Wednesday, November 16, 2011

AI, Are We There Yet?


By: Alon Cohen

Wow how times fly. I remember a discussion I had about 28 years ago as if it was yesterday. At the time, I was working on few computer programs that seem to need human intelligence. One was an automated device testing program that could tell a technician which board to replace based on anomalies in the way that device behaved. The second was a learning algorithm that had to play Tic-Tac- Toe by figuring out the rules of the game by itself (yes after I saw the movie war games in 1983)) and it triggered in me the thinking about what would it take to make a computer more human.
I was talking to few of my friends discussing what would it take to make a computer creative. My idea at that time was that scientists would probably be able to take a human Neuron (or few of them) and place them in a chip, and use that as an intuition co-processor.
Well it seems like the MIT scientists just made it happen. The MIT people did not really took a brain tissue but found a way to emulate the way a Neuron behaves using analog circuits on a CMOS chip. I guess it took more than few years to figure that one out. However, as it stands I can now envision a reality just as the one depicted in Asimov’s books where business entities (like US robotics) will own an artificial brain that will surpass any human or supper (digital) computer in existence, which will help them invent and solve problems not solvable by humans like time travel or teleportation.
As it stands, the human brain has 100,000,000,000 Neurons and about 100,000,000,000,000 Synapse connecting them. The cerebral cortex is the outermost part of the brain. It plays a key role in memory, attention, perceptual awareness, thought, language, and consciousness. The number of neurons in the cortex is estimated to be 11,000,000,000 so about 11,000,000,000,000 Synapses. Synapses are those specialized junctions through which neurons signal to each other.

Surprisingly it only took 400 transistors to create that artificial Synapse at MIT. Imagine a quad core Itanium chip with 2,000,000,000 transistors, take that level of technology and you can create a brain with 5,000,000 Synapses or an equivalent of 5,000 neurons give or take or hmm as smart as Pond Snail.
Since our brain has about 11,000,000,000,000 Synapse in our cortex, we need just about 2.2 Million more of those quad core chips to emulate the human cortex. Sounds scary large number but the truth is that it is not that far if you are an optimist.
If you apply Moore’s law to those numbers you get that in 30-40 years (2^21 = 2M) give or take we will have the capacity to replicate a human cortex that can work thousands of times faster than a human can.
It is not that far I can tell you that, specifically when we live in internet speeds and each New Year end before you even noticed it started. Now, way before those 30 years, say 15 years from now we will be able to compose an artificial brain equivalent in capacity to a dog cortex. I guess all we will need to do at that stage is hope that this technology will not bite us. If it will bite, it will help us realize the corrections we need to make, just in time, so that the first artificial human-level brains will do the right thing.