Is the French defense at a 3700 Elo level "completely refuted"?

Question

...computers will never play the French defense against each other because the French defense at a 3700 Elo level is completely refuted...
_{IM Levy Rozman (YouTube around 3m08s)}

Wait, what? I watched Ding Liren play the French (LiChess) in the world championship thinking "I should play this again", and IM Rozman is like "it's refuted". It's possible this is just an exaggeration for YouTube content. Googling french defense refuted by computers gives this on Reddit:

In the recent TCEC super final both SF and H had the majority of their losses out of playing the French as black.

There seems to be some support for the claim, but still, I'm skeptical. Maybe when IM Rozman says "completely refuted", it means something like "computers think there are better choices, therefore computers don't choose it".

Question: Is the French defense at a 3700 Elo level "completely refuted"?

Answer 1

No

I follow computer chess avidly and so have seen lots of such French games where both engines get massacred as Black.* However, this doesn't mean the French is refuted. With engines as strong as they are, to generate decisive games, it's necessary to introduce highly dubious lines. Called "bias", this leads to openings where engines evaluate the starting position as +1 or more. Naturally, when the opening is so biased, the inferior side will be playing for a draw, without real winning chances (but if they can win with the superior side and draw with the inferior side, they still win the opening 1.5-0.5).

The idea that the French defense is refuted arises because engines often fail to defend high-bias French positions (see example, example). But note these high-bias French positions are high biased. The bookmaker intentionally inserts some inferior moves to create the bias. If you let the engines play the starting position from the French defense 1.e4 e6, you are still overwhelmingly likely to get a draw.

Here's an example of a high bias line.

[FEN ""]
1. e4 e6 2. d4 d5 3. e5 Nc6

Black's last move is a clear inaccuracy, because in the French defense it's imperative for Black to push ...c5. It still shouldn't be a game-losing blunder, however. This lets the stronger engine demonstrate its strength, and is therefore the kind of position that bookmakers give engines. These dodgy French lines are why the opening looks worse than it actually is in engine tournaments.

That said, it's worth pointing out that correspondence chess players don't rate the French defense very highly either. It's not that the French defense is busted, it's just that you'll be suffering for a draw.

*In engine chess tournaments, the opening is pre-determined, and both engines take turns to play the White pieces.

PS: I should add that computer chess moves really quickly. The example games above are only three years old, and yet Stockfish (the losing engine) has gained several hundred ELO in self-play. There's a high chance it can defend those positions against Leela now. As engines get stronger, bookmakers need to insert more and more bias to get decisive games.

Answer 2

For what it's worth, in high-level correspondence play, the French is considered highly suspect. Maybe black can draw with very accurate play, but certainly he has a much harder job than in either 1. e4 e5 or in the Najdorf Sicilian, so most correspondence players don't bother with it anymore.

Another point to consider is that most engines overrate the white side a bit in French structure, since they have a tendency to overrate the impact of a space advantage on a position.

Note that the ICCF (International Correspondence Chess Federation, what FIDE is for OTB chess, ICCF is for correspondence) allows use of engines, which combined with very long time controls (50 days per 10 moves is most usual) leads to very high accuracy.

In human chess, the French is of course fine, there is no direct refutation, even though if white is accurate black will suffer a bit, but so it is with most other openings, except perhaps 1.e4 e5 or th e Najdorf Sicilian. This is what first move advantage is.

A comment to Why is the French defence not played at the highest level? referenced by @Allure: Today in correspondence the best black can get out of any opening is draw (there is one high level black win every few years). With the French, maybe he gets a draw after suffering a lot, at best, if he is lucky.