Sunday, January 6, 2008

For all you singularity buffs out there...

There are two ways to look at this. The Guardian's way of looking at it is to say that:
your computer at home doesn't even come close to matching the power of half a mouse brain: researchers at IBM and the University of Nevada have been using IBM's BlueGene L supercomputer - which contains 4,096 processors, each using 256MB of RAM - and succeeded in simulating a small fraction of the power of just half a mouse brain

Another way of looking at it is that in 1990, one of the fastest supercomputers was the Cray Y-MP 8/8-64. According to a history of supercomputers by the European Center for Medium Range Weather Forecasts:
This system had 8 cpus with a cycle time of 8.5 nanoseconds (166 MHz) , and 512 Mbytes of memory.

Right now I can buy 512 Mbytes of memory on a postage-stamp sized chip on my cell phone, and 166MHZ would have been fast for a desktop a decade ago.

The Guardian article quotes researchers from the Ecole Polytechnique Federale de Lausanne to the effect that the human brain has 100 billion neurons, whereas the mouse brain has 8 million.

Of course, the argument can be made that sooner or later Moore's Law is going to bump up against the far firmer Heisenberg Uncertainty Principle at some point. Since I don't know anything in particular about quantum computing, I won't make any predictions one way or the other on that.

But assuming everything stays on track, it's reasonable to assume that a desktop will be able to simulate a mouse brain full time in 10-15 years or so, and the power of exponential doubling being what it is, a full-time simulation of a human brain might be as near as 20-30 years in the future, at least on a supercomputer.

The Guardian quotes the same Lausanne researchers as saying that they cannot predict if such a model would develop consciousness. I can't say whether the model of the human brain could in fact be simulated. Being able to update 100 billion neurons a second is different from actually programming them the way the brain works, which I believe remains quite mysterious. But if it could, I think it's ridiculous to assume that the outcome would be any less obviously conscious to us than the brain of any other human. That is, from a Turingesque POV, it would be as conscious as you could prove anyone else to be.

All I want to know is when Google will start allowing us to book our personality backups in advance. Then finally my opportunity to skydive without a parachute into an active volcano will come true... ;)

No comments: