Decisions in Tetris are Often Hard: Mapping the Difficulty of Decision Points in a Complex, Dynamic Task

AI tools offer the promise of individualized training and support in a variety of tasks, but to determine when and what kind of assistance to provide, it will be necessary for such tools to gauge the level of difficulty each action poses to an individual subject. This is made more challenging for tasks that take place in dynamic environments where many small-scale actions contribute to a long term goal, but no individual action can be objectively labelled as being "correct". Here, we attempt to map the difficulty of just such a task, the video game Tetris. Using a simple model that is capable of human-like performance at a high level, we take advantage of the model's ability to evaluate all possible actions to determine not only which is the best, but also how that action relates to the other available actions at each decision point. Decisions with a single action rated definitively higher than any other option are considered easy, and decision difficulty increases with the number of plausible actions. We look at the incidence rate of decisions of increasing difficulty levels, and how these patterns vary across players of different skill levels.