Abstract
It has long been hypothesized that perceptual ambiguities play an important role in aesthetic experience: a work with some ambiguity engages a viewer more than one that does not. However, current frameworks for developing and testing such theories are limited by the availability of stimuli and data collection methods. We present an approach to measuring the perceptual ambiguity of a collection of images. Crowdworkers are asked to describe image content, and the variability of responses is summarized numerically. Experiments are performed using several classes of images: images created with Generative Adversarial Networks on Artbreeder.com, artwork datasets from previous studies on perceptual ambiguity, and photographs. Based on the hypothesis that perceptual processing is time-dependent, we measure responses after 500 ms and 3000 ms intervals. From these measurements, we find that several difference categories of ambiguity emerge. Moreover, our results address an intriguing paradox at the heart of modernist art history. Unrecognisable or intermediate images have been consistently celebrated for their aesthetic value by historians, critics, artists and members of the viewing public. Yet these very same images have also been attacked and ridiculed by other critics, artists, and members of the viewing public for the very reason that they are unrecognizable and indeterminate. How can we account for this paradox? Our results find that the raters in our experiments can be clustered into two categories: those that prefer ambiguous artworks, and those that prefer recognizable artworks.