Less is more: Additional information leads to lower performance in TETRIS models

Expert performers in complex tasks synthesize a wide variety of information to select the optimal choice at each decision point. For the task of Tetris, the synthesis includes information about the “next” piece in addition to the configuration of pieces currently on the board. While simple models of Tetris are capable of behavior similar to high level human players most (to reduce the combinatorial explosion in computation time) are only aware of the active piece and its possible placement positions. To explore how additional information contributes to expertise, when placing the current 'on board' piece, our model also considers placements for the “next piece” (visable to humans in the Preview Box). Though we expected this additional information to result in higher performance, we instead observed a drop in performance, and a shift in behavior away from common human patterns. These results suggest that human experts are not incorporating the additional piece information into their current decision. We speculate about the role of next piece information for expert level players.