Fig 1.
The creative foraging game allows high-resolution automated analysis of exploration and discovery processes.
Players move squares starting from a 10-square line with the aim of finding ‘interesting and beautiful’ shapes. Moves keep all squares connected (not by a diagonal). A shape can be saved to a gallery by pressing the gray square at the top-right corner of the screen; in this square, the last shape chosen to the gallery is displayed (see Section B in S1 File).
Fig 2.
Search trajectories are segmented into exploration and exploitation phases.
Segmentation uses the timing difference Δt between gallery shape choices. Periods with increasing Δt are labelled as exploration phases. An exploration phase ends with a decrease in Δt, leading to a sequence of rapid gallery choices, labelled as exploitation phases (see Methods, and Section C in S1 File). Choices of shapes to the gallery are marked with an open circle.
Table 1.
Characteristics of exploration and exploitation phases (Medians and 95% CI).
Fig 3.
Shapes found in an exploitation phase are perceived as similar.
Participants who did not play the creative foraging game chose the odd-one-out among four shapes, three from the same exploitation phase and one from a different exploitation phase found by the same player. Participants chose the outlier twice as often as by chance. Means and std (computed using bootstrapping) are shown (p<0.0001, see Section D in S1 File).
Fig 4.
Participants follow optimal (shortest) paths in exploitation phases.
Shown are moves (circles), gallery-chosen shapes (colored circles), actual (gray) and minimal (red) paths (minimal path marked when different from actual path). Exploration and exploitation phases are labelled. Inset, players’ median ratio of minimal to actual path length between gallery shapes in exploitation and exploration.
Fig 5.
Different players discover similar categories of shapes in their exploitation phases.
Network in which nodes are patches (shapes found in an exploitation phase) from 100 players. Links in the network connect patches that share at least two shapes. Categories are defined as Girvan–Newman modules in the graph, and are shown in colors, with representative shapes (see Section F in S1 File).
Fig 6.
Players show correlated mean exploration and exploitation phase durations.
Mean and error bars of one std are shown for each player. Spearman correlation, r = 0.78, 95% CI = [0.68,0.86], p<0.001.
Fig 7.
Discovery of a new exploitation phase occurs at a transition shape that is ambiguous in the sense that it belongs to multiple categories.
A) Examples of transition shapes at the entry to the category of ‘digits’ by four different players. Transition shapes are non-prototypical, for example a ‘four that is not a four’, whereas shapes found after the transition shape tend to be more prototypical digits. B) A shape is ambiguous if it lies in the intersection of two or more categories (signifying at least two contexts of meaning), as exemplified by the trident shape shared by the categories of ‘airplanes’ and ‘English letters’. C) Transition shapes are more likely to be ambiguous shapes than non-transition gallery shapes (transition shapes: Median = 50%, 95% CI = [49, 50]; non-transition shapes: Median = 15, 95% CI = [15,16]).