Tech

By Daniel in Tech, Thoughts ⋅ November 23, 2007 ⋅  Email this post ⋅  Print this post ⋅ Post a comment

I read a great page today, detailing several ideas for how computing power - and artificial intelligence - will advance. Some of the concepts Eliezer S. Yudkowsky explores here are flat out fascinating - so much so that I've written about a few of them, if you're interested in a synopsis of some of the things he has to say.

Computing Speed

"If computing speed is doubling every two years," he states, "What happens when computers are doing the research?"

This idea is central to the article; it's beatifully simple and makes a surprising amount of sense. The premise is this: one day, we can assume computers will be capable of processing powers similar to the human brain. What if, then, there was a way to let the computers themselves research their own development? To begin with, they will - theoretically - do this at the same rate we do, given similar ability. They will take two years' subjective time, to double their own intelligence. But in doing this, they become doubly as capable, and as a result two years of their subjective time is shortened to the space of one year of ours. The process repeats, the time becoming smaller and smaller, and eventually they reach a singularity of intelligence.

This has its flaws - for a start, no computer has yet been built with the potential to further its own ability, or power. This also ignores a possible physical limit to the amount of possible computing power - though with the rise of quantum computing, these theoretical limits are being pushed back further, continuously. This continually self improving artificial intelligence, asserts Yudkowsky, whether it is desirable or not, is certainly inevitable, and the natural progression for any intelligent species.

—

To assert my own opinion here, the idea of artificial intelligence is one that I am sceptical of, but only with regards to present computers. At the present day, computers are designed to take inputs and give outputs; their reactions, if you discount the layers of abstraction which allow more and more complex things to be achieved with computers, are entirely predictable. The only feasable method I could imagine for an artifical intelligence would be one that was allowed to develop itself from scratch. Perhaps an accelerated version of the natural selection we see in nature, with a built in tendency for mutation? The type of computers within which AI will evolve, in my opinion, will be vastly different from those designed today.

Humans as Machines

Although this particular point seems far fetched at a first glance, it's actually a very interesting and compelling idea. What if, Yudkowsky asks, we develop sufficient nanotech to synthesise and replicate individual neurons? If, say, we were to contstruct a nano-robot capable of scanning each individual neuron in the human brain, replacing each one at a time with a perfect artificial copy. Certainly, we would not be aware of any change, and our brains would continue to function in exactly the same way. So at what point do we stop becoming human?

To extend this argument, when it is possible to create and replace every element of the human body, when a human is entirely replaced by nanotech, will we have discovered the secret to life itself? Or will we have simply turned a human into a computer?

—

I only hope that if such technology does become possible, which given human advancement over the past few hundred years seems infinitely likely, I will be there to see it.

Popularity: 2%

No Ratings Yet

 Loading ...