6

Bobby Fischer:

e4 -- best by test.

What is the current theory or trend in theory behind this statement?

We have seen huge strides taken in chess AI in the past two years. How is this affecting opening theory regarding the "best first move"?

  • IIRC, 1.d4 AlphaZero cycled through a number of openings (e4, c4, ...) as its "best" opening choice, but settled on 1. d4 near the end of its training. But that doesn't really say what is provably "best"... just what some engine happens to think it can play with best. – Mateen Ulhaq Jan 7 at 11:53
  • If chess is the kind of game where a draw is forced with perfect play from both sides (which I believe is widely believed to be the case), then any opening move that doesn't put you into a "losing" game, is equally as good. A similar argument can be made if chess is in fact a game where white always wins with perfect play. – Cruncher Jan 7 at 16:26
  • 1
    @Cruncher If it's the case that White wins with perfect play though, only certain first move(s) might lead to a win, and the others draws. So here the first move you play could matter in this sense. – Inertial Ignorance Jan 8 at 22:27
  • Well of course. That's exactly the same as if it's a drawish game. All moves that keep it drawish are equivalent, and all moves that make it losing are equivalent. This is exactly the case if it's a winning game. All moves that keep it winning are equivalent, all moves that make it drawish are equivalent, and all moves that make it losing are equivalent. I said a similar argument could be made, I didn't think I needed to spell it out – Cruncher Jan 9 at 5:45
12

It's still one of the best moves White can play. There's no clear consensus on whether 1.e4 or 1.d4 is better, but it's played frequently at the top level. Due to advancements in theory, I'd say 1.e4 isn't regarded quite as highly as it was in the past, but again it's most likely White's best/second best move.

Advancements in AI aren't really affecting theory on the first move. Games from AlphaZero/Leela are good for new ways of thinking in the middlegame or late opening, but they're not changing our fundamental opening understanding.

EDIT - also, technically 1.e4 isn't "best by test" currently. In my database it scores 52.8%, while 1.d4 scores 54.4%. But to be fair, 1.e4 has been played in roughly 1.5 million more games, and the types of players who favour 1.e4/1.d4 may differ.

|improve this answer|||||
  • 9
    e4 refuted by AlphaZero would be a great april fool's post. – Michael West Jan 6 at 15:12
  • 7
    For an interesting experiment, try subtracting out all 1. e4 c5 from the e4 games. Last time I did that the e4 score went up almost to the same level as the 1. d4 score was. One possible interpretation of that is that the majority approach to it (open variations following 2 d4 cd) is suboptimal and white should look for a better path. Data analysis often challenges accepted wisdom like that, and doing that sort of analysis is one of the better uses for a chess db. – Arlen Jan 6 at 16:21
  • 1
    @Arlen I think a slightly better conclusion is that 1...c5 is just good, since the open variations are White's best try for a real advantage. – Inertial Ignorance Jan 6 at 16:42
  • A possible conclusion, definitely. But how do you know they're White's best try? Because that's what someone told you? Or because you looked at the data and the positions? Bent Larsen used to call 2 d4 a "cheap trap" and he was strong enough that what he says deserves to be investigated. I don't think either proposition is proven, yet. It's like how Lasker called 3 ... a6 in the Ruy Lopez a "waste of time" yet it was accepted as best for a century until Kramnik dusted off the Berlin and now that's a mainstream option. We should always question whether a move is fashion or quality. – Arlen Jan 6 at 16:59
  • @Arlen Sure, there's no way to definitively prove what I'm saying, but the open variations are most likely White's best try for an advantage. Also, when Kramnik used the Berlin against Kasparov, computers weren't nearly as powerful as they are now (and theory wasn't as extensive). If there was some groundbreaking new idea so early on, it wouldn't be hidden from us. – Inertial Ignorance Jan 7 at 5:54
5

Since 1.e4 and 1.d4 are considered equal by most experts, the value will really depend on your style: If you are a tactical player, you are more likely to get the types of positions that you favor. I am older, and a positional player, so I play 1.d4 for the same reason: I tend to get more positional games, and I am a pretty good positional player.

Whether an opening is truly equal, or even if you have a slight advantage, the real advantage is getting a position that you are comfortable with and have experience with, and that your opponent is less comfortable with. If you are in a tactical position, and the position is technically equal, but you stink at tactics, and are sure to make a mistake, it could be thought of as not equal for you, regardless of whether a computer likes it.

We see Carlsen all the time just trying to avoid book, and get "his" positions.

|improve this answer|||||
  • I would also say that just because the top neural networks don't play 1.e4 doesn't mean that they think it's bad. It simply means that they just prefer 1.d4 or 1.Nf3 or 1. c4. – SubhanKhan Jan 6 at 22:11
  • Is it really the case that AlphaZero doesn't play 1.e4 much? I'm just an interested bystandar, but from reading the AlphaZero paper, the conclusion I came away with is that in self-play training, its most-played opening was the Berlin defense, starting with 1. e4 e5. – amalloy Jan 7 at 0:03
  • @amalloy AlphaZero beated Stockfish with black playing the Berlin Defense. I read the paper and he also was playing the french at first games, but it rejected it as it rejected e4 playing d4 on his match against Stockfish. This was nice for me :/ (irony) as a french defense player and spanish openning player. He refuted all my oppenings (hand to the front). He also refuted my chess idol Fischer "e4 -- best by test" quoted in the question of the poster. – Universal_learner folding home Jan 7 at 0:46
  • @Universal_learner When I look at the Supplemental Data included with the Dec 2018 publication, it lists the positions most arrived at during self-play games, and at the end of its training it was visiting the Berlin most often. This suggests to me that it thinks this opening is fine for white, not just for black - if d4 were better, white wouldn't offer so many e4 openings, right? Am I misinterpreting this data somehow? – amalloy Jan 7 at 1:53
  • @amalloy I might be wrong, but I interpret somewhere, maybe at the end of its study of chess in his 4 million games played against himself, he learned the Berlin Defense was bad for white. That doesn't mean at the beginnig the AI was thinking the Spanish was a good openning for white and that doesn't mean most of his games, even at the end, he played the Spanish as white. He refuted it at the end of the study. The AI played a lot french defense too in the first hours, but somewhere at the last games he played in those 4 hours he refuted it too with white too. – Universal_learner folding home Jan 7 at 2:13
5

No. AI has not changed the evaluation of 1pe4. AI is too limited to be able to see far enough into the game to evaluate the first move that accurately. Best you can do is estimate it even with AI.

The best we can do ourselves is look at top GM results. At the highest levels it gives white 9% more wins but over 50% draws. Net scoring white to black is 55% to 45%.

For all players , meaning good players in tournaments not beginners and casual club players, e4 gives white under 7% advantage with just under 30% draws. Net is just under 53% for white to 47% for black.

And at the very highest levels white seems to do better than when used by players at lower levels.

What seems to matter more than the opening is the players themselves: How good they are, what type of position they prefer to play, and other intangible factors such as what moves followed e4 as all the other various openings would have their own statistics too.

The paucity of games by high rated players also colors the results. There are lots of games with e4, but only a handful starting with 1d3 even though ALL of those games were won by white. Does that make d3 the best first move? NO! It means that nakamura used it a few times and beat 4 other players with it. One was Carauna! Does that mean that d3 is the best first move? NO! It might mean that 1...Nf6 by carauna was not the best reply. Or it might mean that later on nakamura outplayed carauna and the previous moves were just fine after the position became a KID played by white with an extra tempo.

|improve this answer|||||
  • 1
    How did you get those statistics? In the Mega 2020 database, updated to this week, there are 8,068,186 games, but that includes games from some lower-rated players. The scoring is 1.e4 52.8% for white, 1.d4 is actually higher at 54.3%, 1.Nf3 comes in the highest at 55.4%, and 1.c4 is 54.4%. I am not trying to attack you, but I am genuinely curious. – PhishMaster Jan 6 at 16:16
  • 2
    I pulled them off the largest database I have access to. I have access to others with fewer games that will vary some. I am not surprise that Mega2020 or Chessbase or others that require money to use might differ some. Yes some moves theoretically come in at slightly better than 1e4. But that was shown to be irrelevant by my example showing what the players did long after the first move mattered much more. The first numbers were for games by players over 2700. The second was for 'all' games. – yobamamama Jan 6 at 16:57
  • @PhishMaster For Nf3 it shows white winning outright 38.3 for 2700+ players and winning only 30.4 for all games. For overall score it is 55.9/44.1 and 55.2/44.8 respectively. So add in the factor whether a player needs to play for a win versus just playing as good as possible to also disrupt the statistics. – yobamamama Jan 6 at 17:05
  • +1 this time just for taking some time to check your db and answer on more than one text line – Universal_learner folding home Jan 6 at 17:13
  • 1
    @yobamamama I just searched Mega 2020 for all games with both players above 2700, and with the ECO codes A04-A09, which is 1.Nf3. There are 423 games, with a record of 177 wins for white (41.8%), 90 wins for black (21.3%), and 156 draws (36.9%). That is an overall score of 60.2, which is very impressive. – PhishMaster Jan 6 at 17:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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