Aug
28
2011
The question we are asked the most at Thinking Solutions is what is patom theory and why is it different to the current scientific theories of the brain or the available computer models?
What is PATOM theory?
Patom theory suggests that brains work by storing, matching and using patterns. There is no "processing" - it just finds patterns and uses them. Obviously there is a process to match a pattern, but there isn't a process to analyze the patterns found through a hidden homunculus. A Patom is defined as the smallest unit a brain uses to store a pattern - hence it is a Pattern-ATOM. As a cognitive science student in the 1980s, we looked at the ways to represent information in a computer system and how to make sense of the vast amount of information available to a brain and how to analyze that information. Something becomes clear quickly when this problem is considered in light of a brain's physics.
A brain is slow, taking milliseconds for its elements - neurons - to activate. A computer is millions of times faster. For a brain to recognize a threat and take action quickly, within the order of tenths of a second, something is happening in a short path within the brain to allow such slow moving elements to generate a response by the muscular motor system. A hypothesis is that the brain has already stored a response pattern and the lead up to the response is just to recognize the threat and signal the response pattern. Of course there must be more than one response pattern for a successful evolutionary outcome, but the principle is there and there are ways to create sets of responses.
PATOM Science difference
Since working on Project Turing, named after the late great computer pioneer Alan Turing, it has become apparent that human languages need only make uses of stored patterns. By storing these patterns in a hierarchy, very simple elements can create amazingly complex constructions within a few layers. Human languages quickly become rich networks of information in context through the layers of words to phrases to clauses to sentences to stories. If you only get a layer or two, there remains unsolvable levels of ambiguity that mere mortal programmers cannot deal with. Programmers cannot follow large numbers of possible paths to success reliably, while a computer readily stores sets of patterns to enable disambiguation through discovery - the discovery of consistent patterns.
These patterns are comprised of sets, lists, lists of sets and sets of lists.
Considering the structure of neurons, the triggering of a neuron looks a lot like the identification of a set, while a list looks like a set of neurons firing in sequence.
For languages, consistent patterns include finding the correct list of letters in a word (or the correct list of phonemes in a spoken word), followed by the identification of the correct list of words within a phrase and then the correct list of phrases within a clause. Sentences are merely lists of clauses within a punctuated group. Stories are simply sets of clauses within a single context.
The work of David Hume back in the 1700s rings loudly with his claim that, in my terms, brains understand through experience and recognize (a) similarity, (b) contiguity and (c) cause and effect. We'll come back to this another day, but similarity is just the recognition of common elements, contiguity the recognition of a sequence of element and cause/effect is a sequence of clauses. Paraphrasing further, brains store, match and use lists and sets. These patterns are stored in a hierarchy. By verifying matches in reverse (by checking that the best-fit is consistent with observation) the brain can recognize patterns that are inadequate on their own for identification, but which upon confirmation prove obvious.
What is really different between PATOMs and computers?
Here's a thought: computers store binary code while brains store patterns. What this means is that a computer encodes information and stores it generically. A brain recognises patterns and stores them once. ONCE. If all patterns are stored once, similarity comes from the fact that the same elements are referenced. There is no processing needed to find the reference. The reference is the same, so it is found. Rhyming is noticed by the brain because the same pieces are re-used. Words formed by joining other words are noticed because the other words are embedded - the French word 'toilette' reminds an English speaker of the word 'toilet' because the letters are within the French word, not because having looked up the word we further process the pieces. Indeed, the words 'to', 'toil' and 'let' are within toilette.
By storing pieces of experience in a hierarchy, the comprising elements are found along the way when they are a part of the larger pattern.
This difference is fundamental in the approach of modelling brains. By building machines that store patterns once, any information matched is immediately expanded and by expanding information, additional alternatives are immediately identified.
There is no processing, just pattern matching.
Until next time, when we will look further into the science and the new capabilities enabled.
John Ball