A team of California Institute of Technology researchers used a neural network model to predict how people will evaluate different works of art, offering new insight into the potential universality of certain aesthetic principles.
The findings, published Thursday in Nature Human Behaviour, may help scientists understand the process by which humans assess constituent visual components to form an overall impression of a piece of art. The results, although preliminary, may also be of interest to marketers and advertisers as they aim to create visuals that capture the attention of viewers across digital and in-person platforms.
"If you [visit] Facebook or Instagram, you see a bunch of photos taken by random people, but you can actually decide if you like them or not immediately," Kiyohito Iigaya, a research associate at Caltech and the first author on the study, told The Academic Times. "It's not really known how we manage to do that. It's a really complicated problem because the input is very complex."
The Caltech researchers began by breaking down works of art into low-level visual qualities, such as average hue value or brightness effects. They then compiled a team of art experts to assess and weigh more subjective, high-level qualities, such as abstraction, dynamics and emotional valence in various paintings. The scientists found that they could predict how people would rate art based on the low-level conditions alone, although there was a higher degree of accuracy when those conditions were also combined with the higher-level qualities. The results suggested that the subjective qualities of art are mostly built on more objective visual components.
The art world is booming: In the U.S., the arts and cultural sector generated more than $804 billion just in 2016, representing more than five times the value-added GDP of agriculture. And art insiders have a financial stake in understanding which works might be more likely to attract praise and attention and thus fetch higher prices. But there are other factors at play, such as the popularity of a particular artist as well as the scarcity of his or her work, that can influence art buying.
Similar aesthetic dynamics could also influence the fast-growing social media advertising industry, which brought in $41.5 billion in 2020 alone. Advertisers rely on visuals to entice potential customers, who, on Facebook at least, spend an average of less than two seconds on each item in their news feed while scrolling on their phones. Social media companies use recommendation algorithms to feed customers visuals that reflect their buying habits; although those sites aren't necessarily concerned with the aesthetic value of art and photography, they may still benefit by selecting images with basic features that are more likely to attract viewers.
For the study, a total of 1,359 online participants rated their preferences for a random sample of 60 paintings, which were equally dispersed across genres. An additional group of seven individuals assessed a total of 1,001 paintings in a laboratory setting. Both groups rated paintings on a four-point scale, from zero to three: A zero indicated enjoying a painting "not at all," while a three represented liking a painting "very much." The high-level and low-level models could predict preferences among individuals as well as across the two groups of participants. Follow-up tests, which featured 716 photographic images rather than paintings, showed similar predictive power.
The researchers could categorize participants into three distinct clusters made up of individuals who shared overlapping preferences. The largest group preferred concrete paintings rather than abstract ones, although the paintings that they rated the highest were impressionistic rather than photorealistic. Another subgroup preferred abstract works and color fields, such as the works of Mark Rothko or Josef Albers. A final group placed the greatest weight on a work's degree of complexity, favoring cubism and other intricate, layered styles.
The process by which humans interpret art may at first appear straightforward. Our retinas take in a complex visual image. Then, the brain processes the image and renders a subjective judgment: "How much do I like this?" as the authors put it. But attempts to identify the underlying mechanisms by which humans assess the aesthetic value of art remain challenging, especially because judgment about the visual quality of an object may be shaped by prior experiences, historical context and the different ways people ascribe meaning to a particular work of art. "We focused on some universal features that people use," Iigaya said. "But I'm sure that there are cultural differences, age differences and so forth that I think [are] going to be really interesting to study."
This isn't the first time that scientists have attempted to reduce a relatively abstract evaluation to its fundamental components. In 2017, for instance, corresponding author John O'Doherty was part of a study that predicted how humans would evaluate different foods based on their fat, vitamin, carbohydrate and protein content. In that study, the scientists used magnetic resonance imaging data to match activity in the orbitofrontal cortex with participants' assessment of a food's nutritional value. "However, in the case of visual imagery, the sheer visual complexity of one art piece, as well as the enormous variation between pieces, renders the task of identifying the relevant features that underpin this process exceedingly challenging," the researchers wrote in the current study.
The researchers cautioned that future testing would need to determine whether trained experts might assess a random sampling of artistic works in a different manner than the general population. And because they drew from a limited number of genres, they are unsure whether genres that were not included in the study might prove to be a poor fit for the model. The study also could not account for the way that people may shift their attention to different areas of a composition to come up with a final evaluation of a piece.
Besides exploring the pinnacle of beauty and aesthetic pleasure, researchers are also interested in phenomena that occupy the opposite end of the aesthetic spectrum, such as disgust. Disgust — a part of the "behavioral immune system" — may serve as a tool to avoid dangerous contaminants and diseases. Yet people may still find themselves drawn to gross and offensive art that shocks the senses, causing them to reevaluate prior beliefs or view the world in an unexpected way. Thus, a person's degree of "enjoyment" of a particular work of art doesn't necessarily translate to its perceived beauty. "There have to be some features that, innately, you don't really like," Iigaya said. "Using our computational methods, we may be able to probe what sort of features they are actually using to evoke those disgusting feelings."
The study "Aesthetic preference for art can be predicted from a mixture of low- and high-level visual features" published May 20 in Nature Human Behaviour, was authored by Kiyohito Iigaya, Sanghyun Yi, Iman A. Wahle and John P. O'Doherty, California Institute of Technology; and Koranis Tanwisuth, California Institute of Technology and University of California, Berkeley.