The AI box experiment is the logical step after a human has recognized that he will loose any game against the AI. After a chessplayer has lost 10 matches against the computer player, he will come to the conclusion, that he needs a controlled environment in which he can observe the behavior of the AI. This is possible be given restricted ressources to the AI. That means, the human player has the normal chess pieces, while the game AI starts only with a king. Now, the AI can use all it's strength and the human can't be beaten. The ability to think very fast, and the ability to think 100 moves ahead won't help a single king to win against a human controlled setup.
The same AI in a box experiment can be realized with the realtime strategy game OpenRA as well. :The idea is, that the game AI starts with lower ressources than the human. And then the AI has to overcome it's restrictions. Even if the human player doesn't play very well the game, the AI can't beaten him because the AI is in a weaker position. Similar to be restricted to a prison.
What would happen, if the AI is able to understand the situation, can it escape from the prison? Yes and now. Let us observe first the situation with computerchess. A single king, even if it's controlled by the strongest AI player in the world is not able to win against a normal equipped situation which contains of 16 pieces. No matter, which kind of trick the AI is trying to explore, the human is always stronger.
The problem is, that not all prisons are secured very well. In some movies it was shown, that the prisoner can escape. And if a human can do so, an AI has perhaps the same capabilities. The point is, that in an escape game, the AI starts in weaker position. The idea is not to figure out who plays a game better, but the idea is, that the opponent controls the game, and the AI has to overcome the restriction.