Did Plato discover Neural Networks?
Here's a bit of history, I'm dredging up an argument that filled my final year at Uni. Basically, I actually find this interesting, and no-one else seems to, so by putting it out on the net, maybe I'll get some comments. Apologies for the need for joining the dots of this together, but if I have to explain it more than this then either I would be very rich or very stupid, and I'm definitely not the former!
My final year thesis was supposed to be about a face recognition system that parsed language descriptions and searched a database of face elements (facets?? ;) to produce a match. Dull, dull, dull, dull, dull. I decided to do a half-arsed program to do what was required, and focus my attention on the write-up, which turned into a discussion of the Philosophy of form and recognition.
Turns out that going back to Greek times gives a pretty good insight into the problem, perhaps because the thinking was more pure and less polluted by other thought. To paraphrase Newton "If I cannot see the ground, it is because I stand on the shoulders of giants". Plato and Aristotle had a fundemental difference in how they saw the 'Form'. Aristotle believed that the archetypal 'horse', for example, existed in reality. Plato believed the ideal form only exists in the mind (actually he doesn't say exists, but can be seen). In other words, the concept of the 'horse', the form, is reflected in the real horses that exist in reality. We can look at a new animal an classify it as a horse because it fits the form of a horse that we carry in our mind.
This is very interesting because it means that the form of the horse is matched in our mind with the ideal form or paradigm symbol, but that form is not concrete. Plato does not really say it, but I contend that we cannot see the ideal form in our minds, but we are nevertheless aware of it, and thus, if we want to create an image or sculpture of a horse, we are able to create one based on the re-cognition (sometimes just breaking a word down gives an insight) of the horses we have seen in the past. I think Plato would have said that the form we create is not the ideal form but an imperfect reflection of it, however it would also be true that the more horses we see and the more detail we see them in, then the closer to the ideal form we can get. Thus a master builder can build a better house or a master sculptor (of horses) can scuplt a better horse.
There have been articles I have read subsequently that talk about breaking down objects into facets, attributes that all things of this type have in common, but tie these things to later philosophy, rather than the basic Plato/Aristotle divide.
Back in 87 I looked at some of the physiology work done by Lettwin et al on the optic nerves and began to get an inkling of how the brain maps fairly closely to the Platonic philosophy. The pattern of activation of neurons within the brain strengthens with repeated exposure, connections between neurons reinforcing each other and producing the non-concrete form.
Concepts are formed out of the simultaneous activation of many different activation paths, effectively producing recognition of simple forms and then combining these into recognition of more complex ones. Lettwins experiments showed patterns of activation that represented circles, squares and triangles, but these would have been further combined to represent a fly, for example.
The key to intelligence is creativity and here is where it starts to get a bit theoretical ( ;-)
What if connections form between proximate neurons when activations happen contemporaneously? This would create links between forms that originally had no connection.
My contention was that the level of adrenalin and other chemicals in the 'environment' of the brain, facilitate or inhibit the formation of these connections, thus allowing traumatic experiences, such as fire, hot, pain to be linked (learned) in the mind very quickly. It is interesting that in our behaviour we will often perform some self-reinforcement in these cases, touching the thing that was hot one more time - just to make sure.
Back in 87 my conclusion was that in order to build a 'brain' that was intelligent you would need to have a process(or) for every neuron, to handle it's interactions with the other neurons and it would also need a substantial amount of 'state' information. It would also need an 'emotional controller/facilitator' to broadcast to the network.
I'm very interested in the new IBM processor that might be a step along this road. Once I have read a bit more about it, I'll post some more. Or if anyone reads this blog (and knows more than I do) let me know!