Human Shogi players still stronger than computer............

[George]

After 15 years since Deeper Blue beat Kasparov in 1997, in Japan Shogi programers are trying hard to beat Shogi professionals, and they are getting closer but it will be a few years until a computer can beat the best Shogi player and World Champion ==>
http://en.wikipedia.org/wiki/Deep_Blue_ ... 6#The_game

http://www3.nhk.or.jp/nhkworld/english/ ... 11_15.html

[Evert]

My understanding was that there are no official human-computer shogi tournaments anymore, has that changed?

I suspect that if we had something like the Stockfish framework for Shogi, the strength would go up extremely quickly, closing the gap sooner. Not that this is necessarily interesting.

[George]

My understanding was that there are no official human-computer shogi tournaments anymore, has that changed?

I suspect that if we had something like the Stockfish framework for Shogi, the strength would go up extremely quickly, closing the gap sooner. Not that this is necessarily interesting.

The computer system used for Shogi competition vs human in Japan are much much much stronger than any computer used by stockfish so far, but the complexity of Shogi compared to Chess is much much harder since the captured pieces can be used by the opponent and placed back into the board immediately and when you do NOT see a checkmate the captured pieces will come out of nowhere and be placed where checkmate is possible, also to make it harder Shogi is played on a 9x9 board instead of 8x8 like chess is. Long time ago Mr. Capablanca suggested and invented in 1920 a chess board 10x8, but most people do NOT even use it. Can you imagine if the captured pieces rule is added to our standard chess plus an additional square, Stockfish nor Komodo would be able to checkmate Carlsen in the next year or so..............

PS: Here is what Capablanca suggested in order to make standard chess more challenging, but without the rule of being able to place captured pieces wherever you needed them==> http://en.wikipedia.org/wiki/Capablanca_chess

[hgm]

Last year the computers beat the humans, though.

So I don't think you can say that humans are "still stronger". Just that there were not enough games to determine with any confidence wheter humans or computers are stronger.

[Evert]

My understanding was that there are no official human-computer shogi tournaments anymore, has that changed?

I suspect that if we had something like the Stockfish framework for Shogi, the strength would go up extremely quickly, closing the gap sooner. Not that this is necessarily interesting.

The computer system used for Shogi competition vs human in Japan are much much much stronger than any computer used by stockfish so far,

I'm not talking about the hardware used during the tournament, but the tuning effort that goes into the evaluation function (which is harder for Shogi than it is for chess, since material plays a larger role in chess).

but the complexity of Shogi compared to Chess is much much harder since the captured pieces can be used by the opponent and placed back into the board immediately

Yes, but the main problem is not in searching the drops (most of them are pointless) but in initiative (can I mate you before you mate me?), which is more an evaluation issue.

and when you do NOT see a checkmate the captured pieces will come out of nowhere and be placed where checkmate is possible, also to make it harder Shogi is played on a 9x9 board instead of 8x8 like chess is. Long time ago Mr. Capablanca suggested and invented in 1920 a chess board 10x8, but most people do NOT even use it. Can you imagine if the captured pieces rule is added to our standard chess plus an additional square, Stockfish nor Komodo would be able to checkmate Carlsen in the next year or so..............

PS: Here is what Capablanca suggested in order to make standard chess more challenging, but without the rule of being able to place captured pieces wherever you needed them==> http://en.wikipedia.org/wiki/Capablanca_chess

You're aware that my engine Sjaak plays all of these, right? Sure, it's in no danger of winning any major tournaments in the next decades, but that's not it's purpose anyway.

[George]

My understanding was that there are no official human-computer shogi tournaments anymore, has that changed?

I suspect that if we had something like the Stockfish framework for Shogi, the strength would go up extremely quickly, closing the gap sooner. Not that this is necessarily interesting.

The computer system used for Shogi competition vs human in Japan are much much much stronger than any computer used by stockfish so far,

I'm not talking about the hardware used during the tournament, but the tuning effort that goes into the evaluation function (which is harder for Shogi than it is for chess, since material plays a larger role in chess).

but the complexity of Shogi compared to Chess is much much harder since the captured pieces can be used by the opponent and placed back into the board immediately

Yes, but the main problem is not in searching the drops (most of them are pointless) but in initiative (can I mate you before you mate me?), which is more an evaluation issue.

and when you do NOT see a checkmate the captured pieces will come out of nowhere and be placed where checkmate is possible, also to make it harder Shogi is played on a 9x9 board instead of 8x8 like chess is. Long time ago Mr. Capablanca suggested and invented in 1920 a chess board 10x8, but most people do NOT even use it. Can you imagine if the captured pieces rule is added to our standard chess plus an additional square, Stockfish nor Komodo would be able to checkmate Carlsen in the next year or so..............

PS: Here is what Capablanca suggested in order to make standard chess more challenging, but without the rule of being able to place captured pieces wherever you needed them==> http://en.wikipedia.org/wiki/Capablanca_chess

You're aware that my engine Sjaak plays all of these, right? Sure, it's in no danger of winning any major tournaments in the next decades, but that's not it's purpose anyway.

But to point out that Japanese shogi programmers are not in the same level of programmers that program Stockfish or Komodo is without evidence no matter what the tuning effort that goes into the evaluation function, all the Japanese programmers are up to date and probably know more about tuning.

PS: If it was so easy to program a top level Shogi chess engine why Larry Kaufman who is a level 5 dan Shogi player and co-author of Komodo has not tried to make a top level Shogi program in the same caliber as Komodo. Probably he knows what it take to make a top level Shogi program but instead he leave it to Japanese programmers who have years and years of experience tuning their shogi chess program to be in the level where they have accomplished so far.
https://chessprogramming.wikispaces.com/Larry+Kaufman

http://eric.macshogi.com/shogi/handicap ... intro.html

https://www.google.com/search?q=Larry+K ... 11&bih=445

[lkaufman]

Actually Don and I did make a "Komodoshogi" and it played on the server a bit. But we only spent two weeks on it so it was only about 2000 rated there. A Japanese backer expressed interest but didn't follow thru, and Don could not afford to work for free. It's pretty hard for a non-Japanese to get into the Japanese shogi market. The hard part of the program is indeed the eval, which is far more crucial to elo in shogi than in chess.

[Evert]

But to point out that Japanese shogi programmers are not in the same level of programmers that program Stockfish or Komodo is without evidence no matter what the tuning effort that goes into the evaluation function, all the Japanese programmers are up to date and probably know more about tuning.

Of course no one made any sort of claim to the contrary. At the same time it should be clear that the amount of effort that has gone into tuning Shogi engines is smaller than the effort that has gone into tuning orthochess engines.

For one thing, interest in Shogi is mostly limited to Japan. Sure, there's a niche interest in the rest of the world, but it's tiny compared to the interest in orthochess. Japan itself has a population that is about twice the UK, so if the same fraction of people are interested in Shogi as there are in Chess in the west, that's far fewer people working on it. Furthermore, most interest in computer Shogi seems to be academic, but the point there is not necessarily to produce the strongest program, but to investigate search and evaluation techniques. Tuning is not science, it's engineering. Scientifically, the amount of CPU time that has gone into Stockfish tuning is a waste of time: you can get a paper out saying "look, we tuned this program using this method, which has this efficiency compared to this other algorithm, has a confidence interval of such and so to find improvements etc", but you don't get a subsequent paper out saying "look, we tuned it some more and now the program is 10 elo stronger". It's a waste of time.

Look at the CCRL rating list: the top is dominated by commercial and amateur/open source engines. Just by population size, that market is much smaller for Shogi than it is for western chess.

So while it's (much) harder to write a good Shogi evaluation than a good Chess evaluation, that's only part of the reason Shogi programs are relatively weaker than Chess programs - Chess simply has had more CPU hours devoted to it.

PS: If it was so easy to program a top level Shogi chess engine

Who said it's easy?

[Evert]

Actually Don and I did make a "Komodoshogi" and it played on the server a bit. But we only spent two weeks on it so it was only about 2000 rated there. A Japanese backer expressed interest but didn't follow thru, and Don could not afford to work for free. It's pretty hard for a non-Japanese to get into the Japanese shogi market. The hard part of the program is indeed the eval, which is far more crucial to elo in shogi than in chess.

Do you know what the most important evaluation terms are?

My impression (never having had the opportunity to play the game myself yet, just from watching computer games) is that a good king-safety is absolutely crucial, but quantifying that is hard. I try to keep squares around the king under control (and occupied by generals) and that's ok-ish, but it's very far from optimal.

HGM once mentioned that strong Shogi programs know how to attack a well-build castle (they recognise the pattern), but fail to efficiently dismantle a sloppy one, despite recognising that it's a poor castle. That suggests that something is missing in the evaluation (or rather, that a relevant term is present, but not general enough/too specific for handling a particular castle). Do you have any thoughts on that?

[myfish]

Shogi programs are getting stronger. No doubt for the average player, they are VERY formidable and myself I find many engines simply too strong. I have, rarely beaten some of the more common ones but it is, occasional rather than the norm.

Shogi is hugely technical and strategical even by comparison to an already tough and hugely strategical western chess. The permutations are almost, whist not infinite, they seem that way.

Just the other day, I played a human <> human game where it was deadlock for a long time and the focus changed from being a king game, to that of trying to get a piece in hand, held at the time, by your opponent and in a strong position itself, to start the game to mate with the king.

And, I think that is quite common where players play strategies that in the long term will or may rely on a piece that is in your opponents possession.

Maybe it's this depth, that is separating the computer shogi engines really looking as strong as the professional. Like I say, ordinary players have a formidable opponent in many top shogi engines. Too strong in most cases for kyu players but then, there seems to be this almost spiritual leap to 'knowing' and 'seeing' the engines haven't grasped yet.

But, I think in time, IF the pro's play the newer engines more and more and we see the results, this gap may narrow in the next decade.