Did machine learning reveal symbolism, emotionality, and imaginativeness as primary predictors of creativity?

Creativity is a fundamental but ambiguous concept in the visual arts. It has been demonstrated that there is an association between the perceived creativity of a work of art and its overall quality, but the underlying factors that shape individuals’ perceptions of creativity are not well-established. This discrepancy is the central focus of a recent article published in Nature’s Scientific Reports, in which Spee et al. attempt to identify the perceived attributes of a work of art — e.g., color, symbolism, emotion — that contribute most heavily to individuals’ assessments of its creativity.

Spee et al. conducted a study in which 78 non-experts were asked to rate 54 paintings according to 17 different attributes, each on a 100-point scale. The participants were also asked to judge the creativity of each painting on the same 100-point scale. The authors trained a random forest regression model to analyze these data, emphasizing that this approach is well-suited for capturing the nonlinear relationships that are thought to exist between creativity and the selected attributes. Their results support this hypothesis of nonlinearity and suggest that symbolism, emotionality, and imaginativeness are the most prominent predictors of creativity in Western paintings. In this report, we evaluate the reproducibility and replicability of Spee et al’s work.