If you tell the optimal algorithm to only think 3 moves ahead, it's too easy, and if you let it think 4 ahead it becomes very difficult (but not quite impossible).