Artificial Intelligence - What is it anyway?

A few years ago I started getting really interested in the concept of AI, a genuine, thinking, learning, created intelligence. I started playing around with chess engines to see just how a computer could really get as good as a human Grandmaster, and tried to program a simple Connect 4 (or 4-in-a-row thanks to some inane copyright) engine. However, the more I got into the grassroots of how they were set up and how they worked, the more it dawned on me that they simply weren't "intelligence". Don't get me wrong, it is a genuinely brilliant feat that they do so well against the top players, but when we consider that an average chess engine calculates hundreds of thousands of positions a second compared to a top Grandmaster's 1 to 2, it seems a grossly inefficient attempt to play good chess. Rybka is doing rather better, performing to a slightly higher level than the top "mainstream" chess engines and calculating at only 40,000 positions a second on my humble machine. But of course, whilst its performance is clearly improved, it is still in a different league to a grandmaster at pruning out bad moves when thinking - all this at something that a computer should be able to do exceptionally, after all, to the computer this is just glorified number-crunching. I cannot see how this can be defined as Artificial Intelligence.

There are of course the famous "George" and "ALICE" AI humans, who keep trying to be so convincing a conversationalist that a human cannot tell whether they are speaking to. Of course they are still nowhere near that goal, but they are capable of picking up inflections in a human tone of voice and moods - even picking up standard language used in unusual context. Great steps forward, but still fundamentally limited; if the programmers haven't told it what to look for, how to look for it and how to respond, it is completely stuck. It won't come to any conclusion by itself. So even if it does manage to fool a human to thinking they are talking to another human, again I can't see how it can truly be considered AI.

There are good progresses being made in real-time strategy computer games. If you start to win with set strategy types and certain units the computer AI will learn the way you play, and adapt to defeat your strategies (and you thought the profile setups were just to allow multiple users to keep their settings). This is without any doubt a self adjusting engine that learns to adapt to its surroundings, in this case direct competition. However, again, it can only adapt within the confines of pre-programmed parameters; it has a mathematical understanding of unit strength and a spatial understanding of unit tactics, so it is hard to consider it AI. To draw a comparison, a human can pre-empt what a computer might do based on empathetic psychology - what would the human do in the computer's position. A human can observe something it didn't know could happen, and adjust accordingly based on completely unrelated human experiences doing other things - a computer program is still limited to its programming.

So what then would properly consitute AI? Technically of course it could "randomly" adapt, but this is either limited to programming what it is allowed to change (which defeats the AI goal) or having it adapt truly randomly, and allow unsuccessful iterations to die -- sort of Darwinian. This is likely to be unfeasible because without something to guide it the randomness of the changes are likely to give an almost infinite chance of the change being detrimental to the program as a whole.

Ideally, true AI would update its understanding, increase the number of ways it can adapt to situations by drawing off successful reponses to other areas. It would have some level of parallel logical processing different types of information. True AI wouldn't just learn that it did something wrong, it would create a plausible solution based on some concept of artifical common sense. At the moment I don't think anyone has managed to theorise how this is possible, let alone program it, but I suspect it will lie in gaining some greater understanding of the human mind. Until then, I suspect "I, Robot" and the dreams of Isaac Asimov and millions of Hollywood Sci-Fi fans will have to wait a long long time.