In chess, AIs are very superhuman; the best players in the world would lose nearly every game against any modern computer player.
Do humans still have something to add? The continued existence of correspondence chess, IMO, suggests that they do. In correspondence chess players have days to make each move, and play from their homes. Due to the impossibility of policing cheating under these conditions, correspondence players are allowed to use computer assistance.
You might think this would make the games just a question of who has more computing power. But as far as I can tell, that’s not the case.
What are humans adding? Low confidence, but I think it’s mostly opening prep; try to find a line that looks ok on shallow computer analysis, but where deeper analysis shows you have an advantage. The human value-add is telling the computer which lines to analyze. Since the chess game tree is so large, advice like this is quite valuable.
Proof that correspondence chess is still played: https://www.nytimes.com/2022/11/09/crosswords/correspondence-chess.html Interview with a human player (from 2016): https://en.chessbase.com/post/better-than-an-engine-leonardo-ljubicic-1-2
On the other hand: I don’t play correspondence chess, so I’m not that confident in the claims above. And some people don’t find them plausible: https://twitter.com/liron/status/1660890927920201728?s=46&t=UlLg1ou4o7odVYEppVUWoQ
Why should we care? This might provide some indication of what value humans can provide in a world of superhuman AI (at least initially).
Can anyone provide a more definitive account of what value, if any, humans add in correspondence chess?
I believe the answer is potentially. The main things which matter in high-level correspondence chess are:
Total amount of compute available to players
Not making errors
Although I don’t think either of those are really relevant. The really relevant bit is (apparently) planning:
(From this interview with Jon Edwards (reigning correspondence world champion) from New In Chess)
I would highly recommend the interview on Perpetual Chess podcast also with Jon Edwards which I would also recommend.
I’ll leave you with this final quote, which has stuck with me for ages:
Interesting. Note that Jon Edwards didn’t win a single game there via play—he won one game because the opponent inputted the wrong move, and another because the opponent quit the tournament. All other games were draws.
Agreed—as I said, the most important things are compute and dilligence. Just because a large fraction of the top games are draws doesn’t really say much about whether or not there is an edge being given by the humans (A large fraction of elite chess games are draws, but no-one doubts there are differences in skill level there). Really you’d want to see Jon Edward’s setup vs a completely untweaked engine being administered by a novice.
I agree. Judging by the fact that AI is strongly superhuman in chess, the only winning strategy is to completely remove the human from the loop, and instead invest in as much compute for the AI as one can afford.
If it’s a sequence that no superhuman AI would consider, this means that the sequence is inferior to the much better sequences that the AI would consider.
It seems that even after 2 decades of the complete AI superiority, some top chess players are still imagining that they are in some ways better at chess than the AI, even if they can’t win against it.
If you look at the actual scenario there, the game was essentially in a stalemate, where the only possible way to win was to force the other player to advance a pawn. Stockfish can’t look 30 moves ahead to see that it’s possible to do that, so would have just flailed around.
You still need stockfish, because without it, any move you make could be a tactical error which the other players computer would pounce on. But stockfish can’t see the greater strategic picture if it’s beyond its tactical horizon.
This seems needlessly narrow minded. Just because AI is better than humans doesn’t make it uniformly better than humans in all subtasks of chess.
I don’t know enough about the specifics that this guy is talking about (I am not an expert) but I do know that until the release of NN-based algorithms most top players were still comfortable talking about positions where the computer was mis-evaluating positions soon out of the opening.
To take another more concrete example—computers were much better than humans in 2004, and yet Peter Leko still managed to refute a computer prepared line OTB in a world championship game.
Yes, humans still provide value. Correspondence chess players will for example read chess opening books to find if there any mistakes in that book and even if they find just one, they’ll try to lead their opponent into that dubious line, which is often a mistake that computers can’t easily spot. Also as a former highly-ranked chess player, I’d use multiple chess engines at the same time to compare and contrast and also I’d know their strengths and weaknesses and which possibilities to explore.
Time should also be a factor when comparing strength between AI alone and an AI-human team. Humans might add to correspondence chess but it will cost them a significant amount of time. Human-AI teams are very slow compared to AI alone.
For example in low latency algorithmic stock trading reaction times are below 10ms. Human reaction time is 250ms. A human-AI cooperation of stock traders would have a minimum reaction time of 250ms (if the human immediatly agrees when the AI suggests a trade), This is way to slow and means a serious competitive disadvantage.
Take this to strategically aware AI compared to a human working with a strategically aware AI. And suppose that the human can improve the strategic decision if given enough time. The AI alone would be at least a 100x faster that the AI-human team. A serious advantage for the AI alone.
For the more mundane human in the loop applications speed and cost will probably be a deciding factor. If chess was a job than most of the time a Magnus Carlson level move in a few seconds for a few cents will be sufficient. In rare cases (e.g. cutting edge science) it might be valuable to go for the absolute best decision at a higher cost in time and money.
So my guess is that human in the loop solutions will be a short fase in the coming transition. The human in the loop fase will provide valuable data for the AI, but soon monetary and time costs will move processes towards an AI alone setup instead of humans in te loop.
Even if in correspondence chess AI-human teams are better it probably does not transfer to a lot of real world applications.
I think you are underrating the number of high-stakes decisions in the world. A few examples: whether or not to hire someone, the design of some mass-produced item, which job to take, who to marry. There are many more.
These are all cases where making the decision 100x faster is of little value, because it will take a long time to see if the decision was good or not after it is made. And where making a better decision is of high value. (Many of these will also be the hardest tasks for AI to do well on, because there is very little training data about them).
True, it depends on the ratio mundane and high stakes decisions. Athough there are high stakes decisions that are also time dependant. See the example about high frequency trading (no human in the loop and the algorithm makes trades in the millions).
Furthermore your conclusion that time independant high stakes decisions will be the tasks where humans provide most value seems true to me. AI will easily be superior when there are time constraint. Absent such constraints, humans will have a better chance of competing with AI. And economic strategic decisions are often times not extremely time constrained (at least a couple of hours or days of time).
In economic situations the amount of high stakes decisions will be limited (only a few people make desicions about large sums of money and strategy) . Given a multinational with a 100.000 employees, only very few will take high stake decisions. But these decisions might have a significant impact on competitiveness. Thus the multinational with a human ceo might out compete a full AI company.
In a strategic situation time might give more of an advantage (i am economist not a military expert so I am really guessing here). My guess would be that a drone without a human in the loop could have a significant advantage (thus pressures might rise to push for high stake decision making by drones (human lives)).