Knowledge is pattern recognition. And pattern recognition is a process of forming associations. We have different ways of forming associations; one such process I've talked about from time to time is based on Boltzmann mechanisms. The idea here is that the different points on the graph, pictured above, represent different possible patterns of association, and the best patterns are those with the lowest potential energy (the least conflict, the greatest consistence...), as represented by pits. As we weigh one possibility and then another then another, it's like rolling a marble on the graph. Eventually we settle on a low point. But - notice - it might be a shallow low point. Then we actually have to expend energy and shake up our thinking to escape these 'local minima' (this is why I say that learning is practice and reflection - and this is the bit about learning that the 'core knowledge' types miss completely). Eventually, we get to the deepest point - the most stable configuration. The best pattern. (Then tomorrow, we have to do it all again, because in a complex environment the deepest pit is a stange attractor, which means it's always moving).