Why there is no interest in Computer with odds Vs Humans match?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
jp
Posts: 707
Joined: Mon Apr 23, 2018 5:54 am

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by jp » Wed Jul 03, 2019 7:56 am

lkaufman wrote:
Wed Jul 03, 2019 4:27 am
Ovyron wrote:
Wed Jul 03, 2019 4:04 am
Do I gather that chess stopped being interesting so people are finally moving to other variants, and the most interesting one is chess with material odds? What happened here? :|
Chess did not stop being interesting, just chess between computers and humans with no handicap. That is why it is the topic of this thread.
But why is chess with material handicaps more interesting to people than chess with computer resources limited (which is much more interesting for computer science than handicaps)?

lkaufman
Posts: 3647
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by lkaufman » Wed Jul 03, 2019 3:25 pm

Laskos wrote:
Wed Jul 03, 2019 7:35 am
lkaufman wrote:
Wed Jul 03, 2019 5:54 am
Ovyron wrote:
Wed Jul 03, 2019 5:04 am
lkaufman wrote:
Wed Jul 03, 2019 4:27 am
Chess did not stop being interesting, just chess between computers and humans with no handicap. That is why it is the topic of this thread.
I agree on that, but there are many handicaps not being considered, like limiting depth (what is the lowest depth that Komodo needs to beat humans?) or nodes (what is the lowest nodes Komodo needs to visit to beat humans?) Couldn't Komodo boast to have some great chess knowledge if it can defeat humans without much search used?

For this I suggest something like Rebel Decade's move delay. You have two Komodos, one to pick the moves, and one to manage the time. So the one with low depth/nodes picks a move instantly, but the other doesn't know, and thinks it's in a normal game, and spends its time finding the best move to play, but when is its time to move, it plays the first Komodo's move.

Material handicap is the laziest one to implement, and it's sad to see people acting as if it's the only handicap left to try...
There have actually been a huge number of games played already on chess.com in three different events in which grandmasters played against Komodo handicapped primarily by search depth. The levels below level 20 on Komodo (level 25 in future versions) are handicapped by depth and by randomness, but the randomness is only significant below the top few levels. For all practical purposes, levels 17 thru 19 are just depths 8 thru 10 with a little variety. First Nakamura, then MVL, and then many teams of players (including GMs) played against the levels, working their way up. Most of the games were played at 5' + 2" I believe. At this time limit, I would summarize the results by saying that depth 8 is weak GM, depth 9 is strong GM, and depth 10 is super-GM (Naka/MVL). Regarding your delay suggestion, is this just to create the illusion of a normal game? Chess.com wanted Komodo to play as quickly as its search would allow, not wasting anyone's time. It's up to the human player to adapt to this.
I personally prefer material and move handicaps, because limiting depth or nodes has no meaning to ordinary chessplayers, only to chess computer geeks like us. You don't need to know anything about computers to know that if it beats a title contender giving him f7 and two moves, or two pawns, it has vastly outclassed him. But if it beats him giving him 1000 to 1 time odds (either explicitly or via nodes/depth) people just say that shows how fast computers are. But we'll continue to have events with limited depth handicaps. You can play Komodo online yourself with this depth-limit handicap on chess.com. If you want to play depth 10, just challenge Komodo 19.
Taking moves back by the GM was not explored. We don't even know what conditions these take backs can take. As many times as he wishes? How far he can take back (say up to 2-3 moves)? 20 takebacks total in one game (several for the same move being allowed)? I guess that taking back would help more against SF or Komodo than against Leela, as Leela can outplay a GM positionally, without that GM even realizing quickly what happens to him and realizing much too late.
Yes, we should try some takeback handicap, that's never been done by anyone to my knowledge. 20 takebacks with no restrictions sounds reasonable. There may be some practical problems, I don't know how clocks are handled by online servers when takebacks occur (the player should not get his time back as well as the move). SF and Komodo will also outplay GM positionally, but I agree Leela will do it more often. Another problem is that takebacks make for draws by reducing errors. Probably Hikaru or MVL won't win a match from Komodo even with 20 takeback odds, but they might get mostly draws, and if we also give them White pieces every game, maybe they can draw all the games. Not very exciting for viewers I'm afraid.
Komodo rules!

lkaufman
Posts: 3647
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by lkaufman » Wed Jul 03, 2019 3:29 pm

jp wrote:
Wed Jul 03, 2019 7:56 am
lkaufman wrote:
Wed Jul 03, 2019 4:27 am
Ovyron wrote:
Wed Jul 03, 2019 4:04 am
Do I gather that chess stopped being interesting so people are finally moving to other variants, and the most interesting one is chess with material odds? What happened here? :|
Chess did not stop being interesting, just chess between computers and humans with no handicap. That is why it is the topic of this thread.
But why is chess with material handicaps more interesting to people than chess with computer resources limited (which is much more interesting for computer science than handicaps)?
I think you answered your own question. Handicapping by computer resources is more interesting for computer science, handicapping by material and/or moves is more interesting for chess players who know nothing about the inner workings of computers. The first tells us something about computers, the second tells us something about chess.
Komodo rules!

todd
Posts: 36
Joined: Thu Apr 19, 2018 7:09 pm

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by todd » Wed Jul 03, 2019 4:56 pm

Takeback odds is an interesting idea. I play odds games against computers regularly and will have to try it.

Against a top engine, perhaps the human should not play normally and only use the takebacks when they realize they've made a mistake. Instead, we should play in a way that takes advantage of the fact that we know we'll be able to take back moves.

I suppose we still won't be able to play for a win, but how to maximize the chances of a draw? Perhaps "trying out" forcing lines and abandoning them if they seem unlikely to result in either a perpetual or a holdable simplified position is one way to go, while saving a few takebacks for blunders too.

Another funny idea is that you can use takebacks in the opening and then play the exact same move again anyway :) (If the engine plays several different lines and you get one you didn't prepare as much for)

Uri Blass
Posts: 8532
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by Uri Blass » Wed Jul 03, 2019 8:39 pm

todd wrote:
Wed Jul 03, 2019 4:56 pm
Takeback odds is an interesting idea. I play odds games against computers regularly and will have to try it.

Against a top engine, perhaps the human should not play normally and only use the takebacks when they realize they've made a mistake. Instead, we should play in a way that takes advantage of the fact that we know we'll be able to take back moves.

I suppose we still won't be able to play for a win, but how to maximize the chances of a draw? Perhaps "trying out" forcing lines and abandoning them if they seem unlikely to result in either a perpetual or a holdable simplified position is one way to go, while saving a few takebacks for blunders too.

Another funny idea is that you can use takebacks in the opening and then play the exact same move again anyway :) (If the engine plays several different lines and you get one you didn't prepare as much for)
I see no reason that the engine play different move in the opening.

The engine can be programmed not to play different move in the same position.

jp
Posts: 707
Joined: Mon Apr 23, 2018 5:54 am

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by jp » Wed Jul 03, 2019 8:43 pm

lkaufman wrote:
Wed Jul 03, 2019 3:29 pm
I think you answered your own question. Handicapping by computer resources is more interesting for computer science, handicapping by material and/or moves is more interesting for chess players who know nothing about the inner workings of computers. The first tells us something about computers, the second tells us something about chess.
I think this probably overestimates how much handicap matches will tells us about chess, and underestimates how much resource optimization (or whatever we want to call it) will tell us about chess.

lkaufman
Posts: 3647
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by lkaufman » Wed Jul 03, 2019 9:04 pm

I think that by accident I chose an engine, Arasan 18.0, as an opponent for Komodo in these handicap matches, which is an excellent proxy on 1 thread for Magnus Carlsen. Deep Fritz 10 on 4 cpus is rated 2830 on CCRL 40/40, and it performed over 2900 in beating Kramnik 4 to 2 in 2006 on the hardware of that time despite several unfavorable conditions (notably Kramnik using Fritz's opening book during the game!). Arasan 18.0 on 1 cpu is 2860 on that list should be favored under normal tournament conditions in a match with Carlsen (2878 now), especially on say a 2016 xeon, ten years newer, although it might be a close match. Does this seem realistic? Of course if the match were played at a rapid time control, the engine would be even more favored.

So I started a new match at 20' + 10", latest Arasan 18.0 vs. latest Komodo (Contempt 50, Variety 10), both on 1 thread, at pawn and two move odds (f7 off, e4 played, WTM). In a real match with Carlsen, Komodo would use many threads, not one, so this is very biased against Komodo. On the other hand, the human knows to simplify against a stronger opponent or when ahead, but with only one pawn in material, but a huge advantage in time and king safety, energetic play is needed to win, so this handicap is more easily simulated than say knight odds. We'll see what happens. So far Komodo is off to a good start.
Komodo rules!

jp
Posts: 707
Joined: Mon Apr 23, 2018 5:54 am

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by jp » Wed Jul 03, 2019 9:35 pm

lkaufman wrote:
Wed Jul 03, 2019 9:04 pm
Deep Fritz 10 on 4 cpus is rated 2830 on CCRL 40/40, and it performed over 2900 in beating Kramnik 4 to 2 in 2006 on the hardware of that time despite several unfavorable conditions (notably Kramnik using Fritz's opening book during the game!).
That feels a bit strange. Maybe it'd be more 'natural' just to forbid Fritz from using an opening book, but would Fritz be worse off with this more natural condition than the one used?

todd
Posts: 36
Joined: Thu Apr 19, 2018 7:09 pm

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by todd » Wed Jul 03, 2019 9:45 pm

Uri Blass wrote:
Wed Jul 03, 2019 8:39 pm

I see no reason that the engine play different move in the opening.

The engine can be programmed not to play different move in the same position.
Yes, but the drawback of that approach is that it may allow the opponent to prepare very deeply down the only line it plays, which also makes obtaining draws easier for the weaker player.

lkaufman
Posts: 3647
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by lkaufman » Wed Jul 03, 2019 10:00 pm

jp wrote:
Wed Jul 03, 2019 9:35 pm
lkaufman wrote:
Wed Jul 03, 2019 9:04 pm
Deep Fritz 10 on 4 cpus is rated 2830 on CCRL 40/40, and it performed over 2900 in beating Kramnik 4 to 2 in 2006 on the hardware of that time despite several unfavorable conditions (notably Kramnik using Fritz's opening book during the game!).
That feels a bit strange. Maybe it'd be more 'natural' just to forbid Fritz from using an opening book, but would Fritz be worse off with this more natural condition than the one used?
Well, we've had a couple of matches with this handicap (no opening book past move 3), but I think it is a bigger handicap than the one Kramnik got. If the human doesn't want to play against a fully-booked engine, both sides could agree to start with openings like 1.a3 a6 for example, with the computer using no book. Top level chess has become too much of a memory contest, that's the underlying problem here. Human GMs have memorized thousands of opening variations, but of course a computer book has even more in memory.
Komodo rules!

Post Reply