Tuesday, 28 October 2008
A long time ago I read Shadows of the Mind by Roger Penrose. It was pretty tough going, but the one thing I really remember now is the bit about the microtubules inside neurons.
It was interesting because people often assume that neurons are the basic computational unit of the brain.
For example, when you count the number of neurons inside a brain (about 100 billion) and the speed at which they operate (about 1000 operations per second) you can estimate a maximum processing capacity of the brain of about 10^14 operations per second.
Then if you look at the number of transistors on current CPUs (about 410 million on an Intel Core 2 Duo), and their speed of operation, and take Moore's Law into account, you can get quite optimistic estimates for how long it will take for AI to match human intelligence.
Of course everyone knows that neurons aren't that similar to transistors, and its probably much more sophisticated thinking than that which leads people like Ray Kurzweil to estimate that machines will achieve human-level artificial intelligence by 2029, but I'm sure that comparing neuron counts to transistor counts must be a factor.
So one of the most interesting things I learnt from Penrose's book is that neurons, like most cells, are made up of thousands of microtubules. And Penrose believes that the microtubules may be the basic computational unit of the brain, not the neurons. Because there are thousands more of them, and they operate much faster than neurons, we may have underestimated the processing power of the brain by a factor of up to 10,000,000,000.
This theory clearly has its detractors, but if you look for research on information processing within microtubules, there is plenty of it.
So maybe things like Eidolon AI are a long way off yet.