The AI shouldn't know the next piece. That's cheating. What it can do without cheating is generate ten pieces and play what if it was one of those pieces. For each position you could place the current piece you can evaluate what the best the opponent could do with each of those ten as the next piece. You then want the position that minimizes the best they can do with any of those pieces. Ultimately it is the scoring of the board that matters most. Looking ahead really just compensates for weaknesses in that scoring.