I've been getting Stockfish to analyse lines of the KID, and the evaluation gives a score of around +1.5: White has a big advantage. How come this is true, when the KID is a playable(and good) opening? I've also seen the same problem with Stockfish analysing the Neo-grunfeld, and similar hypermodern openings where White gets a large center.
Very related: Traxler Counterattack
The answer is effectively the same: White has roughly a +0.5 advantage in the starting position, and a shift of a few centipawns one way or another is acceptable, but if that advantage has grown to +1.5 there's a problem. Do note that the starting position of the KID, 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 is evaluated at about +0.5. If you are getting +1.5 book exit, you've gone into some dubious territory. It doesn't matter what human analysis thought of the line before the advent of computers - if you go into these lines today, and your opponents are armed with engines, you are asking for trouble.
That said, it's worth pointing out that the King's Indian is a difficult, unbalanced opening that even computers don't understand perfectly. That's why it is disproportionately represented in the opening book for engine tournaments - it gives great chances for one engine to outplay the other. But: even these engine tournaments do not usually start with book exit lines of +1.5. That's very likely going to lead to a 1-0, 1-0 game pair.
For illustration, here are some Stockfish evaluations of known bad lines:
- e4 e5 2. Nf3 Nc6 3. Bc4 Nf6 4. Ng5 d5 5. exd5 Nxd5?! (see TCEC Season 15 games 43/44): +0.8
- d4 d5 2. c4 c6 3. Nf3 e6 4. Nc3 dxc4 5. Bg5 f6?! 6. Bd2 b5 (see TCEC Season 18 superfinal game 8): +0.74
- e4 d5 2. exd5 Qxd5 3. Nc3 Qe6+? (Patzer variation): +1.1
- Pawn odds (White starts without f2-pawn): -1.5
- e4 e5 2. Nf3 f5? (Latvian): +1.7
- e4 e5 2. Nf3 f6? (Damiano Defense): +1.9
- e4 e5 2. Ke2? (Bongcloud): -2.1
As you can see, you have to play some pretty terrible lines to get > +1.5 book exit.
In short, if you are Black, interpret the book exit evals as:
- eval < 0. Obviously the best, but will never happen without cooperation from White.
- 0 < eval < 0.5. The next-best, this is what you will usually be shooting for.
- 0.5 < eval < 1. Most commonly seen in engine tournaments where the aim is to see if the engine can defend the inferior position. Not recommended for actual tournament play.
- 1 < eval < 1.5. These lines come close to being busted. You could conceivably play them if you are after an adventurous game, or if it leads to complicated positions where you could hope to outplay inferior opposition, but it's a bad idea to take them up as a mainstay of your repertoire. A computer can still conceivably defend Black's position, but humans are going to suffer.
- eval > 1.5. If you ever include these lines in the opening book of an engine tournament, be prepared for viewers to start talking about how you've chosen these lines to make the engine with Black lose.
Edit: since some people were unconvinced about the role book exit plays in opening selection, see this article by Jeroen Noomen (who selects the openings used in the Top Chess Engine Championship, the unofficial world computer chess championship):
The complexity of a position and the book exit evaluation are the most important selection criterions. Book exits have been checked with classical Stockfish (development version, contempt = 0) on a 16 core Ryzen system and Leela Chess Zero v0.26.0, with the net that played in the TCEC 18 superfinal. As mentioned above, book exits are significantly higher compared to season 16 and before, to keep the games interesting and avoid boring draws. Nevertheless, there are also some lines with a lower book exit, especially in the theoretically interesting lines.
Low book exits will lead to boring draws. High book exits can lead to wins, which of course bodes badly for the engine that has the worse position out of the opening. These openings might still be viable in human play, but the keyword here is "human play"; objectively they are subpar and against an opponent armed with engines you are asking for trouble.
KID is a notorious difficult opening for a chess engine to evaluate. Forget all those opening lines, just look at the standard positions based on standard positional terms.
- White has an advanced pawn on d5 in many lines. This gives space advantage.
- Even if White doesn't a pawn on d5, White's centre control is apparently stronger. Open d-file for the queen, more squares for the bishops etc etc.
- Better queenside spacing and pawn structure.
- (Important) Black often has to advances the pawns in-front of its king on the king side. But ... this is making chess engines worried because chess engines are taught not to push the king protecting pawns without a clear compensation.
KID is too complicated for a chess engine to evaluate properly. You need to push it into specific lines.
You will see the same problem in an opening setup where Black's idea is to give up the center control but opt for long term counter-play. Grunfield, KID, and hypermodern openings all fall into the category.
Note that Hikaru Nakamura and Garry Kasparov have both stated that engine evaluations in the KID that favor White shouldn't be fully trusted. In the 2015 Sinquefield Cup, Nakamura won a sharp attacking game against Wesley So after which Naka is quoted as saying:
"Wesley, frankly, just needs to learn how to prepare better," Nakamura said. "He just really misunderstood the position." Nakamura said the computer evaluation can't be trusted…
Source: Mike Klein, "Nakamura Blasts So's Preparation, Vachier-Lagrave Also Lurking," Chess.com, August 30, 2015.
In a tweet, Kasparov said:
Congrats to @GMHikaru on his victory today. The King's Indian requires the courage of your convictions & courage to ignore machine opinions!