3

I wonder if anyone has tried to organize a large number of much weaker players to cooperatively beat a top player. Of course weaker players may not even see let alone agree on best moves if they use simple voting, but I think theoretically if the team uses some kind of clever distributed analysis protocol to try more possible moves in future positions, and look more moves ahead than an individual weak player can (much like how engines do it), it can compensate for individually slower tactical calculation and less accurate positional intuition.

I imagine the communication overhead will severely limit improvements no matter how many players team together, especially in faster formats, if they are only allowed to communicate in person with physical boards and pen/paper. If electronic communication and shared online chess boards are allowed, they will be more efficient, although it starts to feel like we are just building a chess engine. Let's assume these players understand the rules of the game and can follow directions accurately. Under ideal logistics circumstances and with the best protocol, how many of them would be needed to team together to have even chances against a player who's say 1000 Elo points above them, at least in some format? How about 2000 Elo?

PS: anyone interested in actually designing and testing such a protocol to see if together we can beat a top player?

5
  • chess.stackexchange.com/questions/21828/… You might be interested
    – Allure
    Commented May 21 at 2:19
  • 1
    My experience suggests than a bunch of players cooperating achieve a weaker playing strenght than the strongest of them if s/he was playing alone...
    – Evargalo
    Commented May 21 at 11:52
  • There is a deepmind research paper on this. For alpha zero reinforcement learning. as well as previous machine learning paper on combinations of sub-optimal experts. I understand those to be weaker players. But that was not using chess. The deepming paper is. I need to find the title. I will put that in an answer. Although it might not be your question, I think it puts some light on it. As experimental model.
    – dbdb
    Commented May 23 at 7:00
  • Not quite what you're asking, and much less formal.... But have you tried watching live broadcast commentary of high-level chess matches? For say, the World Chess Championship, you can find teams of incredible players (GM strength, but not world top 10) breaking down the current position and trying to advise which move(s) are best and predicting the players' moves; either with or without consulting an engine.
    – DongKy
    Commented May 25 at 20:27
  • @DongKy Not what this question is about but I did watch the final round of the recent Candidates. Made me lose a bit of faith in humanity to be honest... Commented May 25 at 20:33

1 Answer 1

0

Firstly, a game between the (then) world's top player and other players was organized back in 1999 - Kasparov vs. The World. It wasn't exactly what you're suggesting, but GM's - including some top ones - were involved on the "World" team. While it was a competitive game, in the end, Kasparov won.

As for providing a number, I don't think there is any way to do so. As noted in the question, there would be issues of communication that could hinder the team of GM's. Further, the super GM has earned that "title" because he is one of the best in the world. He or she will see things on the chess board that a garden variety GM will miss; while a collection of GM's may have a better chance of seeing something subtle, there's no guarantee. Further, another issue that could cause problems is how does the group make a decision?

2
  • Human cooperating players might need a reliable language first. And then they might use arguments to deliberate, rather than vote ensemble learning. I guess that is also what you are pointing to.
    – dbdb
    Commented May 23 at 7:06
  • Thanks! I wasn't really thinking about a team with GMs consulting, more like beginner or intermediate players. Lack of organization&good decision making/overhead are all serious issues at any level, but more so for weaker players. I think the best strategy given a huge 1000+ Elo difference is more like a large group of players using some algorithm to act like a chess engine. Perhaps similar to the one AlphaZero uses, with humans generating possible moves and executing Monte-Carlo tree search with alpha-beta pruning, with added optimizations to better suit a team of humans rather than CPUs. Commented May 23 at 21:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.