# Can a grandmaster still win against engines if they have a really long consideration time?

Can a grandmaster still win against engines if they have a really long consideration time? For example, say that Stockfish gets 20 minutes and the GM receives days, if not weeks.

• From a developer point of view, this question has one flaw: 20 minutes on a phone is a lot less than 20 minutes on a supercomputer. As such, "engines" is a bit ambiguous. Real simply said, an engine given 2 cores will be twice as good (or at least be able to calculate twice as much) as the same engine given one core. (Over-simplification of course.) Dec 27, 2020 at 13:05
• Is the engine allowed to "ponder" (think during the human player's turn)? Dec 28, 2020 at 20:37

Are there asymmetric time controls where a GM can still beat Stockfish? Certainly, as long as you give Stockfish little enough time. I think the time you give to Stockfish is almost more important than the time that the GM gets. At 20ms I would favor the GM even at tournament time controls.

A few years back I played a handicap game against Komodo where I played some blitz time control (maybe 5+0 or so, I don't quite remember) and Komodo had 1 second, which should work out to roughly 20 ms per move and I think back then I could actually win that game. Since then engines have progressed so modern Stockfish is probably 300 rating points stronger than the Komodo from back then, but on the other hand I am more than 300 rating points below a GM so I would think the GM should have chances even if they blitz, nevermind at tournament time controls or longer. As it turns out engines aren't really made for 20 ms per move games so they don't perform very well there. Now if you give the engine a second per move things may be very different.

As a fun experiment you can just try yourself, give yourself sufficiently much time against Stockfish and give it a go, see how it goes.

• About 2 and a half years ago, it was possible for GM to beat leela chess at symmetric time control: lichess.org/study/VlIegy6E/XfqgyZCG Dec 27, 2020 at 4:48
• @Akavall It's not a good example though since Leela is a relatively new engine. It's gained more than a thousand elo in self-play since then and is now unbeatable by humans, as well. Dec 27, 2020 at 8:57
• @Akavall Indeed, at the time it was by far not the strongest engine in the world. Of course, you can always still beat weak engines, for instance I can on a good day beat the engine I wrote. Because the engine I wrote is terrible. :D Dec 27, 2020 at 9:40
• Just to somewhat support this claim more broadly: humans (and all animals) are innately good at quickly making an approximation of a situation. Throwing a ball to hit a moving target, pathfinding logic, image recognition, or even the traveling salesman problem. Humans can't find a perfect solution or be very precise quickly (or even slowly, sometimes), but they are faster than a computer at finding a quick approximate solution. Therefore, giving both the human and AI player less thinking time favors the human player. Giving the AI less time and not the human favors the human even more. Dec 27, 2020 at 10:48
• @Flater sorry but that part is clearly not true, giving both the human and the engine less time favors the engine. Just look back when man vs machine was competitive still, engines were able to beat top humans in blitz long before they were able to do the same in long time control games. Presumably the reason being that calculating 100K nodes or 1M nodes is not a huge difference, however for a human, the difference between calculating 1 move or 10 moves is a huge difference. (of course things are not that simple, but the point stands) Dec 27, 2020 at 12:02

Here's a comment by GM Kaufman, developer of Komodo (emphasis mine).

Q: Author Cyrus Lakdawala suggested I ask: In what respect are the program's move choices human?

A: All the features of the engines' evaluation function have been based on how some human (in the case of Komodo, me) thinks they should be defined. The weights were originally my subjective ones, but gradually got "tuned" by testing and automated methods. So in theory, if the search depth were the same as mine (which of course is variable so this is unrealistic) it should play somewhat like I do, although I don't add up hundreds of numbers in my head when I play, I just estimate everything. The main reason Komodo is a thousand or more Elo stronger than I am is that it searches so much further ahead in nearly every line than I could possibly do. Perhaps if I spent 24 hours per move on a game (moving pieces around freely, but not consulting any engine) I could play as well as Komodo plays a blitz game, but that's just a guess. Note that this does not apply to NN engines, which (in pure form) don't have any human knowledge input.

So: maybe.

The computer's big advantage is its speed. It can look at millions of possibilities in seconds. But the human Grandmaster would have the advantage of being more selective and knowing just what variations to concentrate on. Based upon that, I would think that given unlimited time, a Grandmaster would be able to defeat the machine under the conditions you stipulate, its speed notwithstanding.

The short answer is yes, but there's more to it. For one thing the usefulness of the consideration time stockfish gets is dependent of the computer it's run on. On my phone stockfish takes some time to compute 100m nodes (each node being a potential position or board), on my pc it takes a few seconds to generate that amount of nodes. Also 20 minutes is almost certainly still going to be more consideration time than stockfish 15 needs to beat any human even just being run on a phone. Stockfish would probably have to be limited to a very small amount of time or a relatively small number of nodes before it becomes beatable by any human, but given a small enough limit it will become beatable.