It probably depends on the game and type of problem and how much data it is trained on for this game. It also sees and knows what you do and how often you try and fail. So this could be taken into account. Off course I have no clue at the moment, just speculate.