How does it affect the chances of winning a turn based zero sum game by a AI algorithm such as Minmax algorithm if the opponent doesn't play optimally or rationally?
Is it possible to defeat a Minmax algorithm by playing not so optimal moves as standard Minmax algorithm assumes the opponent to be playing optimally and thus makes decision on the basis of the above assumption?
If the opponent makes a less than optimal move, but then does better than they would by making the "optimal move"... then the "optimal move" is not the optimal move.
Of course, if the opponent does not make the expected (optimal) move, then you have to rerun the Minmax to decide how to respond -- it may be that the optimal response to the opponent's optimal move is no longer suitable (or possible).
