Counter intuitive move

I am quite puzzled as to why the computer rates a counter-intuitive move as being the best move in the position below.

Currently the position is at move 24 (a side-line to the actual game, not the main line played) with Black to move. The computer program Stockfish recommends 24...Reb8 to be the best move with an evaluation of -2.97 at depth 16, followed by 24...Qe2 at -2.74 and 24...h6 at -2.68. The move 24...Reb8 is simply anti-developing and counter-intuitive as the active rook has been brought into an inactive position, also causing the a8 rook to be unable to move. I would definitely not play or even think of 24...Reb8 in this position. Does anyone know why the computer would suggest such an inactive move to be the best move, and how many of you would play 24...Reb8 here?

On a side note, the actual game went like this:

Engines can calculate the variations very far, far beyond the human ability. I think the idea was that the rook pair at a8-b8 allows the a and b pawns to move towards promotion without any obstacles, while protected by the rooks. Obviously this isn't something a human would ever play, but it is quite possibly the best move in this position, however awkward it looks.

Agree with MayCaesar. The 2 vs 1 pawns on the queenside are how black will eventually win the game so Reb8 prepares their advance. It looks scary for black over on the kingside but the fish calculates like a beast so can see concretely that anything white tries can be defended one way or another. Therefore it may as well get on with the business of pushing the pawns forward to victory.

@MayCaesar and @Strangemover

Thank you for your suggestions. Well that does make sense, and perhaps players at more experienced levels might play it, but the formation of White's pieces in that situation does look scary as you mentioned, since all 4 of white's pieces and that e5 pawn seems to target on g7 and e6 and yet the e8 rook moves away from the defence of e6. I guess it takes some courage to convince myself that White's scary looking pieces are not as scary as they appear to be and even more courage to move the e8 rook to b8.

I think this is one of those cases in which I wouldn't follow the Stockfish advice. The move Reb8 probably relies on precisely calculated lines, and a human not able to calculate as well may get in trouble after this. The move 24...Qe2 pretty much forces the queen exchange (otherwise 25...Rf8 follows, and, after a series of exchanges, white is down a piece without any compensation), and white's position collapses rapidly after that. Stockfish knows that, if both players play perfectly, then Reb8 leads to the best position for black (at least, by Stockfish evaluation scheme), but humans probably shouldn't play a move like this, as in practice it might be harder to find the right sequence of moves after it.

@MayCaesar

Computers do compute lines which sometimes appear weird to humans. Computers do not possess any feelings at all, and will play any move to their liking. Humans, on the other hand, are wary of scary-looking threats and usually attend to it even if the threats are not actually that deadly.

In this game https://www.chess.com/forum/view/more-puzzles/quest-for-survival-in-a-winning-game I encountered a line where the computer likes my idea, but in the actual game I nearly wilted as I almost resigned after not being able to find a rather difficult saving move (I do not have super powers like Stockfish). I had to defend move by move just to endure a counterplay from my opponent. Had I gone back in time to play this game again, I would not have played what I have played and would have chosen a safer winning (but not as winning) alternative sequence.

@wfloh,

That is definitely true. If a pawn can be protected without a compromise in safety, why not keep the pawn to your advantage? Stockfish feels that the b7 pawn can be protected safely because White has no funny business around Black's kingside. But it is going to be painful for me to endure an attack from the opponent, even if it does not actually yield anything, to win the game just because I tried to save the pawn. Because from most human players' point of view, it would be wiser to play the safer line and still hold that advantage. Well sometimes I hate an opponent's knights for that reason (knights love to harass queens and rooks around the centre squares).

In the actual game the situation did not occur. I played poorly, turning a win into a draw and then a loss.

This is an interesting aspect showing the fundamental difference between how humans play, and how engines play.  For a human, being a knight up, losing one pawn does not matter much, if it leads, for example, to equal exchanges, rendering the endgame easy to win. But an engine tries to always preserve or increase its current score, and since losing a pawn effectively drops the score by 1, even if the resulting position is completely won, it still sees it as worse than the current one.

That said, even if black decides to save the pawn, it is not immediately obvious that ...Reb8 is the best way to do it. Personally, I would strongly consider ...b5 or ...b6, for example

@eric0022 

I remember seeing Topalov's game in which, as black, having a rook on a8 and the pawn structure a6-b7, in the middle game he suddenly played ...Ra7. This move looked extremely ugly, and it was very hard to even see the point of that move, let alone why it is a good move. And engines strongly disagreed with this move, suggesting something much more intuitive. Yet Topalov won that game. I suppose playing ...Ra7 was more to his style, which is also an important factor to consider: you don't necessarily always want to strive for making the "best" move, often making the move that matches your style better, leading to the positions you understand better, is practically better than the raw engine suggestion!

Well that is why most players stick to basic chess principles, since it is easier to understand than playing some winning but weird moves. It is much easier to play positions that we are more familiar with (example controlling the centre, doubling rooks etc) than to try random sacrifices unless we are sure that these sacrifices will work to our advantage.

I doubt I will ever play Ra7 when there are black pawns on a6 and b7. Probably he intended to push the b pawn in the moves following so that the rook could move along the 7th rank.

If I remember correctly, it was something much more subtle. I think, the idea was that white bishop on the a8-h1 diagonal had to be opposed by the black bishop, and to move on b7, it needed support by the rook - but ...Rb8 move hindered the knight maneuver arising in one of the variations, in which it had to go through b8, so developing the rook on a7 was the better option, and, since the position was very closed and rooks wouldn't get into play in a while in any case, that move was justified.

The problem is, even if we've evaluated and calculated everything and know that this is the best move from our perspective, it is still very hard to physically make it, because of how awkward it feels. Many players, even the strongest players, are comfortable playing much more intuitive positions, and they wouldn't make a move like that, just because it didn't lead to the position that "felt right" to them. I think it is an important factor, and we shouldn't dismiss it in our play; machines can afford making the raw best calculated move, but humans also have to account for personal style. Sometimes a slightly inferior move can lead to a position the player understands better and can play better, so the otherall quality of his game will be higher, even if that particular move isn't the best possible.

In this particular position, even knowing that 24...Reb8 is the best move, I probably wouldn't make it, because I am not sure I will be able to continue from that the right way. My move would be 24...Qe2, looking to simplify the position, even at the cost of 1-2 pawns. But someone more comfortable with this rook setup might very well follow the engine's advice.

---

It is also worth noting that engines' assessment of each position depends on the set of criteria, which are somewhat subjective. Engines value material advantage, open lines, piece activity, pawn structure and so on - in some of the older engines, those criteria were often publicly available and you could look at the algorithm they used. That algorithm, while highly effective, is still somewhat flawed, so the engine will not always recommend the perfect move. It is rare, if impossible, for a strong engine to suggest a wrong move, but when it comes to picking one out of a set of top candidate moves, I think one should take the engine's recommendation with a certain level of skepticism.

Well that is a very important point to consider. It is easier, or more comfortable, to play a familiar position with a clear follow up rather than embark into some unfamiliar terrain where the situation does not allow any room for errors - the sequence of moves is very advantageous to the winning side but it must be played very precisely, and even the slightest of mistakes from the winning side may quickly turn the situation into a draw or even a loss.

was there any way for black to continue without giving up his queen?

in this situation I would choose to move Rf8 and swap rooks and work from there. that way there is less material on the board and more room to work with

@Benedore

Probably ideas like those are easier and more confortable for human players.

As long as we can refute White's funny but unsound attempts to try and push for a win (for example by Nf6), any reasonable series of moves should work.

In the actual game I blundered several moves before the game reached move 23. This position felt extremely difficult from Black's perspective as all three of White's heavy pieces (queen and rooks) were extremely close to Black's king, indicating the possibility of a checkmate or at least some form of damaging attack. On top of that Black's queenside knight and rook were not developed yet. I found it quite difficult to continue the game by moving the queen (and now I highly doubt that there is any saving move to White's potential attack if I chose to save the queen), so I made the difficult decision to let my queen go and chose to play 23...Nc6 instead, because capturing the queen by 24. Rxe5 would result in 24...Nxe5 with a knight fork. The game went ahead with White indeed capturing on the Black queen and the knight fork on White's queen and rook.

After the necessary queen move followed by 25...Nxf7 and 26. Qxf7, preliminary piece counting suggested that I have secured my opponent's two rooks at the expense of my queen and knight. However, closer inspection a few moves down the road revealed that White's knight could not stay on the board much longer due to a back-rank mate threat, as it happened in the actual game. The resulting endgame of queen and 4 pawns wersus two rooks and 4 pawns should be a draw in theory, but of course poor play from me and finally a blunder in the endgame dropped my rook to a strong attack and the game as well.

Another situation where Rb8 was preferred over Ra8, after I have done a "post-mortem" of my game.