I think this article's key conclusion is incorrect, and that is is likely possible for modern AI techniques to beat Pokemon. The main reason a machine has not

I think this article's key conclusion is incorrect, and that is is likely possible for modern AI techniques to beat Pokemon. The main reason a machine has not already beaten Pokemon is that it is a nontrivial amount of work to connect a standard AI algorithm to a new video game, and nobody has crossed that hurdle for Pokemon. Mario is one of the most popular video games of all time, and so it is one of the few games that people have connected AI to. If there was an AI that was connected to a Pokemon game, I am fairly confident the AI would be able to beat Pokemon. The article discusses the problem of having an unclear goal metric. That doesn't seem like a very hard problem to me - you can start with something like, win as many battles as possible. You might beat the game just randomly after you become powerful enough, or you might need some more tweaks to the metrics, but it doesn't seem like that should be a showstopper. A lot of people are pointing out that Pokemon is fundamentally harder than games like poker or go. That is true, but the bar in this article is beating the game. For poker or go, AI is now better than any human. That is a much higher bar, that is not even relevant for the single-player Pokemon game.