computer beats pro

[Larry Kaufman]

     Yesterday was a big day in shogi history. The top shogi program GPS shogi defeated a top pro, Miura, in a high-profile game with a time limit of 4 hours per side plus byoyomi. Miura had sente in a symmetrical aiyagura opening and appeared to be playing for an entering king, but was unsuccessful and got mated. He is now the strongest pro to have lost an official came to a computer. Moreover, this was the fifth and final game in a series of weekly games between increasingly strong pros against increasingly strong engines. The final score was 3 wins for the computers, 1 for the pros, and one jishogi. 

      So it now seems clear that only the very best pros have much chance to defeat the best program. I hope we can see a match (or series of games) between the top computer and the titleholders Habu, Watanabe, and Moriuchi. It is up to them to show that computers have not yet surpassed all human players. 

      Whatever the result of such a match, it won't be long before computers are clearly superior to all humans. I predict that within four or five years the computers will be strong enough to successfully give lance handicap to any human, though whether the pros will agree to play such matches is another question. 

     In my own experience, playing fast handicap games with Bonanza 6.0 (not the very strongest engine but close enough), I find it just as difficult to win if not more so than when playing professional players, even though the opening library for handicap games is very limited. Although I can generally win these fast games at rook and lance handicap (or two piece), I almost always lose at bishop, and rook is an uphill struggle. Perhaps other readers here will report on their experience playing handicap games with Bonanza or with other top engines. 

     Larry Kaufman (U.S. Champion) 

[Tomoyuki Kaneko]

Hi, all.

Just FYI.
http://www3.nhk.or.jp/nhkworld/newsline/201304210206.html

Tomoyuki

[DARREN PAULL]

Hi Tomoyuki san

Nice video , domo arigatou gozaimasu

I hope the humans can strike back next year! :)

Darren

> Date: Mon, 22 Apr 2013 23:35:47 +0900
> From: kan...@graco.c.u-tokyo.ac.jp
> To: sho...@googlegroups.com
> Subject: Re: computer beats pro

> --
> You received this message because you are subscribed to the Google Groups "SHOGI-L" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to shogi-l+u...@googlegroups.com.
> To post to this group, send email to sho...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.
>
>

[Larry Kaufman]

     If they wait another year for such a match it will be too late, even for the titleholders, to win against best computer. Right now they still have a fair chance. The situation now is about the same as it was in chess at the turn of the century, when matches between the best computers and Kasparov and Kramnik were dead even. Now in chess no human has a chance against a top computer without a handicap, and soon it will be so in shogi. Then in another 15 years or so GO will be the critical battleground between best computer and best human.

[Jeff Rollason]

Gripping stuff and thanks Tomoyuki for the posting. I watch NHK but seemed
to have missed that. The next Computer Shogi Championship (CSA) is in a few
days and I'm sure it will be buzzing with this result. I have been involved
in Computer Chess since 1977 and watched the progression to Deep Blue and in
the last 15 years have been involved in Computer Shogi, but have not entered
the CSA in the last 3 years. I will be in Japan shortly after the CSA and
will look forward to catching up with the competitors.

Jeff Rollason - CEO and founder
AI Factory Ltd - www.aifactory.co.uk

----- Original Message -----
From: "Tomoyuki Kaneko" <kan...@graco.c.u-tokyo.ac.jp>
To: <sho...@googlegroups.com>
Sent: Monday, April 22, 2013 3:35 PM
Subject: Re: computer beats pro

> --
> You received this message because you are subscribed to the Google Groups
> "SHOGI-L" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to shogi-l+u...@googlegroups.com.
> To post to this group, send email to sho...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.
>
>
>
>

> -----
> No virus found in this message.
> Checked by AVG - www.avg.com
> Version: 2013.0.3272 / Virus Database: 3162/6263 - Release Date: 04/21/13
>

[Jeff Rollason]

On the footnote on Go: I have been working with an Academic team on a research work that is involved in the critical methodology that also drives recent Computer Go success. That is UCT-based MCTS (Monte Carlo Tree Search). I say success in inverted commas as, although there has been a big jump in performance in the last decade, there is still a very long way to go. MCTS though promises much in so many problem domains, so may well become a dominant AI method in many domains in a few years.

 

For me, I now use this for cards games such as Spades. I have wondered about its use in Shogi (which I have also been involved in for many years). One key aspect to MCTS is that it tends to make move selections on a more "probability on win" basis than the methods such as minimax, which are more like theorem provers, tending to prove which is the best move. In that respect I would expect that future MCTS-type game AI to be more human-like than those based on minimax.

 

Jeff Rollason - CEO and founder
AI Factory Ltd - www.aifactory.co.uk

 

Checked by AVG - www.avg.com

Version: 2013.0.3272 / Virus Database: 3162/6277 - Release Date: 04/27/13

[h.g.m...@hccnet.nl]

MCTS is kind of a very expensive method for evaluating positions with no
other knowledge than the winning condition. For games where a reasonable
evaluation heuristic is known, the latter usually outcompetes MTCS by many
orders of magnitude.

Although evaluation of non-terminal Shogi positions is not nearly as
straightforward as in Chess, it is still good enough that MCTS would have
a very hard time to compete.

[Jeff Rollason]

I confess that the combinatorially explosive nature of Shogi makes the
prospect of a competitive MCTS approach difficult to imagine. My own shogi
program Shotest did quite well many years ago, but based on really tiny
search trees and a huge amount of evaluation. However it was also hard to
imagine that MCTS could possibly deliver in the much harder domain of
Computer Go, but it is now the dominant method in that domain, so I would
not rule it out.

The MCTS I am using is not terminating on a win/lose result, but an
evaluation score, so not the usual situation. I also ponder the possibility
of considering selective termination on non-terminal nodes. Early days here.

----- Original Message -----
From: <h.g.m...@hccnet.nl>
To: <sho...@googlegroups.com>
Sent: Sunday, April 28, 2013 9:39 AM
Subject: Re: computer beats pro

> --
> You received this message because you are subscribed to the Google Groups
> "SHOGI-L" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to shogi-l+u...@googlegroups.com.
> To post to this group, send email to sho...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.
>
>
>
>

> -----

>
> Checked by AVG - www.avg.com

> Version: 2013.0.3272 / Virus Database: 3162/6278 - Release Date: 04/27/13
>

[h.g.m...@hccnet.nl]

Well, with Go there really wasn't much of an alternative, (not even
mobility is a guidance), so anything looks good. I don't play Go, but it
strikes me as a game where many more moves are playable than in Chess or
Shogi, where you really need very accurate tactics, or the game will be
immediately lost. In Go this can also happen (if a large chain is
threatened with capture), but it is much more rare. So it is also more
difficult to do representative MC sampling of a Chess or Shogi tree,
unless your sampler is tactically smart (which slows it down enormously).

But who knows, perhaps there is a way to make this work.

I won't come to the CSA championship; my engine is not yest strong enough
to have a realistic possibility to survive the first round, and traveling
half-way around the world for just one day of Shogi is a bit too much for
me. I will go to the ICGA Olympiad instead (where I can also play
mini-Shogi and Xiangqi), which is in Yokohama in August. But perhaps next
year...

[Jeff Rollason]

Ok I will pitch in that the CSA is a very enjoyable event and worth it for
an excuse to visit Tokyo and just to be among so many strong Shogi
programmers, so I would keep that one well in mind for the future. As for
me, I want to return when I have more time. I have always managed to be in
the second or final round after some 11 times entering, but well understand
now that progression from the first round is no longer a given! :)

----- Original Message -----
From: <h.g.m...@hccnet.nl>
To: <sho...@googlegroups.com>
Sent: Sunday, April 28, 2013 11:35 AM
Subject: Re: computer beats pro

[h.g.m...@hccnet.nl]

Op Zo, 28 april, 2013 1:42 pm schreef Jeff Rollason:

> Ok I will pitch in that the CSA is a very enjoyable event and worth it
> for an excuse to visit Tokyo and just to be among so many strong Shogi
> programmers, so I would keep that one well in mind for the future. As for
> me, I want to return when I have more time. I have always managed to be
> in the second or final round after some 11 times entering, but well
> understand now that progression from the first round is no longer a given!
> :)

Oh yes, it will surely remain high on my list of things I still want to
do. But my engine (Shokidoki) is vastly weaker than yours (which is not
surprising, as I put it together in just 2 months, using my Xiangqi engine
as a starting point, and hardly have spent any time on it since, other
than optimizing 5x5 mini-Shogi and adding 6x6 Judkins Shogi to it). So I
really should do some serious work on it first.

Btw, the ICGA Olympiad was also a very enjoyable event. In 2010 it also
was in Japan (Kanazawa), and although the Shogi tourney was not nearly as
large as that of CSA, there still were 9 participants, all very friendly.
and there were also 8 participants for mini-Shogi, which my engine
actually is a lot better at. I really look forward to being there again.