Analysis of IT news

Tuesday, February 17, 2009

On Singularity

There's been renewed interest lately in so-called Singularity. Singularity is the moment when computers will be as powerful as a human brain. When this day arrives computers will be intellectually superior to humans.

I'll be upfront: I don't buy it - not in the nearby future at least. As a matter of fact, Singularity reminds me of all the buzz around Artificial Intelligence from the '70s and '80s.

I don't buy it not because it sounds unbelievable. If in 1950 someone had predicted how powerful computers would be today and how they would be used, he would have sounded crazy.

I don't believe in Singularity as it's presented to us because the model used by its advocates is fundamentally wrong. Their argument is the following: nowadays we can simulate in a computer how several neurons interact with each other. Following Moore's law, computer power grows exponentially so we should eventually - and sooner than we think - come up with computers that can simulate an entire human brain. Some advocates believe Singularity will be reached withing the next couple of decades.

I however see one flaw in this theory. Put a few neurons together and you indeed have basically an electronic circuit easy to simulate. But when you put billions of neurons together - enough to make a human brain - something weird happens: a conscience emerges. Call it a soul, a spirit or whatever you want but no one can deny there is *something* that appears and which has emotions - something not taken into account at all by any computer model. Is the conscience a side effect of electricity? Of a chemical reaction? Something else? Nobody right now has *any* clue about what really conscience is, why and how it appears.

I'll use a parallel: Newton's laws of motion have been very useful to humanity - they are a cornerstone of modern mechanic. But Newton's model also says we can reach any speed we want provided we provide the necessary energy. Except that reality is more complex than what Newton imagined, and past roughly 100,000 km/s his model doesn't work anymore. That's when we need to switch to another model to predict movement accurately: Einstein's relativity theory.

Well, something tells me that the neuron model we know works for a small amount of neurons, but past a certain critical mass it's not valid anymore and we need a new model. Until then trying to predict when Singularity happens is just ludicrous.

I'll end with another story. Back in the 50's, Jon von Neumann thought that computers would help us not just predict but *control* the weather. His argument was that any system contains points of unstable equilibrium, and that you can control the states of that system by acting at these points. Imagine a ball on the top of a hill that you can make go in any direction you want by giving an appropriate small nudge. Von Neumann's thinking was that it was just about finding out the unstable equilibrium points of the atmosphere and we would know how to control the weather. It was "just" a matter of number crunching. Except that von Neumann's model was flawed. He indeed didn't realize that *all* the points of the atmosphere are unstable equilibrium points (the famous butterfly effect). Once again reality turned out to be much more complex than what he imagined. Half a century later computers are more powerful than von Neumann could have ever imagined but we still can't reliably predict the weather a week in advance. Just a matter of number crunching, uh?

And guess what? "It's just a matter of number crunching" is exactly what Singularity advocates are telling us. Could reality be more complex than what they imagine?


Post a Comment

<< Home