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.
Chessqueen
Posts: 510
Joined: Wed Sep 05, 2018 12:16 am
Full name: Nancy M Pichardo

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

Post by Chessqueen » Sun Jul 07, 2019 11:32 am

Uri Blass wrote:
Sun Jul 07, 2019 3:44 am
jp wrote:
Sat Jul 06, 2019 11:37 pm
Ovyron wrote:
Sat Jul 06, 2019 3:26 am
lkaufman wrote:
Wed Jul 03, 2019 5:54 am
I personally prefer material
Except that's no longer chess. The engine wasn't built with handicap in mind, the human's elo wasn't built from playing material handicap games, so I don't understand how playing a different game altogether is better to people. Like, you don't see tournaments between humans where they're all set at the same level by removing their material (so one has to play against a Magnus Carlsen that has his two Bishops removed; something like this was tried at chesscube without much success), why is computer-human games different?

Time handicaps are natural, lichess allows users to go "berserk" by cutting their time in half, this is clearly still chess played at different levels, but when you make the engine play high chess level moves instantly, you don't let the human play at the level they'd play against an opponent that required that time to find that great move (unlike engines.) I hope move delay's effects get investigated more.
Time handicaps are even more "natural" when engines are playing (than when both sides are humans), because the software has always been decoupled from the hardware (and you can have ponder off if you want, etc.). Time is equivalent to computing power, and there's never been a fixed "natural" amount of computing power.

Material handicaps are less "unnatural" for humans than for engines, because humans have real intelligence, so they can at least partially adapt to the change, while the engines have no intelligence, so they won't even try to adapt. Whether a chess engine plays relatively well with a handicap is then largely a fluke.
I disagree.

Engines are programmed by humans and I believe tournaments with odds may be interesting for the programmers.
I think that it may be interesting also for people who are not programmers in order to know which engine is better with material odds because it is better to analyze with an engine that is better in this situation.

I also believe that there should be a rating of the top 6 engines playing odds versus some weaker engines such as knight odds, and f7 and a move odds etc...... to evaluate which engines can handle a losing situation better and avoid trading pieces and instead create more complications to the weaker engines.

PS I am sorry to have called people between the ages 55 to 66 old, they are simply over a century old that feel full of spirit and are healthy which is that matter the most. I only wanted to point out their ages which is related to their experiences and some of them have beaten players that are between 1900 thru 2050 a few times during their chess career.

jp
Posts: 603
Joined: Mon Apr 23, 2018 5:54 am

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

Post by jp » Mon Jul 08, 2019 11:35 am

lkaufman wrote:
Sun Jul 07, 2019 12:16 am
jp wrote:
Sat Jul 06, 2019 11:37 pm
Ovyron wrote:
Sat Jul 06, 2019 3:26 am
lkaufman wrote:
Wed Jul 03, 2019 5:54 am
I personally prefer material
Except that's no longer chess. The engine wasn't built with handicap in mind, the human's elo wasn't built from playing material handicap games, so I don't understand how playing a different game altogether is better to people. Like, you don't see tournaments between humans where they're all set at the same level by removing their material (so one has to play against a Magnus Carlsen that has his two Bishops removed; something like this was tried at chesscube without much success), why is computer-human games different?

Time handicaps are natural, lichess allows users to go "berserk" by cutting their time in half, this is clearly still chess played at different levels, but when you make the engine play high chess level moves instantly, you don't let the human play at the level they'd play against an opponent that required that time to find that great move (unlike engines.) I hope move delay's effects get investigated more.
Time handicaps are even more "natural" when engines are playing (than when both sides are humans), because the software has always been decoupled from the hardware (and you can have ponder off if you want, etc.). Time is equivalent to computing power, and there's never been a fixed "natural" amount of computing power.

Material handicaps are less "unnatural" for humans than for engines, because humans have real intelligence, so they can at least partially adapt to the change, while the engines have no intelligence, so they won't even try to adapt. Whether a chess engine plays relatively well with a handicap is then largely a fluke.
I don't think it's a fluke at all. For normal engines, it's a function of how much effort the programmers have put into the best way to play when up or down material. For NNs, it seems to depend on the details of how the NN was trained; I'm no expert on this, but perhaps it has to do with how often materially unbalanced positions occur in the training, and the depth of the training games. I would guess that training using very low node counts would produce better handicap play than normal training, because making the best moves when up or down a piece (for example) won't change the result much with deep searches. The losing player has to have some chance to save the game.
"Fluke" may not be the best word, but just that later, stronger versions of the same program can be worse at handicaps... It'd be different if programmers were trying to make engines good at handicaps.

lkaufman wrote:
Fri Jul 05, 2019 4:14 pm
As for Lc0, their first series of networks, culminating in 11248, is very good at giving handicaps; it scored about 70% in blitz games giving handicaps averaging around knight odds to GM Naroditsky, an excellent blitz player. However more recent networks are just awful at giving knight odds, they just make blunders. Lc0 just plays awful chess when the winning expectancy gets down below 2% or so. For the old networks that only happens around queen odds, but for the new ones it happens even below knight odds.

Chessqueen
Posts: 510
Joined: Wed Sep 05, 2018 12:16 am
Full name: Nancy M Pichardo

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

Post by Chessqueen » Mon Jul 08, 2019 7:50 pm

lkaufman wrote:
Sat Jul 06, 2019 3:56 am
Ovyron wrote:
Sat Jul 06, 2019 3:26 am
lkaufman wrote:
Wed Jul 03, 2019 5:54 am
Regarding your delay suggestion, is this just to create the illusion of a normal game?
No, it's because the slower the human's opponent plays, the better the human plays.

Try some extreme case:

Human 1 gets 1 minute on the clock.
Human 2 gets 2 hours on the clock.

What is human's 2 best strategy? Well, it turns out that the longest time used by human 2, the better human 1's level will be, so that in a normal rhythm, human 1 will play at a level much higher than 1 0 bullet. Human 2 would need to find a sweet spot where their own moves benefit from the extra time, but they're played at a rhythm that doesn't allow the opponent to increase their "bullet level" (which is incredibly low. I have actually checked, and my 10 0 blitz level is higher than Nakamura's 1 0 level. But if we played time handicap he'd destroy me, why? Because I can't play against his 1 0 level, because his level would increase tremendously as he waits for my moves, getting close to his 10 0 level.)

With computers it's similar (except that it doesn't work the other way around, with computers being really bad at pondering) so human with 1 0 on the clock against computer with 1 0 on the clock would play at their bullet level, but at 1 0 for them against 120 min for the comp, humans would play at a much higher level, as long as the comp plays the same moves it'd have played with 1 minute on the clock (these are just the extremes). Move delay solves for all this.
lkaufman wrote:
Wed Jul 03, 2019 5:54 am
I personally prefer material
Except that's no longer chess. The engine wasn't built with handicap in mind, the human's elo wasn't built from playing material handicap games, so I don't understand how playing a different game altogether is better to people. Like, you don't see tournaments between humans where they're all set at the same level by removing their material (so one has to play against a Magnus Carlsen that has his two Bishops removed; something like this was tried at chesscube without much success), why is computer-human games different?

Time handicaps are natural, lichess allows users to go "berserk" by cutting their time in half, this is clearly still chess played at different levels, but when you make the engine play high chess level moves instantly, you don't let the human play at the level they'd play against an opponent that required that time to find that great move (unlike engines.) I hope move delay's effects get investigated more.
I understand your point about move delay, but let's do some math. Suppose that we are playing a ten minute game, with the computer actually thinking only a few milliseconds but programmed to delay so as to use about equal time to the human. Let's say that thinking on the opponent's time is worth 30% of thinking on our own time (I think this is roughly what ponder stats say). So the human is really getting 13 minutes instead of just ten this way. But we could also increase his strength the same amount by making it a 13 minute game, with the computer moving instantly, while cutting the total time for the game from 20 minutes to 13 minutes. So from the perspective of both spectators and the player himself, there is no advantage to this delay idea, just raise the time limit 30%.
Regarding material handicaps, it was standard to run tournaments this way in the 1800s, and we did run some rapid events this way even in the 1970s. Garry Kasparov played a widely publicized match giving two pawn odds. My guess is that you also don't like chess960, it's "not chess", but FIDE is putting hundreds of thousands of dollars into a chess960 world championship. I agree that it wouldn't be very interesting to watch a computer or Magnus Carlsen give queen odds to a patzer, but when the odds receiver is a titled player, even a GM, and still has trouble winning, it is really amazing to see how this is possible. I would find it very interesting to see Carlsen give f7 odds in a serious match to a 2500 GM, for example. Judging by how he saves so many bad or lost positions with Black in normal chess against the Elite, I would bet on him.
I found this Amazing that Leela Chess Zero gave a Knight Odds to this Grandmaster and was able to draw, and the most amazing thing was the masterpiece of an endgame that was created by LC0.

PS: I believe that it will take another 10 years for any engine to give a Rook Odds to any strong GM.
https://www.youtube.com/watch?v=dsz7_Al1dXM

lkaufman
Posts: 3438
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 » Tue Jul 09, 2019 3:52 pm

Chessqueen wrote:
Mon Jul 08, 2019 7:50 pm
lkaufman wrote:
Sat Jul 06, 2019 3:56 am
Ovyron wrote:
Sat Jul 06, 2019 3:26 am
lkaufman wrote:
Wed Jul 03, 2019 5:54 am
Regarding your delay suggestion, is this just to create the illusion of a normal game?
No, it's because the slower the human's opponent plays, the better the human plays.

Try some extreme case:

Human 1 gets 1 minute on the clock.
Human 2 gets 2 hours on the clock.

What is human's 2 best strategy? Well, it turns out that the longest time used by human 2, the better human 1's level will be, so that in a normal rhythm, human 1 will play at a level much higher than 1 0 bullet. Human 2 would need to find a sweet spot where their own moves benefit from the extra time, but they're played at a rhythm that doesn't allow the opponent to increase their "bullet level" (which is incredibly low. I have actually checked, and my 10 0 blitz level is higher than Nakamura's 1 0 level. But if we played time handicap he'd destroy me, why? Because I can't play against his 1 0 level, because his level would increase tremendously as he waits for my moves, getting close to his 10 0 level.)

With computers it's similar (except that it doesn't work the other way around, with computers being really bad at pondering) so human with 1 0 on the clock against computer with 1 0 on the clock would play at their bullet level, but at 1 0 for them against 120 min for the comp, humans would play at a much higher level, as long as the comp plays the same moves it'd have played with 1 minute on the clock (these are just the extremes). Move delay solves for all this.
lkaufman wrote:
Wed Jul 03, 2019 5:54 am
I personally prefer material
Except that's no longer chess. The engine wasn't built with handicap in mind, the human's elo wasn't built from playing material handicap games, so I don't understand how playing a different game altogether is better to people. Like, you don't see tournaments between humans where they're all set at the same level by removing their material (so one has to play against a Magnus Carlsen that has his two Bishops removed; something like this was tried at chesscube without much success), why is computer-human games different?

Time handicaps are natural, lichess allows users to go "berserk" by cutting their time in half, this is clearly still chess played at different levels, but when you make the engine play high chess level moves instantly, you don't let the human play at the level they'd play against an opponent that required that time to find that great move (unlike engines.) I hope move delay's effects get investigated more.
I understand your point about move delay, but let's do some math. Suppose that we are playing a ten minute game, with the computer actually thinking only a few milliseconds but programmed to delay so as to use about equal time to the human. Let's say that thinking on the opponent's time is worth 30% of thinking on our own time (I think this is roughly what ponder stats say). So the human is really getting 13 minutes instead of just ten this way. But we could also increase his strength the same amount by making it a 13 minute game, with the computer moving instantly, while cutting the total time for the game from 20 minutes to 13 minutes. So from the perspective of both spectators and the player himself, there is no advantage to this delay idea, just raise the time limit 30%.
Regarding material handicaps, it was standard to run tournaments this way in the 1800s, and we did run some rapid events this way even in the 1970s. Garry Kasparov played a widely publicized match giving two pawn odds. My guess is that you also don't like chess960, it's "not chess", but FIDE is putting hundreds of thousands of dollars into a chess960 world championship. I agree that it wouldn't be very interesting to watch a computer or Magnus Carlsen give queen odds to a patzer, but when the odds receiver is a titled player, even a GM, and still has trouble winning, it is really amazing to see how this is possible. I would find it very interesting to see Carlsen give f7 odds in a serious match to a 2500 GM, for example. Judging by how he saves so many bad or lost positions with Black in normal chess against the Elite, I would bet on him.
I found this Amazing that Leela Chess Zero gave a Knight Odds to this Grandmaster and was able to draw, and the most amazing thing was the masterpiece of an endgame that was created by LC0.

PS: I believe that it will take another 10 years for any engine to give a Rook Odds to any strong GM.
https://www.youtube.com/watch?v=dsz7_Al1dXM
Note that this was a blitz (5' + 2") game. In blitz (3' + 2") games giving knight odds Komodo was able to score about 50% in many games with GM Alex Lenderman several years ago. But in fairly slow games (45' + 15") no computer has scored over 25% against any titled player yet at knight odds, even with ones around 2100 FIDE strength. The time limit makes all the difference.
Komodo rules!

User avatar
Ovyron
Posts: 2143
Joined: Tue Jul 03, 2007 2:30 am

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

Post by Ovyron » Tue Jul 09, 2019 5:13 pm

jp wrote:
Sat Jul 06, 2019 11:37 pm
Material handicaps are less "unnatural" for humans than for engines, because humans have real intelligence, so they can at least partially adapt to the change
"The change" is a change to a different game, with starting positions never seen in actual chess games, and completely different concepts and requirements than normal chess. I'm not even sure the strength of the human matters in odds, because, who knows if I'm the strongest human in the world at Knight odds? (my opponent starts without a knight) it's just that when I don't have such a material advantage I suck, and I suck at getting it. The point is no Master titles exist for material odds, so at this point it makes no sense to find strong human players for a match (you'd want to find those that are best at taking advantage of the odds. At bullet I just saw Nakamura blundering all over the place with material advantage, some other players may do it better, but they're worse at getting that advantage in the first place.)

That's why I suggested at some point that at the very least the machine should play really badly until the human gains an advantage that is equal to starting the game with the material odds, this will be natural and will have the human having plans already in place when the engine switches to full strength mode.

This is how Chessmaster handicap personalities worked, anyway, though, intermittently (full strength most of the time, a blunder now and then.)

It could work as follows:

Have two Komodos running at the same time. They play until out of book (for variety). Then, one Komodo plays at Depth 1 (or whatever is deemed appropriate - I have no problems beating this depth, so the strong human could demolish it). The other Komodo checks the positions seeking for an evaluation point at which the advantage is the same as in material handicap, and once this happens, Komodo will play at full strength.

This would be natural and organic handicap. Like a "drunken Komodo" that snaps out of it after realizing it's being beaten badly, and interesting for the human that would try to reach a position with an advantage big enough to win once it's as big as the material handicap.
Make someone happy today.

jp
Posts: 603
Joined: Mon Apr 23, 2018 5:54 am

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

Post by jp » Tue Jul 09, 2019 5:44 pm

Ovyron wrote:
Tue Jul 09, 2019 5:13 pm
That's why I suggested at some point that at the very least the machine should play really badly until the human gains an advantage that is equal to starting the game with the material odds, this will be natural and will have the human having plans already in place when the engine switches to full strength mode.

It could work as follows:

Have two Komodos running at the same time. They play until out of book (for variety). Then, one Komodo plays at Depth 1 (or whatever is deemed appropriate - I have no problems beating this depth, so the strong human could demolish it). The other Komodo checks the positions seeking for an evaluation point at which the advantage is the same as in material handicap, and once this happens, Komodo will play at full strength.

This would be natural and organic handicap. Like a "drunken Komodo" that snaps out of it after realizing it's being beaten badly, and interesting for the human that would try to reach a position with an advantage big enough to win once it's as big as the material handicap.
I have no idea why people would want to do something like that. What would it tell us about chess, or the way humans play chess, or the way computers play chess? Almost nothing, I'd guess.

I'm not sure that material handicaps tell us much about chess, either.

Limiting the computer's hardware resources is natural and could tell us something about chess.

User avatar
Ovyron
Posts: 2143
Joined: Tue Jul 03, 2007 2:30 am

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

Post by Ovyron » Tue Jul 09, 2019 6:13 pm

jp wrote:
Tue Jul 09, 2019 5:44 pm
I have no idea why people would want to do something like that.
Because at least it's only one side playing stupidly (drunken engine). This is how the start of the game of material handicap looks like:



What's the problem with those players??

Here's what I propose looks like:



Here, Komodo at full strength is switched on, because it thinks white's advantage is as big as in the material handicap position.

Note this was human as black playing the best moves that they thought could play (knowing engine will play badly and would reach big advantage no matter what), it may be a good idea to reach a position they think they can't lose (and if so, then the advantage given was too much and the engine should play at full strength sooner.)

This is different from the first game where black would play 4...Nxh5?? allowing the rook to capture the knight, and 5.Nc3?? where white doesn't punish it. So it's no longer chess.

In the second game, black did no such thing, it's still chess, drunken Komodo had a choice of what to play, and played as best as it could once out of book, but with its search limited.

At least we could learn something about the Silician Dragon, I don't think 1. h3 Nf6 2. g4 Nxg4 3. h4 Nf6 would ever be played by people trying to win.
Make someone happy today.

jp
Posts: 603
Joined: Mon Apr 23, 2018 5:54 am

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

Post by jp » Tue Jul 09, 2019 6:33 pm

Ovyron wrote:
Tue Jul 09, 2019 6:13 pm
jp wrote:
Tue Jul 09, 2019 5:44 pm
I have no idea why people would want to do something like that.
Because at least it's only one side playing stupidly (drunken engine). This is how the start of the game of material handicap looks like:

Well, I'm not one to push for material handicaps, but I wouldn't feel that a handicap game must have a crazy back-story or history that connects it the chess starting position to justify itself.

How many moves out of book is the second example? I don't think the moves between the end of book and the end of move 16 must be "played" by the engine. It seems a bit artificial to claim they are. We could produce them by any means and just start at the position at the end of move 16, which would be fine.

User avatar
Nordlandia
Posts: 2325
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

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

Post by Nordlandia » Tue Jul 09, 2019 7:10 pm

Ovyron: the days when human can compete against the computer in the classical starting position is by a long shot over. Material odds or cripple the hardware is needed to give the human any practical chances.

lkaufman
Posts: 3438
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 » Tue Jul 09, 2019 8:38 pm

Here is a practical proposal. Anyone reading this with a chess.com blitz rating over 2200 (arbitrary, I know) who would like to play handicap games with Komodo for public viewing on chess.com please post your name, rating, and what conditions (time limit, handicap) would be ok for you. As long as there is no money involved, I can arrange it myself. I trust that no one would cheat with no money on the line in this situation, I mean why would a cheater want a handicap? Any material and/or move handicap is possible as long as it just requires stating the initial start position, no special conditions. If several people reply, I'll choose from among the most interesting. I think that most interesting is knight odds, alternating b1 and g1, with a time limit depending on the human's rating. Once we have agreed on terms and starting time, I'll post that info, and all you have to do is look for the games of "Playkomodo". Note that for those interested in handicapping by limiting Komodo's depth of search, you can challenge the top komodo levels and if you wish post when you plan to play the games.
Komodo rules!

Post Reply