Thread: Advanced AI

But that's not a learning algorithm then, is it?

The definition of AI is rather vague.

I think people initially thought that a computer winning a good human player in chess would be a major breakthrough in the field of AI. Then it turned out that you get a stronger and faster computer player if you just let the computer do what it is best at - mindless number crunching.

As to your idea, how are you planning to do that?

You would only have to
1 select a piece from your set at random,
2 make a legal move in some random direction

How about
1. select a random legal move (consulting with a database, though)
2. store the position and the move in a database: if this move leads to a major worsening of the position (next moves or a number of moves later), don't use it again in the same situation; if it improves the position, always make this move in this position (as long as the opponent doesn't devise a winning strategy).

Now this might give it an appearance that the AI is actually learning.

(What if you let two AIs play each other like that - would they become exceptional players? )

Possibly, but since they're starting at random, the number of possible positions is so absurdly high that it's not feasible to store them.

I agree with the bee: when does a learning "algorithm" stop being just an algorithm and start becoming intelligent? Our notion of AI is too vague and undefined. I believe it stems from our belligerent miscategorization of it as being "artificial"; in contrast, when an algorithm uses calculus techniques to solve a system of equations, is this deemed to be "artificial mathematics"?

Originally Posted by 691175002

An AI that simply replies with an identical canned response based on the positions of the pieces cannot be called intelligent, and would probably not be very effective either.

My suggestion was not quite serious, but if it would pick from a collection of more or less equal moves, and the evaluation of positions would change over time?

I wonder how that would work for something simpler, like Tic-Tac-Toe.

I don't however believe that the term "unbeatable AI" is quite correct. Chess programs are technically beatable alright, since chess cannot be solved to the end (yet). And for human players, chess programs let you set the playing strenght - noone likes to lose all the time. I've even seen a program with a particular mode where it would make intentionally weak moves occasionally, let you know of that and then tell you after your move whether you managed to take advantage of it.

A lot of intelligence comes from position evaluation. The decision how good or bad a position is involves a lot more than material count.

As for playing by the book (computers use databases for openings) it is the same thing with human players. A strong player needs to know a huge number of openings. If your opponent knows exactly what they are doing in the opening phase and you don't, you are at a great disadvantage. Unless you are extremely talented, you'll most likely never achieve a position where you might hope to gain the upper hand by intelligence only.

In addition a strong player needs to be familiar with a large number of tactical patterns and strategical positions. This is, however, where some fuzzy pattern matching is required because you won't neccessarily get the exact same positions described in the books.

Lol I have no passion for playing chess my self :P . there fore thats what I do. I just make a most random ever possible. the rule I keep is not to let the king be alone. of course ppl with strategy own me badly but if the other player was poor like me it will take a long time to End

anon: I didn't meant that to be taken literally :S

Btw I started the Chess AI today
its gonna a written in a Mix of C, ASM(mostly) , vb.net (Lazy to create a GUI in C)

I remember playing chess with the TV console ("Chessmaster"?) which had several handicapped playing styles for the computer. If I'm not mistaken, "drunken" was one of the personalities and that basically followed rogster's strategy. As this is very similar to how complete beginners play, then you might say it was the most human-like AI in the program. Except it never got sober.

The term artifical applies to AI in that it necessarily distibnguishes it from intelligence that is not created by man, i.e. natural intelligence.

As natural as genetically enhanced offspring, or shall we call them "artificial babies" now too? Currently this is an easy distinguishment to make, but in the (possibly near) future it will become increasingly harder, so the better we ween ourselves off it today and stop encouraging its use, the better our ability to grasp the concept of intelligence (and not just the kind that pertains solely to a mass of organic tissue, as many homo-centric philosphers will have you believe).
Granted, we are no where near this point, and perhaps the term "artificial" reminds us of this, but being all of scientific minds, we know that a Turing machine does not have to exist before we can begin to discuss its nature.

Mathematics does not exist in nature.

fractals, magic numbers, golden ratios..., just because we give nature a name and symbols doesn't make it any less natural or any more artificial ( a rose by any other name ...)