Arimaa Forum
        (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi)
      

        Arimaa >> General Discussion >> Will Arimaa pass shogi?
        
(Message started by: Fritzlein on Mar 10th, 2007, 11:59am)
Title: Will Arimaa pass shogi?
Post by Fritzlein on Mar 10th, 2007, 11:59am
From the results of the Challenge match, it seems there are a couple dozen humans who can beat the best computer at Arimaa.  For shogi, the number of humans who can beat the best computer is somewhere around 300.  If their numbers continue to dwindle while our numbers continue to expand, we might be able to say Arimaa is "more computer-resistant than shogi" as well as "more computer-resistant than chess".
Title: Re: Will Arimaa pass shogi?
Post by chessandgo on Mar 11th, 2007, 7:06am
I'm looking forward to being able to say the same for go as well :)
Title: Re: Will Arimaa pass shogi?
Post by The_Jeh on Mar 11th, 2007, 9:46am
Yes, Fritzlein, but how many programmers even heard of Arimaa?  Much more time and money has been spent developing programs for Shogi, which is much more popular at this point.  I actually think that after many years of research, computers WILL be able to defeat even you consistently.
Title: Re: Will Arimaa pass shogi?
Post by chessandgo on Mar 11th, 2007, 10:09am

on 03/11/07 at 09:46:26, The_Jeh wrote:
Yes, Fritzlein, but how many programmers even heard of Arimaa?  Much more time and money has been spent developing programs for Shogi, which is much more popular at this point.  I actually think that after many years of research, computers WILL be able to defeat even you consistently.


Well, I think Karl already pointed out that the ratio time spent on bot programming / game studying time was probably much bigger for arimaa than for many other games, and we might expect that humans will progress at least as fast as bots do ... thus I'd trust Karl to stay much ahead on bots ;)
Title: Re: Will Arimaa pass shogi?
Post by Fritzlein on Mar 11th, 2007, 10:59am
Yes, much more developer time has been spent on shogi than on Arimaa.  There is much more money and glory in producing a strong shogi program, so of course there has been much more effort in that direction.

On the other hand, there are several million people who know how to play shogi, compared to a few thousand who know how to play Arimaa.  As chessandgo said, the ratio of developers to players is probably higher for Arimaa.  Could shogi really have a thousand times the programmer effort that Arimaa does?


on 03/11/07 at 09:46:26, The_Jeh wrote:
Yes, Fritzlein, but how many programmers even heard of Arimaa?
I don't know how many programmers have ever heard of Arimaa, but Omar was recently interviewing a programmer for a position at Yahoo, and the guy said, "Are you the Omar Syed who invented Arimaa?" :P

There are problems with any metric, but I think the absolute number of computer-beaters is a reasonably fair way to measure computer-resistance.

I believe there are fewer OCTI players who can beat JDB's bot testme than there are Arimaa players who can beat Bomb.  I wonder how many competent Havannah players are out there...
Title: Re: Will Arimaa pass shogi?
Post by Fritzlein on Mar 11th, 2007, 11:06am

on 03/11/07 at 09:46:26, The_Jeh wrote:
I actually think that after many years of research, computers WILL be able to defeat even you consistently.

I also think computers will eventually be able to beat me consistently.  However, by the time that happens, I probably won't be the last human in the way.  Even now, if a bot came along that could beat me consistently, people would say, "Ah, but can it beat chessandgo?"

What percentage chance do you give for a computer to win the Arimaa Challenge by 2020?  Right now I'd put it around 40%.  But if you extend the time frame to "many years" (say 2050), I would change my bet to 90% or more.
Title: Re: Will Arimaa pass shogi?
Post by arimaa_master on Mar 12th, 2007, 2:52am

on 03/11/07 at 11:06:30, Fritzlein wrote:
What percentage chance do you give for a computer to win the Arimaa Challenge by 2020?  Right now I'd put it around 40%.  But if you extend the time frame to "many years" (say 2050), I would change my bet to 90% or more.



Given that somebody mentioned that we are with arimaa in situation like we were in chess in 1960's and after 50 years that is possibly no human who can compete with computers in chess (with rybka playing at strong hardware). Given that I am quite with your numbers. However, I would like to point out that there is much more strategy involved in arimaa (like everybody knows). So my guess is 10 percent till 2020 and only 30 percent till 2050 (I am sorry to be too poor to make a bid for arimaa challenge :)).
Title: Re: Will Arimaa pass shogi?
Post by NIC1138 on Mar 12th, 2007, 4:14pm

on 03/11/07 at 11:06:30, Fritzlein wrote:
What percentage chance do you give for a computer to win the Arimaa Challenge by 2020?  Right now I'd put it around 40%.  But if you extend the time frame to "many years" (say 2050), I would change my bet to 90% or more.


Following the Law of Accelerating Returns (http://en.wikipedia.org/wiki/Law_of_Accelerating_Returns), today  I would say a 38% chance that we will lose the challenge in 2020. In 2010, it will be a 27% chance that they will win at 2020. Then in 2015 it will aready be a 8% chance, and in 2020 they will have won in 2018 already! ;)
Title: Re: Will Arimaa pass shogi?
Post by The_Jeh on Mar 13th, 2007, 12:24am
By 2020, I would say there's about a 30% chance of the challenge being won by computers.  (That is, unless the program involved is not named bot_Bomb, in which case the probability rises a few points.  This would indicate increased activity in the programming community.)

I think perhaps there is some flaw in predicting Arimaa's computer-resistance based on its game-tree complexity.  Sure there are thousands of combinations on each move, but the movement and interaction of the pieces themselves are actually quite simple.  For this reason, I think there is probably an ingenious way for the game tree to be trimmed just waiting to be discovered. Maybe.

On the other hand, something that makes Arimaa harder to crack is the fact that the board can be seemingly quite stale for several turns, having no captures and having obscure "cold-war" situations, making comparison of different move choices more difficult.

And Go, or weiqi as it's called in China, is going nowhere.  It is the "rock in which I trust," as far as board games go.
Title: Re: Will Arimaa pass shogi?
Post by NIC1138 on Mar 15th, 2007, 8:56am

on 03/13/07 at 00:24:58, The_Jeh wrote:
And Go (...) is going nowhere.


Is this an intended pun? :)
Title: Re: Will Arimaa pass shogi?
Post by IdahoEv on Mar 16th, 2007, 8:57pm

on 03/13/07 at 00:24:58, The_Jeh wrote:
I think there is probably an ingenious way for the game tree to be trimmed just waiting to be discovered. Maybe.


Or to say the heck with a deep tree search and simply directly recognize strategically good positions that will lead to eventual capture or win.   If you can write a good enough eval you don't need to prune the tree.

I think the simplicity of interaction you point to may in fact make this possible.  Can't prove it, but it's the direction I'm working on for Zombie.
Title: Re: Will Arimaa pass shogi?
Post by The_Jeh on Mar 16th, 2007, 11:31pm

Quote:
Is this an intended pun? :)

No, but now that you mention it...

I would like to see the result of a collaboration of different programs.  They each give a quantitative opinion about different moves, and the move with the best average on a standardized scale gets played. I don't think this would improve performance by any means, but it would be interesting.

By the same token, I would like to see one of the top players face off against the entire Arimaa community.  Everyone votes on the next move.  The problem with this, of course, is that there are far too many moves to choose from.  To overcome this problem, I would have three to five other top players or bots suggest moves, and a move is elected among these.  This would weaken the experiment, but it would still be interesting.  Also, the ensuing debates would be healthful.
Title: Re: Will Arimaa pass shogi?
Post by IdahoEv on Mar 19th, 2007, 1:15pm

on 03/16/07 at 23:31:56, The_Jeh wrote:
By the same token, I would like to see one of the top players face off against the entire Arimaa community.  Everyone votes on the next move.  The problem with this, of course, is that there are far too many moves to choose from.  To overcome this problem, I would have three to five other top players or bots suggest moves, and a move is elected among these.  This would weaken the experiment, but it would still be interesting.  Also, the ensuing debates would be healthful.


This would be a lot of fun.   We'd have to set up a thread in this forum for each move, and the challenged player (Fritz?  Chessandgo?) would have to promise not to read them so as not to be spoiled.   There would probably have to be a single person who was the central coordinator for the opposition team to organize the discussion, coordinate votes, etc.


Arimaa Forum » Powered by YaBB 1 Gold - SP 1.3.1!
YaBB © 2000-2003. All Rights Reserved.