But fine, let’s say they do. In that case pick a game where completely random moves don’t win, eg poker. My point was that in a battle scenario, you can’t really surprise a computer by bluffs and feints in the long run.
Poker indeed has a higher skill ceiling, but there is a difference between high-level and low-level play. Low-level players are cheating each other out of an understanding of the game; they win by playing smarter hands, making progressive bets, and cheating lesser players by playing smarter. Realistically you can only expect an AI to succeed at this low-level of play, since it inherently relies on a knowledge imbalance that AI could potentially exploit. But against high-level players? You're playing a different game, one that even Stockfish won't really grok. Poker masters don't behave like an AI at all, but instead like a computer. Their hand has a relative value which signals the importance of positioning a bet. That bet doesn't always have to be a bluff; sometimes it's a pressure tactic to make a smart player depend on bad cards. Multiple high-level players against each other may simply try to win by attrition, sticking to theory-optimal play to slowly grind away profitable bets. After a long enough time (or high enough value), these wins can be cashed-out apropos of nothing. How do you play smart in the long game while defensive in the short? How do you exploit bad odds and maximize favorable ones? If any of these questions were answered completely, Poker would be an obsolete game relegated to robots. But people play it, because there is indeed a game to be played once you fully understand what's on the table.