Computers' best opening

Does anyone have information on what the strongest computer programs, without being fed human opening theory, think the best opening moves are?

That is interesting, though it seems that today's best engines would be able to at least see any opening move through to quiescence. If what you say is true, it is encouraging, as it suggests computers are far from "solving" chess. Seeing Komodo's results would be very interesting. Is it possible to let it "think" without any human input? I feel it would choose 1.e4, but I am interested to know what it's other options might be.

Thank you, Firebrand! I'm pleased you find this question interesting. I wonder if someday down the road, when computers are up to the task you describe, some disgraced opening might prove to have been the answer all along! You also have me wondering about the implications for a human player today. Does the computer's acknowledgement of its own limitations suggest that we are better off playing simpler openings? I left the Giuoco Piano behind forty years ago. Maybe it's time to go back!

There are a lot of misconceptions about engines it seems.

(Not directed at you FbX) for example "my engine reached depth 30, so this is the correct move!"

Sure, it reached depth 30... by ignoring 99% of all possible moves lol. Engines can't possibly look at every variation. And in particular they're not designed to play openings well.

It may also encourage you to know (some) 100 year old human analysis is still good. Engines don't play or evaluate perfectly.

If the answer escapes human players for 100 years, it's likely to be so complex that it's impractial to play in any case.

Maybe a good example is B+B vs N pawnless endgame. Humans long considered it to be a draw. Only after engines solved it did we discover it's a win... but that doesn't mean you memorize 12 moves and are done with it. It's very difficult to win even after you study it.

Professionals work with engines all the time to explore openings. But if an engine, by itself, found a disgraced opening is playable, it's likely to be like the endgame I described, too bizarre to be playable in a practical sense. (Well, the 5 piece endgame is not unwinnable, but an opening will be exponentially harder!)

This has opened up a can of worms I wasn't expecting. While computers aren't perfect, they are better than humans. I am surprised, if it is true, that no engine has tackled this problem in earnest. While the compounded history of human chess thinking has produced amazing results through the collective experimentation of millions of players, we are still guided by concepts (e.g. control of the center, rapid development, king safety) that are by-products of chess thinkers 400 years ago, who surely could not see as deeply as computers. The evolution of chess theory, including Hypermodernism, has not strayed from these tenets, as to us humans, they make perfect sense. Thus, all grandmaster games overwhelmingly follow rather narrow patterns and very little investigation has been done on less familiar paths. When I solved the Rubik's cube, the essential realization was that I had to make it worse before I could make it better. So much has been discarded because in the short run, things look bad. Perhaps these early setbacks lead to superior positions later, if one knows how to navigate them. But as they are abandoned once the position seems unfavorable, we will never know. I'm not speaking of moves with a clear and decisive refutation but of the nebulous moves that databases tell us are favorable but lacking in statistically signifigant samples. What a fascinating experiment it would be to ban all "sound" openings from grandmaster play for twenty years, and see what theory emerged.

I have bad news for you  engines are programmed by humans, who tell them to favor things like the center and king safety! So even with computers you're not free thinking yet (so to speak).

Also, the underlying logic is hard (impossible) to overturn in most positions. The center is the closest to all other parts of the board. The king's mobility is directly linked to the winning/losing condition.

That said, it is exciting to think about the exceptional cases that must exist, where bizarre or counter intuitive moves are actually the best. To some extent this happens already. Humans are used to ignoring "engine" moves in favor of more practical approaches. Even someone like me, who is far from a professional, has the habit of setting the engine to show me its top 3 choices... and I only explore moves or lines that make sense. (It's not possible for me or my opponents to play based on huge calculations.)

Yes, I noticed the good word choice  I also know you're an accomplished ICCF player.

But when you scrape the bottom of the barrel (like chessbomb chat hehe) you see people who say strange things ;)

It is always amusing to see how quickly these threads stray and become combatative. To return to the OP, and perhaps phrase it better, is anyone aware of research being done with computers aimed directly at the opening, with no human heuristics added? For those who find this topic worthy of exploration, I suggest an, admittedly, imperfect book called Play Unconventional Chess and Win by Noam A. Mandella and Zeev Zohar. It looks at the seemeingly bizarre and counter-intuitive moves computers play, and what we might learn from them. It is not the Holy Grail, but it is the only book I know of that addresses these matters, as they apply to praxis.

The problem is that the opening can ultimately only be evaluated based on the edgames that result from them and then have to traverse a multitude of middlegame positions, with all the complexity those positions have. 

There are a number of openings where computers think one side has an advantage but as the position gets father along, that begins to dissipate, often to no better than a slight advantage. 

But again, the opening can't be looked at in isolation. Barring some literal quantumn leap in computing, chess won't be solved for a very long time, if ever. That said,  there is likely no best opening and there will be a number of them, maybe most standard openings and possibly many unorthodox ones,  that will lead to equality with best play.

See Post #11 at the following link, which might have some useful links for you:

https://www.chess.com/forum/view/chess-openings/computers-know-about-openings

For me, a far more interesting question is which openings do computers still not understand. That's an ever-shrinking list, but dynamic openings such as the King's Indian are still beyond their ken.

Interesting, I wouldn't have guessed an open Catalan.

Why is it important to add salt during cooking? Apart from the fact that Bc5 gets never even mentioned by most engines.

Reality is merely an illusion,albeit a very persistent one! 2...Nf6 followed by Bb4. I'll take a closer look.

You can always buy yourself a coach, and hundreds of tournaments. Pawn takes pawn on f4.

You were right some years ago, but now there are some computer programs (starting with Google's AlphaZero program) whose only human input is chess rules and then the machine "learn" (between quotes because it is completely different to how a human learn) all they need to play good by themselves. So the programmers do not "program" the value of the pieces, openings, center control, strategies, etc.. so it make sense to check which openings does those types of machines favor (1.e4 vs 1.d4 for example)

For those of you wondering what Stockfish 13 NNUE has decided (Depth 50+), it is quite partial to the English (1. c4). It can't seem to find consistent advantage in the Kings Pawn Opening against the Sicilian, or in the Queen's Pawn Opening Against the Three Knights Variation of the Queens Gambit Declined. 

It's second favorite is the Zukertort Opening (1. Nf3) but mostly because the lines it chooses tend to transpose back into an English.

This may be a bit of a deceptive result and as a programmer I am confident that we are quite a ways away from being certain as to what the best opening move is.

My personal recommendation to new players is to find ONE opening, preferably an aggressive but solid one, and learn it really well, with general knowledge on correct responses to a variety of white openings. New players can often get caught up in long lines of the Najdorf or the Dragon in an attempt to eventually learn how to play 'perfectly'. This simply is not the best way to get better, short term or long term.