AI takes on the game of Go
 

I'm surprised nobody posted this, but apparently the guys who created all the super computers that beat the best of the human chess players are bored with that game and have moved on to a new game. Go. Yup, Go. Here is an article talking about what they're planning on doing and what's involved. Should be interesting to see how this does and how well they perform against regular Go players and masters.

Oddly, once they beat everyone at Go, then what will they have next to beat us at?

Tiddlywinks.

Xiangqi/Shogi, which is like a Asian verison of chess.

Who beat me at GO ? noooooo

That would be easier than Go, which is much deeper and takes much more strategy and skill.
My guess would be either American Football or Counter-Strike,
"It's third and 5 should we go a pass or a run, or a trick play?"

"Should we rush again? We won with that last round but won't they be execting that? Or should we camp? Or will they expect that we expect that they expect us to rush again? Then we should rush..."

Well, they're currently starting out with a process of elimination technique, which I feel won't stand up against good quality GO players. However, what they learn from that may just push them to a whole new level and help them better learn how to attack the game with a formula that wins. I admit I'd hate to see the worlds best Go player get stomped by a machine, but seriously, what will be gained in the world of computer AI will be well worth someone getting their pride ground thuroughly into the floor.

I think it would be far harder to program a computer to play go effectively than to play chess effectively.

If they manage to make something that can tackle 1 dan players, I'd be surprised.

Go is in a totally different league than chess: Consider the branching ratios.

In chess, at the beginning (with the most # of pieces on the board) you have 20 possible moves. In go, you have 361.

Each step multiplies that number, so as an approximation, go has 361^180=2.2*10^(460) total number of moves in the game tree (for one player)
(about a google to the forth power).

and chess has 20^50= 10^65 (chess games average about 50 moves max)

That's nearly 4 google times less total number of moves you have to search.

I.e, Go is in a totally different league.

Has anyone seen the current Go playing programs available? I don't think anyone's managed o make anything better than 10 kyu. I don't know how many stones handicap a 10 kyu would have against a 9p, but it would be a hell of a lot.

Yes, the best commerically available program is "Many Faces of Go" I think, and it can play at 10 Kyu or so (bots on the online go sites running it vary around 10-12 kyu).

A 9p versus a 10kyu is basically unfair at any handicap. The general rule is 1 stone per level difference, so that'd be a 19 stone handicap :).

Surely the handicap would be more than that, given that the amateur dan ranks come before the professional dan...

I make it a 25 stone handicap... :p

Oh yes, right... that's what you meant by 9p. Unfair at any case :).

To play a game clever, you do not necessarily need to know all the possible moves in the game. I doubt a human being is playing this way. Its not always about bruteforce power, infact I wish for some software piece that is cleverly written, relying on non-deterministic, semidecidable algorithms. So it doesn't have a 100% chance of winning, but enough power to be a strong opponent.
Bruteforce power is so inelegant :D

Sheesh, I don't know how you guys play the computer with a 10kyu advantage. I get my *** kicked regularly even with a 25 stone advantage. Then again I've never really picked up on Go yet either. ;)

I don't have a 10 kyu advantage... I was just pointing out that some people do. In fact, I'm about 20 kyu (so pretty crap).

But, that's EXACTLY the basic method that Deep Blue used to (pretty much) beat Kasparov. Sure, there was some "knowledge" in the algorithm, but the basis was a good 'ole (pruned) DFS (Depth first search). My point was that for Go, such a brute force method is completely impossible.

I agree, however it would be more possible the smaller the board.

Indeed. I think most computer go tournaments are still played on 9 x 9 boards :).

On a 9x9 board, I can play Many faces of Go to a draw with no handicap on the highest level. Not sure if that's 'cause I suck or if that's because MFoG is pretty decent on a 9x9 board.

What rank are you supposed to be?

Oh, I agree. My point is that while the old methods won't work, it will however spawn newer methods for the computers to use by providing the programmers with a much tougher challenge.