|
AI
Aug 18, 2005 19:46:24 GMT -5
Post by Dragon*of*Heaven on Aug 18, 2005 19:46:24 GMT -5
Hello all. Ok recently I have become pritty interested in AI. Do any of you guys know alot about this technology or others? Im realy curious.
|
|
|
AI
Aug 19, 2005 15:50:51 GMT -5
Post by vertigo on Aug 19, 2005 15:50:51 GMT -5
I don't know the technical details, but I do have an opinion on it. Usually AI is regarded as something mysterious or miraculous. AI techniques are used to solve problems, for instance. Unfortunately there's a pit to be avoided. When an AI technique like a genetic algorithm is used, the result is only as good as the metric you put in. This means that while it might seem to you that it found a really good solution, it is not necessarily so. You might realise something about the metrics, adjust them, and then find a much better solution. Let me give an example. Most people use keyboards which have the QWERTY layout. Now a long time ago Dr August Dvorak decided to design a new keyboard because QWERTY wasn't efficient. For instance, type the word 'reference', it is all on the left hand and isn't any good. Now, Dr Dvorak choose some metrics about how our hands work, and designed a new keyboard layout. His metrics were good, let me give an example. If I number the fingers from index finger to little finger as 1 to 4, pressing 3-2-1 is much easier than 1-2-3. Now if you look at a letter combination like 'th', 'th' is very common in english, but 'ht' isn't. It's one-way, so the proper place to put it would be 2-1, which is the best position for a one-way digraph. Anyway, he used metrics like that and letter frequencies to design a new layout, called the Dvorak Simplified Keyboard. He did this a long time ago (about 70 years ago). Now more recently there have been people who didn't understand the metrics he used, and thought they could do better. This is why you find layouts like ARENSITO, which are actually worse than Dvorak's layout. Now, if a human could come up with a good layout, an AI should be able to come up with a great layout, surely. I mean, if a computer can beat Kasparov in chess, it can surely beat Dvorak in designing keyboards. Well, here you can read about somebody who tried. He used a genetic algorithm to 'evolve' a keyboard layout. A genetic algorithm mimics evolution, but I don't know the technical details. I know parents pass genetic code to children, mutations occur, etc, much like animals evolve. Now what he didn't realise is that the result is only as good as the metric. So when he had finally produced a layout he was satisfied with, he decided that he would compare it to the Dvorak layout, to see which was better. He took real keyboarding data, the time it took for key combinations to be pressed, and using that information compared the two layouts, to see which would be faster. It turned out that Dvorak couldn't be beaten. The reason for that is because the metrics used to design the Dvorak layout were better than the metrics he used. His fancy AI methods didn't mean a thing. So when you think of a robot being operated by an AI process, don't think of it as something mysterious. Remember that a human provided the metrics, and better metrics mean better operation. Given a set of metrics, AI techniques are superior at finding an optimal method, but it is easy to forget that the method is no better than the metrics.
|
|
|
AI
Aug 21, 2005 1:02:57 GMT -5
Post by Dragon*of*Heaven on Aug 21, 2005 1:02:57 GMT -5
Thank you vertigo, but another question if I may. I always AI was the state at which a computer mimics the human brain. I always thought of it as the state at which a computer is able to break the binding of its programing. For instance, if I build a robot and I progam him to pick up a glass he will do so. And no matter how good the programing is anything it dose even that which mimics the human body or a human trait is just another bit of programing. If that is true then isnt actual AI the ability to jump those bonds so to speek? Am I right or have I got onto the wrong track of ideas?
|
|
|
AI
Aug 21, 2005 6:47:35 GMT -5
Post by vertigo on Aug 21, 2005 6:47:35 GMT -5
Yes, an AI will slowly train itself. You give it metrics which determine how good an action is. You might say that the degree to which the task was completed is important, that the time it takes to complete is important, also the amount of power the robot uses, the noise produced, or whatever.
The robot might have a preprogrammed default state or it might start from scratch. It will try various actions, trying to improve. The more it does it, supposedly the better it becomes. This is not always true. Sometimes it can get into a position where it can't get out because it's method is vastly wrong and no mutations would prove beneficial. Some intervention is require by the programmer to track back to a former state, and perhaps to adjust the metric so it doesn't get stuck again.
Anyway, SolidSquid will probably have my head because I don't know the details of evolution, so what I say could be completely untrue.
|
|
|
AI
Aug 21, 2005 12:41:03 GMT -5
Post by solidsquid on Aug 21, 2005 12:41:03 GMT -5
AI is out of my area of knowledge. So the "evolution" (meaning change over time) of such wouldn't really jive with my knowledge of biological evolution (the scientific theory of evolution). So I can't really say much on the subject of artificial intelligence without it being an undirected, laymanistic speculation.
|
|
|
AI
Aug 21, 2005 12:59:02 GMT -5
Post by vertigo on Aug 21, 2005 12:59:02 GMT -5
What knowledge I have of AI is from a computer programming perspective, and AI methods are overrated in my opinion. I would always favour a traditional approach to solving problems.
|
|
|
AI
Aug 22, 2005 23:36:48 GMT -5
Post by Dragon*of*Heaven on Aug 22, 2005 23:36:48 GMT -5
Thank you both for your answers. You were both very helpful thank you. ~^!^~
|
|