Improving novice Tetris players with feedback from AI model based tutors

Abstract

Computer based tutoring systems have largely been focused on static tasks, where the tutor guides the student through a single solution. Tutors for dynamic tasks are much more difficult to create, as the solution constantly changes as the task progresses. This study uses Tetris™ as a testbed for a tutor of a complex, dynamic task. Subjects were trained under a number of different feedback conditions using a computer model designed to approximate the behavior of human players. While each feedback system was hypothesized to affect human behavior in a particular way, no significant differences were observed in player behavior after training, but several interesting trends were suggested and continue to be explored.