Is King's Indian Defense now obsolete?

The DeepMind paper is well worth reading.  Not just for the impact to Chess, but it has some interesting insights into learning techniques (their point was not to make a powerful Chess engine, but use that to continue developing their AI).  When you let it choose its own path, it naturally migrated towards the things that were easier to understand.  When you force it to play more complicated things, it took it longer to understand them, but eventually played them well, too.  An interesting insight into human learning: if we are left to our own devices, we will naturally avoid things that we find too difficult, when it is precisely those things we need to explore in order to continue growing.

I think it causes growth. My reasoning is really that there will be those who will never excel. They may become ok but without aptitude, they won't excel. Memorising moves is part of the process of improving. You can go at it from both sides. Try to work out moves on first principles and also memorise them. If a good player memorises them, then they get further into the game before they're thrown on their own resources. That's a big advantage because it means more thinking time for moves they have to work out. On average they will win more games, their ratings will improve and they'll be playing better players in league chess and tournaments, from whom they're more likely to learn.

Therefore NOT memorising moves automatically hands yourself a disadvantage and slows your improvement.

When I was a tournament player, after I'd taken up chess and when I was improving, I also was doing a degree and acting as primary care giver for our son, while his mother worked full time. Therefore I didn't have much time for studying chess and had to use the time well. In a tournament, if I lost a game and could trace it to a sequence of moves in an opening I didn't fully understand, I would try to memorise that opening two more ply further. Well, really one more move for me and then my opponent's moves and my preferential move for each of my opponent's moves. Looking back, that's how I moved forward to being a strong player. It turned out to be just enough to gain an advantage at the level I was playing, which would have been around 1700 FIDE. My rating shot up and I was 1800 FIDE (149 BCF or ECF) in no time, and since I was in the North of England that was 160 BCF or 1880 FIDE and I was playing fairly strong players and using that method I devised, also improving. Soon after, Northern Counties players were given an extra 10 BCF to stop us winning the counties championships all the time. That actually worked and it stopped us (Lancashire) winning the county championships at all levels except Open.

Not only do they not understand openings, the post-game computer analyses on this site tell me should be grabbing a free pawn instead of pressing on with an attack on my opponent's king. Computers don't think.

When I was playing chess960, it constantly wanted me to play h4 or a4 because it could see it was grabbing space. And as white in a KID it wants me to play h4 all the time. Not a clue.

Hope you are keeping well. You've been missed.

"Who's been missed, I'm a new account?"

Oh sorry, of course you are. A pleasure to meet you.

I think what he was getting at is the idea of just memorizing moves without understanding why you must play those moves and why your opponent must play certain moves will get you into trouble.

I took a look at your profile after I posted that and saw. Hope you had a decent Xmas and New year.

Perhaps, but credit should be given where credit is due.

AlphaZero stomped onto the chess scene using self-taught neural network technology. At that point, the developers of Stockfish and Komodo were still using laborious, human-written code.

AlphaZero dominated Stockfish in a way that we all are now familiar with. Since then, the DeepMind team has moved on to other things. (Creating a strong chess program was never their end goal. They simply used chess as a testing ground to evaluate their self-learning alogrithm. Their end goal is a general AI that can help solve bigger human problems, like disease, poverty, etc ... )

Now Stockfish, Komodo (and Leela) all use their own replicated versions of AlphaZero's NNUE technology, which nods to the significance of AlphaZero's quick arrival and departure.

Yes but I found that memorising without understanding helped me to become quite a strong player. People didn't talk so much about who is strong but I know I was considered very strong on my day.

Memorising without understanding helps the understanding to come, or it should do.

• In my opinion, In high level and tournaments, I still believe King’s Indian with 1.d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 is still so powerful and not obsolete especially with:
  • • Averbakh Variation with 5.Be2
    • Kramer system 5.Nge2
    • 5.Bd3
• In medium level it depends on your opponent and how he prepared for counters of course :
  • Sämisch Variation with  5.f3,
  • Four Pawns Attack with 5.f4 
  • Makogonov system with 5.h3

155 wins for AlphaZero.

6 wins for Stockfish.

At G180+15 time controls.

It's quite impressive, when you consider the fact that AlphaZero only took 9 hours of self-learning to reach it's level, after only being taught how the pieces move. It discovered how to play chess on its own.

Stockfish, meanwhile, had been tweaked and fine-tuned by human chess masters and programmers for 8 years. And it still came up short.

All you have to do is look at what happened after the AlphaZero vs. Stockfish match: developers scrambled to add neural networks to their engines.

Stockfish did it (they even made an excited blog announcement about adding NNUE to their engine, in 2020).

Komodo jumped on the NN bandwagon, as well.

Most top engines now use neural networks ... because their developers recognized the power of what AlphaZero had done.

(For anyone interested, there are many fascinating insights about AlphaZero's development, its chess approach, and its impact in the book Game Changer, by Matthew Sadler and Natasha Regan).

Regarding the thread itself, though ... I agree with those above me who say that the KID is not obsolete, by any means. With best play, I'd wager a bet that not even Stockfish (or AlphaZero, for the matter) should be able to beat it.

It should hold for a draw.

Engines are brutal. Yes, its still obsolete.

That was right a few years ago. Now it is a bit different: they do understand the positional plans of the closed pawn structures, and do find the best moves, but still they are spaceaholics and grant White a large advantage (something ilke ~ +1.0) in many lines, while in reality he doesn't have much of anything.

Since it achieved some sort of AI by adding a neural network in parallel to his traditional Alpha-beta heuristic. Now their percentage of good moves rejected due to pruning has decreased a lot (to the point that, competitively speaking, ICFF games have become a boring drawfest), and it is quite likely that in a few years 99% of the games there will be draws. Not all, as there are still factors like careless players, programming bugs, opponents getting sick, or dying, and such...

To sum it up, due to AI/NNUE, there were no openings which were refuted- of course you can exclude certain variations. Quite the contrary: Many openings which were considered bad have been fully revived. What engines cannot do is evaluate the human playability factor of variation ABC: Engines have absolutely no problems getting through a tactical jungle, where a wrong step means defeat. For a human, it would be rather stupid going there and expecting to calculate anything like 1% of what the engine can achieve.

The question of whether an opening is sound at the highest (engine) level is an interesting one.

From a practical point of view, the KID is great, but players in our rating range should worry more about playing these openings well if we chose to play these more sophisticated ones. I've seen too many play the first 5-6 moves automatically and then continue in a rather random way, without regard to the ideas.

the ways to counter KID are far better understood now than in its hayday. Now the amount of lines black must know by heart to get a playable game at the super GM level is too high for most to bother.

and that's really the rhythm of a lot of rarely played stuff,  its rarely one big refutation but a set of practical nagging problems that require too much effort to get away with. 

think of it this way , at the highest level the pool of players you will encounter is mostly closed off. You start accumulating folders for your top competitors repertoire. If you start playing KID with too much regularity, you will have other 2700's looking for novelties left and right to knock you off your game. its not a defense like a ragozin, or a berlin or a QGD where you know a novelty may prove to have practical venom but is likely to be fine, a novelty in the KID can possibly knock out an entire sideline out of commission at any moment.  Its exhausting to play stuff like that with any regularity.

'King's Indian Defense is riskier for black than King's Gambit for white' - Bronstein

I was a Modern Benoni player but am reverting to the QGD because of the above. I like the problems the MB sets but it isn't good for practical use except in fast chess. Exactly the same as this aspect of the KID.