Slashdot Mirror
Computer Defeats Human At Japanese Chess
Calopteryx writes "A computer has beaten a human at shogi, otherwise known as Japanese chess, for the first time. As New Scientist reports, computers have beaten humans at western chess before, but that game is relatively simple, with only about 10^123 possible games existing that can be played out. Shogi is a *bit* more complex, offering about 10^224 possible games."
pointless comment text
a computer could have beaten me at shogi a long time ago, but it never asked to play.
"National Security is the chief cause of national insecurity." - Celine's First Law
First time "a computer" has beaten "a human", eh?
I'm sure they mean: first time a computer has beaten a 1st dan (or whatever shogi ranks are called) grandmaster in an offical tournament setting...
Also, I don't think the theoretical number of games is very relevant. Paper-scissor-rocks has an infinite amount of possible games, ie 1 draw followed by a win, 2 draws ... inf draws. Much more relevant would be branching factor, difficulty of estimating positional strength, horizon problems, long term dependencies etc.
... design and write another computer program to beat a human at chess or shogi - THEN i'll be worried.
Computer Defeats Human At Japanese Chess
Human, my friend. Human.
"I'm just here to regulate funkiness."
Chess has a natural limit since the number of pieces monotonically decreases during the game. Shogi lets you drop (add) pieces that you capture, so a game can go on for a long time.
Intron: the portion of DNA which expresses nothing useful.
I saw Shogi's show in Branson, that guy plays a mean fiddle.
soooo irritating whenever a go player brings this up.
Go only wins through brute force.
go is 19x19
shogi is 9x9
chess is 8x8
If a game like shogi or chess was extended to 19x19 it would be vastly harder for a computer.
Computers playing Go on 9x9 have beaten 9th dan.
And if it was 8x8 it would be even easier.
What makes Go hard isn't anything particularly neat about the game.
Is just a boring brute force exercise.
Not sure about large numbers, but they certainly had math geniuses
http://www.cut-the-knot.org/proofs/jap.shtml
Intron: the portion of DNA which expresses nothing useful.
I'm in the process of joining the AGA ... that is, I'm strategically holding off until I get more Go literature under my belt (I can bank life-and-death problems against high level players; but my initial set-up and my capture race performance is weak, so my territory boundaries are not far reaching enough and creating wider ones stretches me thin). Maybe in 2 months.
Support my political activism on Patreon.
I spent a summer once working for a professor who has spent his life trying to develop an AI for Go!
In particular I was compressing read-only hash tables of end states. He was basing his approach on the work of someone who had developed AI for checkers but I think it's obvious that Go is a little bit bigger problem.
(To be specific: http://lie.math.brocku.ca/twolf/home/publications.html#3)
Ugh. What's with perpetuating this nonsense? A computer did not beat the top ranked Western chess player. Rather, a group of people _reprogrammed the computer after each match_ to beat the top ranked Western chess player.
TFA, it is annoyingly vague on an important point: What is the rank of the Japanese player that lost?
And as others have pointed out, let see a computer take down a top ranked (10th Dan) player at Go. The best a machine has done (I think) is winning against a 5th Dan.
Go is a simple game.
Mind numbingly simple, in fact.
It's just a LARGE game.
Chess has actual complex rules. It is a hard game.
Mind-numbingly hard, in fact.
It's just a relatively SMALL game.
Sure they can.
The rules just need extending.
Is no different than fischer random chess dramatically increasing chess complexity for an AI.
That's the problem for me with go. It is a simplistic game that, yes, takes a lot of skill for a human. No doubt.
But the number of varying interactions is, well, limited by the tiny ruleset.
Depends on what your definition of "good" is. Efficient? Easy? Fast? etc
If you can map out every possible outcome of a game given every possible move (calculate every ply), you can play optimally. You might need multiple super computers, lots of time, etc (for now!), but if you can do that, you can pretty much guarantee optimal play. Other "smarter" methods are of course faster, more resource efficient, etc, but not as optimal if you know every possible outcome.
If you bother to read the article:
"IBM say they have improved artificial intelligence enough that Watson will be able to challenge Jeopardy champions, and they'll put their boast to the test soon, says The New York Times. "
Do you realize what this means? Ken Jennings versus robots. They could make an entire new show out of this and I'd watch it religiously.
You're bored by the relatively fast advance of computer intelligence? Humans have for the first time in recorded history lost their title of "Best at Shogi" to computers (and orangutangs have presumably been bumped down to 3rd). That may not have any real-world significance, but in the grand scheme of things, it wasn't too long ago that computers couldn't beat us at math.
You're on a forum with a focus on computers, and you say that's boring? Jesus, what WOULD interest you? If it ran linux using a beowulf cluster? Simpsons quotes?
Well fine, I for one welcome our new shogi-playing computer overlords.
Go vs. Chess. RISC vs. CISC all over again.
The actual accomplishment, not specifically stated until the FOURTH paragraph of the New Scientist article with the same terrible headline, is that it's the first time a computer has beaten a professional human player; in this case, Ichiyo Shimizu, the female shogi champion.
What makes Go hard isn't anything particularly neat about the game.
Incorrect. There are many things that make go difficult for a computer to play: positional evaluation is tough. The branching factor is huge (unlike Chess and similar games, the number of available moves in a given board configuration is very large, as a stone can be played virtually anywhere on the board). Life-and-death is difficult to calculate. There are interactions between local and global play...
Go's board size is certainly a factor, yes, but if it were the only one, computers should excel at 13x13 or 9x9 games, and yet they don't.
So Go programs are getting there. Not as fast as chess, but they're still getting there.
"The urge to fly from modern systems, instead of moving through them to even greater, fairer things is, I think, an indi
See Arimaa , a new game with a board and set similar to Chess *but* with specific rules made to be difficult for a computer to play, and easy for a child.
How many options do you have when it's your turn to play with chess ? The average branching factor in a game of Chess is about 35, whereas in Arimaa it is about 17281 !
This is why a computer which can search to a depth of eight turns for each player in chess, can only search about three turns deep for each player in Arimaa...
This game is the new challenge for IA, easy for a child, difficult for a computer. A average human player wins against best programs.
The difference is that nobody would want to play a chess game on a board that size. Go grew to 19x19 by player preference, not as some artificial limit to make it hard to beat the computer.
Don't be so sure.. The most common Shogi is played on a 9x9 board with 40 pieces. True enough.. Just as the most common western chess is played with an 8x8 board and 32 pieces. That is far from the only Shogi though.
Maka-Dai-Dai Shogi has a 19x19 board, with 192 pieces.
There are plenty of variant rules that make for an even more interesting game, one of which has the piece take on the move of the piece in front of it. Others have specific rules about drops, others don't have drops..
There is a Shogi variant, Tai Shogi which has a 25X25 board, and 354 pieces. Something I've wanted to make for years, even if only as a display piece. And there may be bigger I haven't heard of.
Or at the other end of the scale, a 4X5 Micro Shogi board with 10 pieces.
http://trout.customer.netspace.net.au/ Old VB program that works great on Linux under WINE too. So you can try lots of different variants
Chess is a complex game, but there are a huge number of variants. Most are unknown outside the few who play them.
It is difficult to get a man to understand something when his job depends on not understanding it.
We don't know how our minds work well enough to say that we don't use brute force. Obviously, consciously, we're not thinking about it that way, but who knows what kind of processing the brain does to produce those conscious thoughts? When you get a knack of intuition like "ah that move would win" - is that just a brute force algorithm in the subconscious signalling termination with a result?