. . . I spent 18 months in a neuroscience lab. This might be useful background. James's intutition is spot on. Part of what I leanred is that the mammalian brain is at its base level an hypthosis generating machine. Here's a passage illustrating that:
(Gary) Lynch was introduced to Rick Granger, a computer scientist recently arrived in
“Flabbergasted hardly begins to describe my reaction to this,” Lynch said. “I knew from some obscure course at Princeton that Piaget was total crap, and Rick's model had no neural in it, at least as seen on this particular planet. ‘Son,’ I says, ‘no idea what goes on back there at MIT and Yale, but you needs to get yourself into a man's world. Let's put those empirically derived learning rules into a simulation.’ ”
Granger went to work building a computer model of a learning system incorporating what Lynch had learned about LTP, (long term potentiation, a theorized process by which memories are made) in particular the timing rules dictated by the interactions of the AMPA and NMDA receptors and theta rhythms. The simulation used the random access storage system of the olfactory cortex as its base model. Granger ran multiple simulations on the model. He installed a computer terminal in Lynch’s office, with very precise rules about which keys Lynch could touch and which he could not, so that Lynch could examine the results of the simulations.
“One night, I am sitting in my trailer across from the university credit union, punching away on the machine . . . . and a network using LTP rules starts making categories - and then categories of categories,” Lynch said. “There's a moment to live for.”
It was astonishing. The machine – with heavy 2001: A Space Odyssey HAL implications about exactly who was really in charge here - took over. Or, more precisely, Granger’s simulated network took over, and began building an utterly new system. The system, remarkably, was self-organizing. It was downright spooky.
“If you take the olfactory cortex and model it biologically and realistically and you use LTP rules, the thing automatically makes hierarchical categories,” Lynch said. “In other words, it goes, animal, bird, robin in three successive iterations without us telling it ever to do that. It just does that. The capacity of the thing is just dumbfounding. . . . It is a magnificent machine for assembling disparate things in the environment into categories and treating them as wholes and stacking categories on top of each other. It’s astounding.
“Fundamentally built into the circuitry of sensory type cortex, again with random connections, built into it is the way we feel the world. You don’t walk out onto the street and look up and see a car bearing down on you and say, `Oh, a Chevrolet Impala.’ You say, `Car.’ And then if you want more information, you go down the hierarchy, you take another look and another look and another look and you just go down the hierarchy using this first observation. You go through your life in thought using broad concepts and then you go down the hierarchy.”
Lynch was fond of saying that if you weren’t being surprised, if you weren’t learning completely unanticipated things, you weren’t making discoveries, you weren’t really doing science. The results of the neural simulations were utterly unexpected, or even imagined. Granger and Lynch had stumbled upon what seemed to be fundamental insights into a mammalian learning process that reshaped the world even as it was taking it in. If the model was correct, categorization took place inside the brain unconsciously at the neural cellular level. LTP sculpted the world even as it was being experienced. Such a system ought to multiply the storage capacity of the cortex many times by the mere fact of making the categories. H