Saturday 5 July 2008

The Race for the Brain

From Rupert Goodwins at ZDNet:
Meanwhile, other intensive data crunching is cogitating away with IBM's Blue Brain project, where the team has built "a rat-scale cortical model (55 million neurons, 442 billion synapses) in 8TB memory of a 32,768-processor BlueGene/L" That's a long way from the ten billion or so neurons of the human brain – 180 times smaller, in fact – but in Moore's Law terms, that's around ten years. And the researchers are most definitely looking ahead...
We may have missed the starting gun, but the race to produce a full working model of the human brain has most certainly started.
No comment necessary.

Ok, one... a person in a pesistent vegetative state has the same amount of raw processing power available. The problem has never been getting enough neurons and synapses: it's been in how to arrange them, and how to get them to change themselves in a systematic and meaningful way.

If I can make an analogy... we have invented oil paints, canvas and brushes, and in sufficient quantities. We have not yet painted the Mona Lisa.

2 comments:

mythusmage said...

Actually, I rather doubt we've even advanced as far as a single neuron. The brain, once you've had a good look at it, isn't equivalent to a computer, it is equivalent to a network. In the more advanced cases, an internet.

A neuron is, in and of itself, a computer. Furthermore, a neuron operats using principles our computers don't, and which we wot not what of. We don't even know how an ordinary cell processes information. A neuron is even more advanced. Just remember, a computer has trouble learning how to recognize a single face, a neuron does it as a matter of course.

At this point in time we are proceeding under false assumptions. Only when we discard our erroneous thinking and learn how neurons, glials, and astrocytes, and the brains they form a part of will we make any real progress in the IA field.

Anonymous said...

You know, I still not sure that this is the right way to go... MI is something which is going to happen at some time, but is trying to reverse-engineer a biological brain the best way to do this?

Biological brains work in ways which we simply do not understand... The complexities of the neural interconnections are just so far beyond our comprehension at this point that we're grasping at straws here.

Years back when MI was even more in its infancy than it is now, my thesis for my AI degree was on neural networks. Yes, I made a neural network that could play noughts and crosses, but the biggest lesson I learned from the process is that MI can be different... Why shackle it with our limited understanding of how an organic brain works?