~They say AI can beat any player in poker, but I personally doubt it, and it would be very interesting to try to play poker with such a bot.
If the AI can destroy players at a game like Dota, why wouldn't it completely annihilate humans at a much simpler game, like Poker?
~
Maybe because poker is not a simpler game, or, I'd rather say, it's not a simpler game for computer. Bluffing may seem irrational from a computer's point of view, and the machine will perform it only if programmed to do so. But what a self-learning bot will be doing after several terrible outcomes of bluffing? Right, it will stop bluffing. But every good poker player knows that you can't win without bluffing in the long run.
OpenAI's Dota 2 bot defeated over 99% of all challengers, and they where one of the best pro players around. There are no doubts that AI can beat anyone in Dota in the end.
In January 2017 an AI called LIBRATUS beat 4 pro poker players after playing 120k games, and then later another AI, DEEPSTACK, beat 10 out of 11 expert human players after playing 45k games. And yet I think it is much less certain that an AI can beat a good poker player than that of Dota. There are more unique situations in a game of Texas No-Limit Hold'em than there are atoms in the Universe. There are 10 to the power of 160 outcomes for each game. There is no way even the most powerful computers could actually consider all of those possibilities, but even if they could, they would never know for certain whether an opponent is bluffing or not.