The answer is that Magnus would win. Eight pawns may be almost as many points as a queen, but they won’t have much luck fighting her from their starting squares. White’s lack of pawns is almost a bonus here, because along with the queen it plays into the plan of quickly mobilizing the pieces and mopping up Black’s forces. AlphaZero is much stronger than any human, even Magnus, but it still needs something to work with. In this position if White plays decently Black will never have a chance. The position is so winning for White that it doesn’t take a Magnus Carlsen to beat AlphaZero.
I had some preparation for this particular question based on a class I attended a long, long time ago. I was a little kid in Michigan and the instructor was Ben Finegold, now of internet fame. He set up a position on the demo board with queen vs. eight pawns (the same as the Magnus vs. AlphaZero question, but with no other pieces on the board besides kings) and asked who would win. I figured it would be the queen, but most of the other kids went with the pawns, so I wasn’t feeling too confident. Even less so when Ben invited me up to play the queen with him taking the pawns. I guess he sensed my nervousness because he muttered under his breath, “Don’t worry, you’re going to win.” And sure enough I won without too much trouble. It’s actually a very easy position to win: the queen just swoops around and picks up Black’s pawns. As with Carlsen vs. AlphaZero, the queen is just too powerful.
I was curious how many pawns you have to add before Black is favored. The answer, it turns out, is 12. So have we been wrong all along? Is the queen really worth 12 points? Not really. The issue isn’t that nine is the wrong number of pawns, it’s that expressing the queen’s value in terms of pawns just doesn’t make sense. There’s no number of pawns you can stack up in a trenchcoat to make a queen. Asking how many pawns a queen is worth is like asking how many avocados it takes to make an airplane.
Let’s forget everything we know about chess for a moment and think about what is the value of a pawn, really. Well, a pawn can become a queen, but usually not until the endgame, and as Tarrasch said, “Before the endgame the gods have placed the middlegame.” If the only thing your opponent’s pawn can do is promote, you have a fair bit of leeway. A pawn that is doing little but threatening to queen in a distant endgame is what I’d think of as “just a pawn.” But pawns can also function as sort of mini-pieces, taking up space and controlling squares. In this case a pawn is worth more. To me that’s “not just a pawn.”
In this sense, the starting position of the Stafford Gambit is the poster child for “not just a pawn.” White is up only one pawn, but has a two-to-zero pawn advantage in the center. Additionally, White’s center pawns are linked up into a nice chain. One feature of a pawn chain is it cannot be broken down by any number of pieces without a piece sacrifice. That is to say, Black can pile up as many attackers on e4 as they want, but if they ever capture, White will recapture with the d3 pawn, and at best Black will get two pawns for a piece. What this means is that Black can’t really contest the center in any meaningful way and is reduced to speculative piece play on the wings. With best play this shouldn’t work, but in practice it often does.
This is what got me thinking about the value of pawns in the first place. If White can’t score well against the Stafford Gambit, which has almost as many refutations as YouTube videos explaining them, how much is a pawn worth really?
Further evidence that pawns aren’t all they’re cracked up to be came from BookBuilder, a software tool created by Alex Crompton that automatically generates an opening repertoire based on Lichess game data. The program suggests a lot of pawn sacrifices because they score well in practice. Here’s a representative example.
This is like a Stafford Gambit on steroids. As in the Stafford, White has a two-to-zero advantage in center pawns, but here Black doesn’t even have the doubled c-pawn, it’s just a full two pawns up. In return Black has a lead in development. White’s king and queen lined up on the half-open e-file look like they could get in trouble, but the computer isn’t impressed. It says White can just play Bg5, Nbd2, and 0-0-0 and there’s essentially no compensation for the two pawns.
Soon after I looked at this position an FM rated over 2800 blitz tried it against me. Bad luck for him that I was prepared - this certainly isn’t mainstream theory! Still, even armed with the correct plan, I wasn’t able to keep control of the position. I ended up winning a messy game, but he had some chances. Overall, at least in blitz, I feel better development or piece mobility can be worth more than pawns, even if the computer doesn’t agree. Not many blitz games are decided by one player maintaining a single pawn advantage throughout the whole game and promoting it in the endgame. Usually, one side or the other gets hit with a tactical haymaker that ends the game. The more active your pieces are, the more likely you are to be delivering rather than receiving the haymaker.
Additionally, the idea that you could sacrifice a pawn or even multiple pawns for piece activity isn’t inherently unsound, even if the gambits we’ve looked at so far don’t really hold water. Here’s a position where AlphaZero sacrificed not one, not two, but three pawns to defeat Stockfish in a celebrated game.
All of which makes me wonder if we shouldn’t be trotting out gambits more often. It gets at a fundamental question of chess improvement: Should you try to do your best impression of grandmaster play from the moment you pick up a pawn, or should you embrace the fact that different things work at different levels? Currently, the mini-GM approach seems to hold sway, based on the theory that you shouldn’t learn bad habits you’ll only have to unlearn later. That makes sense, but even with the most principled approach, improvement is all about learning and unlearning. There’s simply no way you can understand chess at a grandmaster level when you’re just starting out, regardless of what you do.
There’s an idea in biology that "ontogeny recapitulates phylogeny" meaning that the development of a single organism retraces the evolution of the species as a whole. As a biological theory this has been largely discredited, but it could work in chess. That is, your personal growth as a chess player might reflect the development of chess theory historically. Before the development of positional chess, there was a romantic era where everyone tried to sacrifice all their pieces every game. You could do a lot worse than having your own romantic era.
This was an interesting read, thank you. I love the Ben Finegold anecdote.
I think about this often, and I've also heard the advice to play more gambits and open games when starting out. I think it does help ingrain the importance of initiative and piece activity. In addition, those positions are often more fun to play than trying to fend off an attack and trade pieces to get to an endgame up a pawn.
I have a hard time with it personally. I rarely play gambits and I almost always accept them. If I think they shouldn't work, I'm determined to prove it (of course they often do, esp in shorter time controls, but I try to view it as a learning experience. I no longer fear the Stafford Gambit, but only after many losses against it).
I had the privilege of experiencing grandmaster play last year when I was paired against Brandon Jacobson round 1 of the US Open. I played a Rossolimo as White and the game felt very balanced. He slowly built up a better position and undermined mine, eventually launching a kingside pawn storm that slowly suffocated me, but material was equal the whole game, if I remeber right. Of course, I'm certain he was in complete control of the game the entire game, but I felt good, and I went into subsequent rounds with my head held high, only to lose pieces in opening traps to opponents rated 1100-1300.
At the very least I don't underestimate such gambits and dubious attacks anymore, and I often decline them in blitz. In a recent OTB game I played the Scotch Potter Variation and my opponent played a sac I had never seen: 5...Bxf2+. I thought "that shouldn't work" but spent the entire game getting to a slightly better endgame that was just enough to win.
I'm reminded of your "there are often many decent moves in a position" and I have really come to appreciate that a move that looks dubious (or that the engine screams is a blunder) doesn't just give away the game. At least the opponent has a clear plan and is all in on it. The burden is then on you to find the refutation, and this can be a real challenge, especially for those of us who tend to get in time trouble.
(Sorry for the long comment, but I'm really enjoying the blog: it gives me a lot to think about. Keep up the good work!)
I agree with all this but I think something else is also going on with the score of these gambits at low levels. Someone who is rated the same as me who always plays objectively bad openings like the Stafford is probably stronger than me tactically to begin with - else they’d lose more than half their games and be rated lower. Then you add development and activity compensation in a dynamic position and in practice they win about half the time - as expected.