Project are looking in the wrong place - and why blue sky research which looks for radical alternative solutions "outside the box" looses out.
The article on the Human Brain Project is interesting in that it highlights the problem with modern brain research. Professor Markram says “The fact is, we're still in the Dark Ages. We're a little better than we were 6,000 years ago” which fits entirely with my view that there is a black hole in brain research.

The problem, in my view, is that virtually everyone in the field is trying to find the “Philosopher's Stone” of Intelligence – that extra special something that makes us different to animals. We take it for granted that we are clever – and that there must be something “very clever” in the brain to make us what we are. We are so big-headed that in the past we thought the earth was the centre of everything and the sun and stars went round the earth. Few people still hold that view but we still cling desperately to the idea that there is something exceptional about our brain apart from its size, and what culture allows us to do with it.

The Human Brain Project is a good example of the search for “elixir of intelligence” which is basically a rat race for academic prestige – bigger – better – faster – more computers - more expensive teams of narrow minded specialists – sorry but no time to stop and think outside the box or someone else will beat us to the limited research funds, or worse still, be successful in their search and published first. Even if it succeeds in mimicking the brain there is not guarantee that it works in the same way! In fact millions of people are looking for “the answer” but as far as I can found out no-one has come up with a model, stage by stage, of how the brain's electrical impulses lead to support human [and animal] intelligence. I suspect that you hold this view and are unaware of anyone who has succeeded in such research.

The key to good research is to find the right questions to ask – and not just to charge off in the direction the establishment thinks the answer lies because everyone else is going in that direction. In fact often the most interesting scientific questions are those that fall way outside the box of ideas that the establishment deems to be acceptable. Unfortunately the science rat race, with enormous sums (and prestige) going to a small number of major projects in “top” institutions can mean that genuine blue sky research can get thrown out with the bath water.

The model I am working on [See Brainstorms and information about CODIL in right column of blog] assumes as a starting point that the only significant difference between a typical mammal brain and a human brain relates to capacity (number of neurons and number of interconnections per neuron). I start with a minimal pattern matching model that any animal that can memorise and use facts about the environment must have in order to do anything useful with the its brain. I then look at what extra (if anything) one needs to be able to, for example, support language. Even if the model will not go all the way, it points to the areas where the model suggests possible approaches. After all I work on the basis that any good research model should have predictive power to suggest further research.

My difficulty is not with my model – but rather in persuading people that [the ideas expressed on this blog] represent a viable approach, particularly as I am not in a position to follow it through to a conclusion myself.

The major obstacle is that most people fail to realise that a simple model can help you to understand very complex problems. They jump to the conclusion that as because the human mind can handle very complex tasks the drive mechanism must be equally complex. They forget that the basic idea behind evolution is so simple that it can be written on the back of an envelope yet a detailed study of what has happened (given millions of years of slow changes) can be extremely complex. In my model the equivalent to evolution's million of years is an understanding of the power of significant levels of recursion – which allows the same simple mechanism to be used again and again and again and hence achieve a significant level of sophistication in the concepts the brain is processing, by taking a large number of individually very simple steps. Because of the way recursion works the mechanism is simple even if the ideas being processed are extremely complex.

I also have problem in getting my ideas over because the key features are counter-intuitive in terms of the algorithmic approach to science and mathematics – and in particular the way everyone is taught to think about computers. The Turing model of computing boils down to answering the question “Which came first, the program or the data?” with a very clear answer - “The program”. When I was working on CODIL I was saying that, for large poorly defined problems, all you had was “information” and that when the system made decisions the nearest you got to a task-specific program was to observe the dynamically generated “virtual program” by monitoring how decisions had been reached! For my pains I could not get the support and funding I needed because the establishment was tied to the idea that the Turing model was the only valid model, and was so financially successful that there couldn't be a better information processing model.

What happened in my “Eureka moment” was I realised that the information stored in the brain is also “virtual” in that it only exists in any meaningful form while the links between neurons are active. In particular any neuron might be involved in many different but related “memories” at different times. This “virtual parallel” model of the brain's working memory is so different to the “concrete words in a serial context” of natural language that it can be quite difficult to explain in natural language!

All this suggests that one of the reasons the workings of the brain are such a mystery is that, in order to understand how memories are created and used in the brain, one has to “unlearn” the basis of natural language and the scientific method as applied to individual culturally inspired tasks!!!

I am currently drafting a paper to take my model from individual links between neurons, via building complex concepts by recursive pattern matching, to making decisions (including a consideration of consciousness), the need for accelerated learning, the foundations of language, and the switch to the cultural evolution of powerful mental skills. Hopefully this will be ready in about a week's time and I will let you know when it is ready. 

1 comment:

  1. Almost immediately after drafting the above I came across the piece by Partha Mitra entitled "What’s Wrong with the Brain Activity Map Proposal" which is very relevant - as he clearly sees the danger in "Big Science" approach to brain research. (see ) and I posted the following as a comment:

    I agree with many of the points Partha makes about the inappropriateness of the “big science” approaches to brain research – but he does not go far enough in pointing out the disadvantages of such major projects. Modern science is becoming a rat race for funding and prestige and one can only justify putting vast resources into very big projects if thery are certain to work and if there is enough money “left over” to support genuine blue sky alternative approaches which deliberately look outside “the box”. What happens in practice is that anyone with original ideas who questions the establishment views is very likely to left out in the cold.

    In this context I was interested when Partha uses an analogy with physics research - mentioning topics such as thermodynamics and statistical mechanics and the study of macroscopic behaviour. My own approach to brain research started many years ago with “macroscopic” observations of how sales staff thought about sales contract in a vast sales accounting system. Initially research was directed towards building inherently user friendly information processing systems – and recently I have started to look at the relevance of what I did to brain research. The result is a model which in some ways reflects the statistical dynamic models of gases. One starts with an infinite brain filled with identical neurons and each neuron is linked to a finite number of other neurons. Electrical pulses pass through the resulting network and the key is to find a set of “rules” to control the process. Because of the random elements in the model every brain (both human and animal) will end up with different connections – even when relating to the same external phenomena – and many of the connections will change with time. If such a statistical model is correct it is clear that trying to map individual pathway in an individual brain is about as meaningless as trying to understand gas mechanics by trying to plot the movements of every molecule in an individual gas-filled container.

    The problem I face is that my current research is unfunded “blue sky” research and falls way outside the establishment box on a number of different counts (for instance “I am obviously too old to have original ideas”) – so will presumably my idea swill never be properly investigated.