Friday 23 June 2023

080. Stockfish vs. Stockfish


Black: I. Rees - C&DCCC Ward-Higgs Trophy, 2023

Towards the end of 2022 I was minded to respond to an article in the New York Times, in which the author wrote: “Out of the 136 games played in the 32nd World Correspondence Chess Championship, 119 were draws.”

Looking at the crosstable you can see that that 119 is superficially correct. However, some qualification seemed warranted, so I tweeted:

Except that Bock withdrew, so all his "losses" can be discounted; similarly Neto-Pessoa where Black "resigned" a level position. So out of 119 real results, only *three* were decisive. Of these, two were one-move blunders (Edwards-Michálek and Schwetlick-Lecroq). Only Osipov-Schwetlick was a win "over the board". Basically Edwards came first because Michálek, who gave him a knight for nothing, only lost one game, whereas Schwetlick lost two. Well done.

Someone then asked: “Do the humans have any input or is it all down to computers?”
To which I replied:

There's a surprising amount of leeway in most positions, so humans can direct how they want play to go, while engines keep it within the bounds of soundness. But as a real contest between humans it's ultimately futile: if a position is at all sound, Stockfish will defend it.

We went into this a bit more in a thread on the ChessPublishing forum, which included a discussion of the strength (or otherwise) of a “centaur” (that is, a human using an engine) compared to that of an engine alone. My opinion was (and is):

That the centaur might be very slightly stronger relies on quite a lot of expertise: that the human knows how to use the engine, does so diligently, and is sensible enough not to overrule it in critical positions.

But note that I said "not significantly stronger" above. I meant that in a mathematical sense; i.e. there's no significant statistical difference; i.e. even in a best case scenario the centaur is hardly any stronger.

To put it in personal terms: I'm a CCSIM (title from 20 years ago), currently rated ca. 2200 over the board, and have been analysing with engines for many years. But if a CC game is, say, me + Stockfish vs. 1200 player + Stockfish – a prime example of centaur vs. engine – I know from the outset that, if the opening is sound and all my opponent's subsequent moves are approved by Stockfish, there's virtually no chance I'm going to win.


Hence, as I also said, and as I've indicated before: “I stopped playing serious CC in 2006. My only games are a couple each year for Nottinghamshire in the C&DCCC.” The game below is from this season's competition.

There is nothing interesting about this game. I hoped my opponent might play 3...f5. He didn't. He played the Berlin. I tried a line that Magnus had won with, and then – once Black had deviated – an odd-looking idea (14 Qc1) from an earlier correspondence game.
It didn't get me anywhere. The game was drawn. The companion game was slightly more entertaining but still a draw. My two games the previous season, against someone rated 1650 OTB (roughly 130 in old money) were also draws.

To reiterate: As a real contest between humans, correspondence chess is ultimately futile: if a position is at all sound, Stockfish will defend it.

Or to put it another way: Stockfish vs. Stockfish is a draw.


No comments:

Post a Comment