Ray Kurzweil – SXSW.
Ray Kurzweil is a legend.
He invented OCR, text-to-speech synthesis, speech recognition and loads of other stuff. He popularised the notion of the "singularity", the point in time when computers become more intelligent than humans and all bets are off. He's probably the world's best know futurist.
I like that he takes a systematic approach to futurism. He doesn't seem to act on intuition like a lot of futurists do. His predictions often seem over optimistic, but end up uncannily accurate as he looks at hard numbers.
Over the years a lot of philosophers and psychologists have said we'll never get proper "human style" artificial intelligence because there's something "other" about it, something that is too complex for any computer to simulate. I remember when I was studying cognitive science in the early nineties that it was still thought by many that a chess computer could never beat the best in the world because there was some ill-defined humanness required. Kurzweil predicted that this would in fact happen in 1996, and it did. He knew it would because of the relentlessness of Moore's law.
Another great example was the sampling of the human genome. Having spent something like 10 years to do 10% of it, researchers thought it would take decades to finish. Kurzweil, taking Moore's law into account, said it would take another year or so, and it did.
So a recurring theme was this idea that people are over pessimistic about AI and technology in general because they fail to take into account the sheer power of exponential growth. Worries about being able to create the peculiarities of intelligence get blasted away by the massive increase in things like processor speed, bandwidth, materials etc.
The recent success of IBM's Jeopardy playing supercomputer, Watson shows just this.
It beat the World's two best human Jeopardy players. According to Kurzweil this isn't due to particularly new AI techniques, but more due to raw processing power. It gained it's knowledge by reading the internet, including the whole of wikipedia. It then selected articles via a fairly simple probabilistic algorithm. It worked because it had the speed to process the data in real time. Again it's down to Moore's law. The interesting thing that Kurzweil highlight though was that this is pretty much how we work in our own brains anyway. We are fundamentally probabilistic machines with a lot of processing power.
If you plot our intelligence on a graph against machines, they pass us in 2045. So that's it then, after that point we have no way of knowing what will happen next. Machines might not care to tell us what they are up to, and we may well not understand it if they did.
On the bright side, there will be no reason why we won't be augmenting our own brains with billions of extra neurons, so we may end up having a more symbiotic relationship with machines than some people fear. And at least we'll be able to back it up.