Stockfish play very decent giving pawn odds.........

[Uri Blass]

My question is if Carlsen need three takeback moves, how many takeback will a master rated player needs at lest 20 or he will need to replay the same game one hundredth time until he draw .

Maybe Carlsen needs 20 to draw. Maybe 100 aren't enough. We can't even start to speculate because this terrain hasn't been explored at all.

Also, as a matter of depth, I can beat Stockfish depth 1 with no takebacks - so there's some Stockfish depth that I can beat with 3 takebacks, and it has some time handicap equivalent (assuming I get 10 fixed minutes per game, and Stockfish gets whatever time it takes it to reach that depth. Which could be 5 seconds/game or something.)

I think that you need to define the time control and number of take back that you allow that can be also in theory infinite.

The simplest thing against humans is to have a rule that stockfish use the same time that the human use and the human has some limited time.
With fixed 10 minutes per game I guess that part of the humans are going to lose on time even with infinite number of takebacks and I am not sure how many can win(I guess many can get a draw).
The problem is that if the human try to take back too many moves he is going to use more than 10 minutes and lose on time.

The problem is also that I do not know of some interface that support the idea and you also need to modify stockfish to use the same time that the human use and you need interface that allow the human to click take back during his move so it take back both stockfish's last move and the human last move and the interface also know that the human can lose on time.

Note that time that the human use include the time that he used for take back.

Example suppose you play against stockfish.

If after 1.e4 e5 2.Nf3 Nc6 3.Bb5 it is your move and you use 3 seconds and click take back
then you see the position after 1.e4 e5 Nf3
If after 2 additional seconds you click another take back you get the position after 1.e4 and if you think additional 4 seconds and play 1...c5
then your total time for 1...c5 is 3+2+4=9 seconds and stockfish is going to use 9 seconds for the reply.

Uri, another thing that I noticed is that no matter how many takebacks a GM takes against any top engine rated above 3450 the replacement move will always be played at his strength, and either Stockfish, or Komodo will always reply to the human replacement move with a better move since those top engines look at the replacement move by the human player and recalculate the best response to the replacement move made by the human. I will give you an example, I try yesterday a friendly game against my friend Jorge Sammour rated FIDE 2458 and asked him if I can take 15 or 20 takebacks during the game. So he agreed and we played a 30 minutes game and whenever I told him that I would like a takeback to any of my move, I stopped the clock and started the clock after I replaced the move as if the time froze from that moment so my thinking time for the replacement move did not counted at all. I lost the game even after taking 20 takebacks. The explanation is simple whenever I made a replacement move I still made it at my strength and he always found a better move being way way a better player by at least 450 rating points than I am. During the game he did not made a comment at all that was our agreement, but he was taking notes so he could comment later on and told me why my replacement move was still weak or slightly better but not the best one. I am too embarrassed to post the game here, but I learned a lot more than reading an entire book of chess by listening to his explanation and comment after the game. Conclusion by taking x amount of takebacks the human player will NOT defeat the top engine, since the response of the engine will always be superior to the human, and a human can NOT think of hundred of moves every time it calculate, but the engine does every single time.Therefore, takebacks is not the way to defeat a top engine unless you replace your brain with another top engine of equal strength every time you replace your moves.

Note that with infinite take back you can take back not only a single move but many moves and not because you think that you made a mistake but because you feel that you do not understand the position.

The idea is that hopefully you can use another line to get a position that you understand better or a position that the computer does not understand in order to get a better result.

I am not sure that with infinite take back the top players are not able to win against stockfish.

[Uri Blass]

Takeback odds with some amount of allowed takebacks are great to fix short-term blunders that the master can understand quickly after seeing the engine reply. It won't fix a long-lasting positional weakness, but it's obvious that takeback odds will noticeably increase the elo performance of any human chess player.

The main problem is that chess isn't easily sorted into two categories, "tactical" and "positional". In general, engines beat top players because they see that they can achieve positional advantages by tactical means. The GM may evaluate the positions as well as or better than the engine, but they cannot see the long narrow variations needed to achieve a superior position. So yes, takebacks will raise the human elo, but only by a small fraction of what is needed to be competitive with top engines. I suppose the most natural takeback handicap would be to allow every move to be taken back, once only, until you make the next move, with no time recovery. I would imagine this is close to a class handicap, enough to make a match between the top rated woman Hou Yifan and Carlsen reasonably competitive. Or Carlsen with some engine with a human rating around 3060, so maybe 2950 or so on CCRL 40/15 list. Give him White every game and no opening book for engine, and we're maybe up to 3100 on that list.

It is clear that the most natural takeback handicap is too small against humans and it is the reason that I suggest the infinite take back handicap that is infinite in theory in number of take backs but not infinite practically because the human can lose on time(you use no increment).

[Ovyron]

With fixed 10 minutes per game I guess that part of the humans are going to lose on time even with infinite number of takebacks

The idea is that a take back sets back your clock to where it was before you made the move, so along with the position your clock is as if your other move was never played.

Infinite takebacks wouldn't work in this concept (game would last forever.) If the GM realizes it's losing, then the value of the take backs can vary (say, if they start with 20 takebacks, and 10 moves ago they blundered, they have to burn half their takebacks to go back to a normal position.)

I believe lichess already has takebacks implemented as wished, the players would just need to agree before hand how many will be allowed, and the side with the engine would just stop accepting them once the quota has been used.

[lkaufman]

With fixed 10 minutes per game I guess that part of the humans are going to lose on time even with infinite number of takebacks

The idea is that a take back sets back your clock to where it was before you made the move, so along with the position your clock is as if your other move was never played.

Infinite takebacks wouldn't work in this concept (game would last forever.) If the GM realizes it's losing, then the value of the take backs can vary (say, if they start with 20 takebacks, and 10 moves ago they blundered, they have to burn half their takebacks to go back to a normal position.)

I believe lichess already has takebacks implemented as wished, the players would just need to agree before hand how many will be allowed, and the side with the engine would just stop accepting them once the quota has been used.

I think that for any human to be competitive with top engine by takebacks, he would have to try so many lines with so many takebacks that it could hardly be considered as his game, it's just a test of the ability to use engine analysis rapidly in ways that might be better than just looking at the engines final move choice. It's sort of like the fact that correspondence players using a top engine can generally score more than 50% against the same engine unaided. It is certainly a skill, but not one that has much correlation with chess strength. Now if we just allow some small number of takebacks for the whole game, or even unlimited takebacks of the latest move until you make another move, that probably correlates a lot more with chess strength, but then it isn't nearly enough by itself, it has to be combined with other advantages.

[Ovyron]

It is certainly a skill, but not one that has much correlation with chess strength.

Ha! That's what I've been saying about the starting positions with missing handicap pieces for the engine side all along.

"The human drew against the engine after using 10 takebacks to see what it would move and find a drawing plan" sounds more natural to me than "the human drew against the engine which started the game with its a2 and f2 pawns missing". At least with takebacks the human had to earn the draw with those skills, instead of being given 2 free pawns without doing anything.

At least with takebacks the human gets to decide how they're used, instead of the engine side deciding what pawns to give.

We have to accept that this isn't chess anymore, as in chess many games, the most beautiful, are won by a side low on material but with positional compensation, and material handicap goes completely against those chess principles. If it's not chess then we can arrange something that works, which hopefully starts from the opening position of chess, at the very least.

[Uri Blass]

It is certainly a skill, but not one that has much correlation with chess strength.

Ha! That's what I've been saying about the starting positions with missing handicap pieces for the engine side all along.

"The human drew against the engine after using 10 takebacks to see what it would move and find a drawing plan" sounds more natural to me than "the human drew against the engine which started the game with its a2 and f2 pawns missing". At least with takebacks the human had to earn the draw with those skills, instead of being given 2 free pawns without doing anything.

At least with takebacks the human gets to decide how they're used, instead of the engine side deciding what pawns to give.

We have to accept that this isn't chess anymore, as in chess many games, the most beautiful, are won by a side low on material but with positional compensation, and material handicap goes completely against those chess principles. If it's not chess then we can arrange something that works, which hopefully starts from the opening position of chess, at the very least.

The question is that if the skill of getting a good result with material handicap has more correlation with chess strength relative to the skill of getting good results with many take back.

The claim of larry is that the skill of getting a good result with material odds has more correlation with chess strength.
Knowing if he is right or wrong can be done by testing and I think that it may be interesting to have a tournament between humans when the strong humans play without a knight against the weaker humans and another tournament with no material handicap when the weaker players are allowed many take back and see if the correlation between rating and results is clearly higher in the case of a knight handicap.

[Ovyron]

The claim of larry is that the skill of getting a good result with material odds has more correlation with chess strength.

This is nullified by the inverse correlation, i.e. a side that sacs a knight to win the chess game will win it a knight down, as opposed to the human which is supposed to beat the engine with a knight up.

I'd like to see how many top level humans beat others because they managed to win an entire knight, this seems like an artificial construction because chess games are mostly never won because the losing side dropped a piece (like engines that start the game as if they already dropped a piece.)

[Chessqueen]

With fixed 10 minutes per game I guess that part of the humans are going to lose on time even with infinite number of takebacks

The idea is that a take back sets back your clock to where it was before you made the move, so along with the position your clock is as if your other move was never played.

Infinite takebacks wouldn't work in this concept (game would last forever.) If the GM realizes it's losing, then the value of the take backs can vary (say, if they start with 20 takebacks, and 10 moves ago they blundered, they have to burn half their takebacks to go back to a normal position.)

I believe lichess already has takebacks implemented as wished, the players would just need to agree before hand how many will be allowed, and the side with the engine would just stop accepting them once the quota has been used.

I think that for any human to be competitive with top engine by takebacks, he would have to try so many lines with so many takebacks that it could hardly be considered as his game, it's just a test of the ability to use engine analysis rapidly in ways that might be better than just looking at the engines final move choice. It's sort of like the fact that correspondence players using a top engine can generally score more than 50% against the same engine unaided. It is certainly a skill, but not one that has much correlation with chess strength. Now if we just allow some small number of takebacks for the whole game, or even unlimited takebacks of the latest move until you make another move, that probably correlates a lot more with chess strength, but then it isn't nearly enough by itself, it has to be combined with other advantages.

We saw Komodo 9( f7 odds ) playing Versus Nakamura, Now what we need to do if play Komodo 14.x Vs Nakamura with the weakest computer that Komodo can play with, NOT your 32 cores, but probably a an intel 486 or the first Pentium system which you can purchase for $99.00 on Ebay.

[Chessqueen]

With fixed 10 minutes per game I guess that part of the humans are going to lose on time even with infinite number of takebacks

The idea is that a take back sets back your clock to where it was before you made the move, so along with the position your clock is as if your other move was never played.

Infinite takebacks wouldn't work in this concept (game would last forever.) If the GM realizes it's losing, then the value of the take backs can vary (say, if they start with 20 takebacks, and 10 moves ago they blundered, they have to burn half their takebacks to go back to a normal position.)

I believe lichess already has takebacks implemented as wished, the players would just need to agree before hand how many will be allowed, and the side with the engine would just stop accepting them once the quota has been used.

I think that for any human to be competitive with top engine by takebacks, he would have to try so many lines with so many takebacks that it could hardly be considered as his game, it's just a test of the ability to use engine analysis rapidly in ways that might be better than just looking at the engines final move choice. It's sort of like the fact that correspondence players using a top engine can generally score more than 50% against the same engine unaided. It is certainly a skill, but not one that has much correlation with chess strength. Now if we just allow some small number of takebacks for the whole game, or even unlimited takebacks of the latest move until you make another move, that probably correlates a lot more with chess strength, but then it isn't nearly enough by itself, it has to be combined with other advantages.

We saw Komodo 9( f7 odds ) playing Versus Nakamura, Now what we need to know is if play Komodo 14.x Vs Nakamura with the weakest computer that komodo can play with, NOT your 32 cores, but probably on a AMD HP Pavilion Slimline Desktop PC - AMD E1-2500 (1.40 GHz) can still beat Nakamura with the f7 pawn removed at 45 minutes. And you can even get this system for $99.00. https://www.ebay.com/itm/HP-PAVILION-SL ... 4611684320. This system and the latest Komodo 14.x can probably beat any GM rated 2495 to 2600 FIDE with c2, f2 removed and 45+ 5 to the GM and Komodo with a time handicap of only game in 15+ 10 and on top of that with one takeback duing the entire game.

PS: Now My question is why do we all need to purchase the most expensive computer when a simple computer can beat 97% of us with any of the 15 top engines? I do NOT know what computer system was Komodo 9 using vs Nakamura here , but with the latest Komodo 14.x we can get the same result with this weak system ==> https://www.ebay.com/itm/HP-PAVILION-SL ... 4611684320

[lkaufman]

With fixed 10 minutes per game I guess that part of the humans are going to lose on time even with infinite number of takebacks

The idea is that a take back sets back your clock to where it was before you made the move, so along with the position your clock is as if your other move was never played.

Infinite takebacks wouldn't work in this concept (game would last forever.) If the GM realizes it's losing, then the value of the take backs can vary (say, if they start with 20 takebacks, and 10 moves ago they blundered, they have to burn half their takebacks to go back to a normal position.)

I believe lichess already has takebacks implemented as wished, the players would just need to agree before hand how many will be allowed, and the side with the engine would just stop accepting them once the quota has been used.

I think that for any human to be competitive with top engine by takebacks, he would have to try so many lines with so many takebacks that it could hardly be considered as his game, it's just a test of the ability to use engine analysis rapidly in ways that might be better than just looking at the engines final move choice. It's sort of like the fact that correspondence players using a top engine can generally score more than 50% against the same engine unaided. It is certainly a skill, but not one that has much correlation with chess strength. Now if we just allow some small number of takebacks for the whole game, or even unlimited takebacks of the latest move until you make another move, that probably correlates a lot more with chess strength, but then it isn't nearly enough by itself, it has to be combined with other advantages.

We saw Komodo 9( f7 odds ) playing Versus Nakamura, Now what we need to do if play Komodo 14.x Vs Nakamura with the weakest computer that Komodo can play with, NOT your 32 cores, but probably a an intel 486 or the first Pentium system which you can purchase for $99.00 on Ebay.

I don't know what is the point of determining how ancient a computer would have to be for Komodo 14 to drop to the same level as top human players. Obviously there is some year's best hardware that would do it, but is it really interesting to know whether that year is 1988 or 1990 or 1992?
But you are right that we don't really need a $7,000 computer for handicap matches, using one that costs under $1,000 today would probably not hurt results by very much in these events. I don't see much point in using hardware inferior to that, except perhaps there is some interest in how Komodo runs on a cellphone, and I understand that now even good cellphones are as fast as cheap computers.