AlphaZero taught us that we are playing chess wrong

I think that AlphaZero wasn't doing anything special or mysterious when it defeated Stockfish, it just showed us how wrongly we have recently been playing chess. Now, I know that some of you will notice my low rating and tell me that I shouldn't have an opinion. But I'm going to say this anyway because I think it could be a valid theory. I was watching IM Rensch analyze the AlphaZero game against stockfish and there was one recurring theme that I noticed. AlphaZero didn't seem to care that it was down pieces. It played for the long term. It played positionally and strategically. Stockfish played like modern grandmasters do: tactically and piece heavy.

You see, the last I examined chess theory, everyone was talking about how you need to get advantages, control spaces, and use tactics to achieve your strategic goals. But it looked like AlphaZero didn't follow any of that advice. Instead, Alpha seemed to only care about one thing, controlling the board. It didn't care about material exchanges that much, and it didn't try to use tactics to make Stockfish melt. Instead, it only focused on destroying Stockfish's position and achieving a better long-term board advantage. If that meant fewer pieces it didn't care. So, at least to me, it looked like Alpha beat Stockfish by destroying Stockfish's position, not by winning pieces or using tactics, or making small strategic goals. It only had one goal: destroy the position and force checkmate. 

This got me thinking that, maybe, we humans have been looking at chess totally wrong. We focus on thinking ahead and making little advantages for ourselves and thinking about how we can utilize tactics to make small gains, but none of this can securely contribute to the endgame of checkmating the king. Maybe if we started to think about what our endgame is, and how our current position could give us control and get us closer to checkmate, we will play better chess.

I mean, a Queen sacrifice always looks stupid until you see someone use it to checkmate someone 10-15 moves later. Carlsen seems to understand that you can give up pieces to win games. Maybe we should stop thinking tactically and start thinking about solely destroying the opponent's structure and checkmating their king, regardless of conventional openings, learned board structure setups, or other things that we use to make our thinking shorter.

Thoughts?

Since Google won't grant a rematch, I say Stockfish should play itself, with one side having the hardware and databases given in the AlphaZero match, and the other side given the hardware and databases recommended by the Stockfish team. Then see if the stronger side dominates as strongly as AlphaZero did. And use slower games so the memory difference matters more.

In doing what it did and creating the controversy it has, Alpha Zero has already had a significant impact on chess. It will be interesting to see what lies ahead.

As a newer, lower rated player (and I suspect my rating may even be a little higher than warranted), I cannot differentiate between positional play and tactical play. It seems to me the two are very closely related. I would agree with you that, "[AlphasZero] focused on destroying Stockfish's position and achieving a better long-term board advantage." But isn't this done by tactics, even the tactic of giving up material for a superior position? As I said, I don't pretend to understand any of this to a great degree and I have only read a few beginner books on the topic.

Anyway, there should be plenty of interesting theoretical chess discussions to follow in the coming years. Most of which will probably be way above my level of understanding. 

Alpha zero didn't took material advantage trough tactics because Stockfisch didn't let it, chess engines are wonderful tacticians, and can see tactics that are 20 moves ahead. 

Actually A0 used some nice tactics, trapping SF bishop was a tactic, the kind of tactic that engine can't recognize... Probably SF team is already working to make ot able to recognize those positional traps and tactics. 

I have to disagree with the Queen sack part. When there's compensation, humans can see it: two monstrous bishops and two center pawns is just one possibility.

What you described AlphaZero as doing in these games is called "positional sacrifice" and while AlphaZero's sacrifices are amazing in depth and even mysterious to GMs, the idea is hardly new. The more advanced you are as a player, the more likely you'd make such positional sacrifices, and human GMs make them regularly. A positional sacrifice is not "opposed" to tactics as if you have to choose one over another. Rather, a positional sacrifice is an advanced type of tactics which becomes relevant when more basic tactical opportunities - like winning a piece for nothing - are unavailable. You don't see such basic tactical play in the AlphaZero games not because tactics is unimportant, but because AlphaZero and Stockfish are super-tacticians who avoid tactical blunders that can be exploited immediately. So what you called "tactics" are the basic material concerns that must be examined first, before you consider the more advanced positional sacrifices.

Did you see AlphaZero dropping pieces for nothing or allowing Stockfish to fork or skewer unprotected pieces? Of course not. And that's what is meant by "basic tactics" which AlphaZero always analyses.

Yes, yes it did.  Some learned nothing, some learned everything...  

So you think giving AlphaZero a faster computer with more memory is nothing?

The takeaway I got from this Stockfish v. AlphaZero match was that any formula-based system of analysis, however detailed and refined, is going to have gaps in it, which careful probing will be able to uncover.  Stockfish uses an algorithm with fixed values based upon material and position, but AlphaZero beat it on those nuances - when a knight is better than a bishop (or two), or rarer still when a minor piece is better than a rook - which come from playing a lot of games and experiencing the wins, draws, and losses.  That's how AlphaZero learns - by (self-)playing a ton of games.

I think you're confusing the fact that chess engines can explore deeply into variations without flaws where humans need "signposts" to progress.

So if AlphaZero, which may be much weaker than Stockfish, taught us anything, it would only apply to people with tremendous memories.

If you are human you STILL want to control space, tempo, and development.  I just played a move in a recent game that I learned from Cyrus Lakdawala who said, "The main advantage of this move is its disruptive element."

THAT is what you do when you're a human playing against a human.  The "disruptive element" would NOT apply to playing against a computer.

Additionally, since we humans can't calculate (or effective set-up thousands of electronic boards to play on simultaneously where its illegal for humans to set up more than 1) like computers, we use guides: overprotection, avoid backward, doubled, and isolated pawns,  don't move the Queen out too early, middlegame tactics and strategies (a Knight on the fifth rank is usually stronger than a Bishop, etc.) and endgame concepts like the Rule of the Square or the Principle of the Opposition.  Speaking of the opposition, check out a winning move I made, Qg7!, that Stockfish 8, analyzing at 20 ply, indicates is a "blunder" compared to it's preferred move. But my move maintained the Opposition and guaranteed my Pawn would Queen and win through straightforward human thinking where Stockfish's move was more likely to lead me astray:

Kg7 ! u mean lol

Yes, yes and yes.  Keep looking deeper and you might find something!

Perhaps some will find this little post pertinent!

Was a very unfair match. Too much is being read into it.

A0 taught you nothing,it was rather Google ads in promoting their neural network products. If you run any top ten engines on $25 millions hardware vs $ 3000, you will see similar amazing results from one side.

Would be nice if GM Larry Kaufman wrote a book about this match.