[pps43]

I thought people stopped doing Centaur chess when it turned out that human+computer usually loses to computer alone.

There are still positions that people evaluate correctly but computers misjudge, but those are pretty rare.

[posterboy]

Pretty is nice, tell more please.

[pps43]

As early as 2013: http://www.businessinsider.com/computers-beating-humans-at-a...

Next year: https://www.chess.com/news/view/stockfish-outlasts-nakamura-...

And computers were only getting better since then.

[xoroshiro]

It's not as far as people might think though.

If I recall correctly, Nakamura used an older version of a weaker chess engine (Rybka). He also tried to force a relatively equal position (since he was behind in points anyway) which led him to lose because of risky play.

I wouldn't be surprised at all if Stockfish would lose to Stockfish + Nakamura or any of the leading chess players. These people have been using chess engines for analyses for quite a while now and certainly know their limitations.

[pps43]

The key idea of human+computer is that human does strategy and position evaluation, and computer calculates tactics and acts as a safety net to prevent human blunders. A weaker chess engine would still perform that function.

Granted it's all conjecture since no large competitions took place, but the mere fact that no such competitions took place suggests that the outcome probably won't be pleasant for humans.

[xoroshiro]

My point was tat with the same chess engine, man with machine beats machine.

I do see your point about position evaluation though. And yes, it would be interesting to see a tournament.

[pps43]

I actually have not seen any recent cases when a human with chess engine was beating the same chess engine more often than a random chance would imply.

It used to be the case before about 2012-2014, but now the gap is so large that a human can only make things worse.