Several years ago I read a book called the 'Singularity is Near' by Ray Kurzweil, which both shocked and amazed me. Kurzweil, a transhumanist, predicts that well within our lifetimes, both true AI and other far fetching phenomena such as immortality through reverse biomedical engineering, as well as downloading our consciousness into a fully realistic shared computer simulation, through the law of accelerating change, otherwise known as Moore's Law. At the time I was heavily engrossed in philosophy of AI, cognition and consciousness; and the book left me with the belief that much of Kurzweil's predictions could be true, because I saw nothing philosophically impossible about such rapidly evolving phenomena, and his arguments were so convincing.
Now I'm not saying that this guy is definitely full of ####, nor that he is a nutter (though he probably is), but something smelt fishy about his predictions. See Kurzweil, as well as having some ridiculous number of degrees (I think its about 20) and having a history of solid predictions about the future, is also a very successful entrepreneur. One of his ventures is a range of medicinal long life products, which he sells at a pretty steep price. Could he have written three best sellers telling people they could live to see immortality through nano-technological reverse bio-engineering and hinting at his own endeavours to stay alive through a range of daily supplements? Maybe. Maybe not. We couldn't rule out his ideas purely based on that possibility.
Yesterday, I ended up re-igniting the debate with some friends over a few drinks, and I found out a few things I had probably been over-looking in my previous ignorance. I would consider, both the guys I were chatting to far better computer and AI experts than I am.
First it was pointed out that the idea of the law of accelerating returns, i.e. that CPU power doubles every 18 months, leading to exponential growth and huge improvements in surprisingly short spaces of time, was grinding to a halt in our current state of technological growth. Apparently they just can't fit any more transistors (or whatever they are called) on the circuit boards without overheating. This burst the AI in 50 years bubble to some extent for me, though the debate got far more philosophical than just technological possibilities, i.e. What is intelligence? Do we need embodiment of an agent? Turing tests, Etc.
I'm not going to go into all of that (I'm done with philosophy), but if we give a loose definition of true AI being functionally equivalent to an average human being in all respects, with functions isolated, so that you couldn't tell the difference between the two; is it technologically possible in 50 years? This is similar to what Steve Harnad calls T3 Turing machines I believe. So there might be a language Turing machine, the classic test where an AI converses with you over a text interface to convince you its real. Another one might be an AI with visual processing that can recognize patterns and objects as-well as us. Another might be to produce a piece of music or art that evoked genuine emotion as creative humans can do, and so on and so forth.
I think it might be for many of those tests to be passed soon. The original Turing test for example, I would predict a pass in 20-30 years and perhaps similar for visual and audio pattern recognition. Producing sophisticated music and art maybe 10 or so years after that. But true intelligence, links all of those functions together, arguably in an embodied system, to produce one multimodal integrated intelligent entity. Whilst we might be able to produce a load of functional t3 Turing machines, since the human brain is about a billion times more sophisticated than the biggest super-computers we have now; linking the functions together in one AI to replicate true intelligence will be impossible until we make massive break throughs in processing capacity. One possibility could be some giant parallel processing cloud computer, made possible through the internet or similar, but it would be difficult for us to conceive of that as one entity. Let alone an intelligent one.
However 50 years really is a long time so we agreed it wasn't impossible. Think how far we have come in the past 50 years and its surely only going to speed up. There are so many new possibilities in processing technologies on the horizon; 3D chips, DNA processing, nano-technology, quantum computing. I just don't think processing power will be an issue in 50 years. Its going to be more of an issue of understanding how to make a truly intelligent system and then some how programming it so that it is sophisticated enough to pass for intelligence. That is where Kurzweil may be over optimistic. The task is absurdly complex.
What about immortality through reverse bioengineering. We quickly dismissed this idea. Ok so we might be able to slow down ageing processes through eating the perfect balance of nutrients through supplements. We might be able to develop antibodies and medicines to stop us getting diseases. We might even be able to build nano-bots that live in our blood stream protecting us from any internal threat and communicating to act as swarm intelligence (another one of Kurzweil' excentric ideas). But sooner or later something will get us. Be that a rapidly mutating virus that out does any nano-bot immune system (natural selection always finds a way), or by a car crash, a heart attack or a million other things.
So.... Can we live forever? No its nonesense. Can true AI happen in 50 years? Maybe. But there isn't enough evidence right now to suggest it will happen. And what about one of Kurzweil's other ideas... Living in the Matrix by plugging us all up to a giant supercomputer that stimulates our senses to produce a shared experience that is indistinguishable to real life... I'm not even going to go there...
