Abstract
Intuitively, various factors contribute to beauty judgments. However, there is no empirical evidence that indicates whether rating beauty is unidimensional (e.g., like rating luminance) or multidimensional (e.g., like rating color). A few years after Shannon’s (1948) introduction of communication theory, Miller (1956) used mutual information to describe the limits of human perceptual judgment. Mutual information is the amount of information (bits) obtained about one variable from another. Miller’s review showed that for a single-feature judgment, mutual information plateaus at about 2.5 bits as the number of categories increases. For a multi-feature judgment, the limit is much greater. For example, people can classify over 10,000 syllables, around 13 bits. By placing beauty judgment on the absolute scale of mutual information, we allow comparison with other perceptual judgments. Additionally, the mutual information of beauty hints at its dimensionality. 50 participants rated the beauty of images in four blocks, each containing five occurrences of 3, 4, 6, or 10 images. We selected images based on the mean of their crowd-sourced beauty ratings (Brielmann and Pelli, 2019). Each block contained a low-beauty image (rated 2.7 out of 10), a high-beauty image (rated 9.8 out of 10), and other images rated uniformly from 2.7 to 9.8. Participants rated, on a scale from 1 to 10, the beauty of each image. For each participant and each block, we calculated the mutual information between the mean rating (input) and individual beauty rating (output) of each image. Our results indicate that as the number of images per block increases, the mutual information plateaus at 2.15 ± 0.05 bits, which is slightly below Miller’s estimate of 2.5. Thus, the 2.15 bits conveyed by beauty judgment is close to the 2.5 bits of unidimensional and far less than the 5 to 23 bits of a multidimensional perceptual judgment.