Figure 2. .

Real-time generation of the moves. a: After the initial board was shown, two distributions were produced: one based on gameplay model that computes the probability of the move to be played by a strong player (yellow), and one based on the participant’s predictive fixations during the initial board period (blue). Darker colors indicate higher values in the squares. The move probability and prediction distributions or their inverses were then combined to generate four possible distributions from which the presented move was drawn: improbable and unpredicted (used in Exp. 1 only), improbable but predicted (Exp. 1 only), probable but unpredicted (both experiments), and probable and predicted (both experiments). A move was sampled from one of the distributions. b: Distribution of move probability and prediction accuracy in each of the four or two conditions in Exp. 1 and 2. After a move was sampled, it was evaluated with the move probability distribution and the prediction distribution. This generates a move probability and prediction accuracy for each move (in this example, move probability is −2.4 and prediction accuracy is 0.11). Each point in the scatterplot is a move, and the color of the points correspond to the distribution from which the move was drawn. The ellipses represent the 3σ confidence ellipses for each condition, with colors corresponding to the conditions in (a), showing where most of the points in each condition are located. The correlation between prediction accuracy and move probability is close to zero.