Ontonix: Model-free methods – a new frontier of science
Tuesday, 13 December, 2011 2 Comments
ORIGINALLY PUBLISHED in September (and again in November 2010): I thought it worthwhile revisiting this piece by Dr Jacek Marczyk, (Founder & CTO of Ontonix Srl.) as it tied-in with some of my recent thoughts regarding “complexity paralysis”, procrastination and belief systems.
When we make decisions or when we think our brain does not use any equations or math models. Our behaviour is the “fruit” of certain hard-wired instincts and experience that is acquired during our lives and stored as patterns (or attractors). We sort of “feel the answer” to problems no matter how complex they may seem but without actually computing the answer. How can that be? How can a person (not to mention an animal) who has no clue of mathematics still be capable of performing fantastically complex functions? Why doesn’t a brain, with its immense memory and computational power, store some basic equations and formulae and use them when we need to make a decision? Theoretically this could be perfectly feasible. One could learn equations and techniques and store them in memory for better and more sophisticated decision-making. We all know that in reality things don’t work like that. So how do they work? What mechanisms does a brain use if it is not math models? In reality the brain uses model-free methods. In Nature there is nobody to architecture a model for you.
There is no mathematics in Nature. Mathematics and math models are an artificial invention of man. Nature doesn’t need to resort to equations or other analytical artefacts. These have been invented by man but this doesn’t mean that they really do exist. As Heisenberg put it, what we see is not Nature but Nature exposed to our way of questioning her. If we discover that “F = M * a” that doesn’t mean that Nature actually computes this relationship each time a mass is accelerated. The relationship simply holds (until somebody disproves it).
As a person gains experience, the rules (links) become stronger but, as new information is added, they can also become more fuzzy. This is the main difference between a teenager and an adult. For young people – who have very few data points on which to build the links – the rules are crisp (through two data point a straight line passes, while it is difficult for 1000 points to form a straight line – they will more probably form something that looks like a cigar). This is why many adults don’t see the world as black or white and why they tend to ponder their answers to questions. Again, the point is that there is no math model here. Just example-based learning which produces sets of inter-related Fuzzy Cognitive Maps that are stored in our memory. Clearly, one may envisage attaching a measure of complexity to each such map.
OntoSpace, our flagship product, functions in a similar manner. It doesn’t employ math models in order to establish relationships between the parameters of a system or a process. Essentially, it emulates the functioning of the human brain.