It is my understanding that if a computer sees that the move it is considering gives the opponent a forced win it won't do it if there is an alternative, regardless of the unlikelihood that the opponent will find it. More generally chess computers, like most poker computers, assume their opponents play like them. In poker that means that the best human players outperform computers in ring games when some of the opponents are weak. It isn't presently true in chess tournaments. But wouldn't that change if there were a bonus for quicker (fewer moves) wins?

Strong computers still get in bad positions against humans where this might matter? What year is this?

The scenario you are describing is somewhat abstract - you're basically asking if the computer would use hope chess (play a conventionally tricks/bad move that exploits weak players hard for a cheapo win) - while you can program that functionality into engines (you can make an engine play more risky for tricks and weaken its objective strength but make it more tricky and human like).

Sub 2000 players make too many tactical blunder early on, that a traditional chess engine would probably punish it faster than any human just because it can calculated deeper) - chess is a highly punishing game where GTO would still destroy weak players quite fast (unlike poker). But yes theoretically if you have a very tactically gifted human, who also plays a brand of chess that is extra brutal and tricky (e.g. Tal or Nezhmetdinov), they could potentially beat up noobs faster than an engine with tricks. However most human players (even masters) are unlikely to punish as hard as engines and ballsy enough to make unsound gambles.

Don't think this would really work in chess as someone could play out completely lost positions that could negate the extra credit. Would also cause certain openings to go out of practice that are interesting but less dynamic. Not to mention collusion where one player agrees to get into a "tactical fight" and lose fast. They do have tournaments where a win is 3 points and a draw is 1 point which gives extra credit to play more agro. In terms of a strong player playing a sub optimal move knowingly happens all the time. For example if you are losing you seek complications and try to mix things up. While the move might not be optimal given engine analysis wise it gives you the best chance to win the game.

I agree with the above post. There is no practical way to assign extra credit in any reasonable manner for shorter wins so none of this makes much sense.

In the odd game or two? Sure. Over thousands of games averaging fewer moves to victory over amateurs than a top engine? No chance.

Chess is a game with complete information. That's the crux right there. Let's assume an engine is able to find the objectively strongest move in any given position (it's not, but it gets significantly closer to this than any human being), that move will always be the one that leads to checkmate the quickest with the best possible reply in mind. If the human opponent chooses an inferior reply, checkmate will be delivered quicker.

Whereas a human in a position in which he holds a winning advantage over his human opponent might consciously make a slightly inaccurate move (which is still totally winning but the computer would assess it as +5 instead of the possible +14 or whatever) to reduce any ounce of counter play his opponent might have, a computer will never do that. The computer doesn't care about counter play, initiative or any of the other esoteric concepts humans have invented to make this highly complicated game more manageable for our brains.

The analogy doesn't transfer because there are no other players or random elements to which any player has to adapt. I suppose if you were dying to have a tournament where both humans and computers participated (which doesn't happen now), then your suggested solution might theoretically work, where gambling humans could take shots against other humans and hope for a quick win while computers would play theoretically correctly at all times, thus allowing humans to play exploitatively.

However, it seems like a solution in search of a problem. No one wants to have tournaments with humans and computers.

This is ludicrous. The way to beat an amateur quickly is not to make the best move in a position, but to aim for a position where they have the greatest opportunity to make mistakes. The blunders it is possible for an amateur to make in a complicated position are orders of magnitude greater than the extra advantage a computer can eke out over a master. Increasing the chance that an amateur will make a bad blunder dwarfs all other considerations when winning quickly is the priority.

Think about it. You're basically saying that to demolish his amateur opponents, Morphy would have been better off going for a quiet variation of the English Opening, rather than the King's Gambit, which is an objectively worse opening but offers the chance for people to quickly go badly wrong.

It would actually be an interesting AI problem to try to write an engine which aims for faster victories. Let's say the engine knows its opponent's ELO. You would have to have a means for the engine to predict what moves an opponent of a certain ELO is likely to make. Traditional engines are very bad at this. A machine-learning approach would probably do well. It would be an even more interesting problem if the engine didn't know its opponent's ELO and inferred it dynamically from their moves during the game.

Even this post is better than most of the stuff DS writes.