Automated scoring of figural creativity using a convolutional neural network.

AB One of the abiding challenges in creativity research is assessment. Objectively scored tests of creativity such as the Torrance Tests of Creativity and the test of Creative Thinking-Drawing Production (TCT-DP; Urban & Jellen, 1996) offer high levels of reliability and validity but are slow and expensive to administer and score. As a result, many creativity researchers default to simpler and faster self-report measures of creativity and related constructs (e.g., creative self-efficacy, openness). Recent research, however, has begun to explore the use of computational approaches to address these limitations. Examples include the Divergent Association Task (Olson et al., 2021) that uses computational methods to rapidly assess the semantic distance of words, as a proxy for divergent thinking. To date, however, no research appears to have emerged that uses methods drawn from the field of artificial intelligence to assess existing objective, figural (i.e., drawing) tests of creativity. This article describes the application of machine learning, in the form of a convolutional neural network, to the assessment of a figural creativity test-the TCT-DP. The approach shows excellent accuracy and speed, eliminating traditional barriers to the use of these objective, figural creativity tests and opening new avenues for automated creativity assessment.