Bots at BGA

[Meeplelowda]

Why would a bot wait until the last second to make its move?

Well, there are two reasons I could think of.
(1) It is programmed to troll people. (I think trolling is never something you should rule out in any context.)
(2) What Jellyby said. It needs processing time depending on its own complexity and the machine it is run on. (Although it is indeed weird if it takes several minutes per turn.) You don't know what kind of bots people create or what is the mechanism for the bot to recognize the game state. I mean, yeah, a well-made bot would not take minutes per turn but some people may take BGA as a playground to "fool around" with some crappy AI bots run on a mediocre system, who knows.

I'd certainly suspect trolling first. I would have thought that taking more than a few seconds to make a move would be superfluous, given how fast machines operate, but I guess it depends on several factors.

It doesn't matter how fast a machine is if you write inefficient enough code, as non-professional programmers are prone to do (no shade, I count myself among them). There are literally decades of research into efficient algorithms for searching trees, for example, but if you are just vibe coding with not much CS background and don't even know that body of work exists, you won't even know the right questions to prompt your AI with to make it more efficient.

[teallite]

I'd certainly suspect trolling first. I would have thought that taking more than a few seconds to make a move would be superfluous, given how fast machines operate, but I guess it depends on several factors.

No matter how fast machine operates you can't solve chess and many other games. If a goal of a bot is only to win, it should use all the allotted time. The only difference is a better bot will calculate a better move.

[tbhp]

I'd certainly suspect trolling first. I would have thought that taking more than a few seconds to make a move would be superfluous, given how fast machines operate, but I guess it depends on several factors.

No matter how fast machine operates you can't solve chess and many other games. If a goal of a bot is only to win, it should use all the allotted time. The only difference is a better bot will calculate a better move.

That makes sense, though it is a case of diminishing returns. I mean, would a bot that takes minutes to make a move be significantly better than a bot that stops at a few seconds?

Maybe the first bot would reach 800 Elo, while the second bot would reach 900 Elo (the numbers are arbitrary, I'm just trying to say it may not be a very big difference).

The same goes for humans, in my experience. When I take a long time to make a move, the move I come up with is generally not that much better than the move I would have made after only a minute (and often times it is exactly the same move). So it is interesting how the duration of thinking doesn't increase the quality of play in a linear manner.

But yes, if the goal is for the bot to be the best it can be, then you'd want it to use all the allotted time (and we should too when we humans play, but we don't have the discipline for that).

[Meeplelowda]

I'd certainly suspect trolling first. I would have thought that taking more than a few seconds to make a move would be superfluous, given how fast machines operate, but I guess it depends on several factors.

No matter how fast machine operates you can't solve chess and many other games. If a goal of a bot is only to win, it should use all the allotted time. The only difference is a better bot will calculate a better move.

That makes sense, though it is a case of diminishing returns. I mean, would a bot that takes minutes to make a move be significantly better than a bot that stops at a few seconds?

Maybe the first bot would reach 800 Elo, while the second bot would reach 900 Elo (the numbers are arbitrary, I'm just trying to say it may not be a very big difference).

This is something you can actually hypothesis test. You can have a chess engine play itself at different search depths (it isn't linear, so you'd have to tune it to find ratios equivalent to using 5 seconds vs. 3.5 minutes, or whatever) and see how the different depths perform relative to each other and whether it indicates a difference at, greater than, or less than an 100 Elo difference. Given the nature of the internet, I would be surprised if someone hasn't already done this.

Less robust than that, but a proof of concept, just look at what move it selects within 5 seconds of eval time and how often it differs from the move it selects at 20 seconds. Of course it depends on the board state. If someone has hung a queen, maybe taking the queen is always the move whether it evaluates for 5 seconds or 100 seconds. But in more nuanced positions you may be surprise how often it bounces around options at lower search depths, then looks like it has settled on a move at say depth 16 to 18, then when it gets to depth 24 it chooses something else.

[lefalaf]

The same goes for humans, in my experience. When I take a long time to make a move, the move I come up with is generally not that much better than the move I would have made after only a minute (and often times it is exactly the same move). So it is interesting how the duration of thinking doesn't increase the quality of play in a linear manner.

Speaking only for myself, I would disagree. If I don't take the time, or am rushed for whatever reason and cannot, I am more apt to miss things, not notice something new and important with the game state, make blunders, and/or chose a less than optimal play. So I play almost exclusively turn-based, or real-time with friends where we allow each other whatever time we choose to take.