To many people’s eyes, artist Mark Rothko’s enormous paintings are little more than swaths of color. Yet a Rothko can fetch nearly $100 million. Meanwhile, Pablo Picasso’s warped faces fascinate some viewers and terrify others.
Why do our perceptions of beauty differ so widely? The answer may lie in our brain networks. Researchers have now developed an algorithm that can predict art preferences by analyzing how a person’s brain breaks down visual information and decides whether a painting is “good.” The findings show for the first time how intrinsic features of a painting combine with human judgment to give art value in our minds.
Most people—including researchers—consider art preferences to be all over the map, says Anjan Chatterjee, a neurologist and cognitive neuroscientist at the University of Pennsylvania who was not involved in the study. Many preferences are rooted in biology–sugary foods, for instance, help us survive. And people tend to share similar standards of beauty when it comes to human faces and landscapes. But when it comes to art, “There are relatively arbitrary things we seem to care about and value,” Chatterjee says.
To figure out how the brain forms value judgments about art, computational neuroscientist Kiyohito Iigaya and his colleagues at the California Institute of Technology first asked more than 1300 volunteers on the crowdsourcing website Amazon Mechanical Turk to rate a selection of 825 paintings from four Western genres including impressionism, cubism, abstract art, and color field painting. Volunteers were all over the age of 18, but researchers didn’t specify their familiarity with art or their ethnic or national origin.
Using an algorithm to reveal patterns in connections between datapoints, the researchers found that paintings preferred by the same groups of people tended to share certain visual characteristics. These characteristics all fell into two categories: “Low-level” characteristics, like contrast and hue, were intrinsic to an image. “High-level” characteristics, like the emotion a painting elicited, required human interpretation.
Once the algorithm was trained, it could analyze these characteristics in new paintings and accurately predict which works a person would like, the researchers report this month on the preprint server bioRxiv. It also correctly grouped the works into categories that corresponded to the paintings’ characteristics and volunteers’ preferences, across and within art genres. People tended to group into three clusters: one that liked concrete, clear images; one that liked dynamic images; and one that preferred abstract art. Even within these genres, however, the algorithm was able to predict an individual’s specific preferences.
Next, the researchers repeated the experiment with six volunteers, showing each person 1000 paintings while using functional magnetic resonance imaging to scan their brains. The scans revealed that the visual cortex—the part of the brain that receives visual information from the eyes—was active in ways suggesting it was integrating the low-level information with the high-level characteristics, Iigaya says. This information, he adds, then feeds into brain regions known to be associated with value judgments, allowing the person to form an overall opinion of the painting.
Finally, to see whether the same process was happening with other kinds of images, the researchers showed a set of 716 photographs to a new group of 382 Mechanical Turk volunteers. The algorithm was similarly good at predicting individuals’ preferences, based on their previous ratings and characteristics of the photos like contrast and motion. Iigaya says this suggests the factors that contribute to whether a person likes an image are universal.
Using brain imaging on something as ambiguous as artwork is ambitious, says Lesley Fellows, a neurologist at McGill University who studies the neural basis of value judgments. “We know a lot about how the brain carries out actions,” such as deciding to buy artwork or spending time looking at it, she says. Why we do things is far less well understood. “The ‘why’ is really fundamental.”
Iigaya acknowledges the sample was too small and not diverse enough to represent all people: Factors such as age, education level, and culture can also affect art preference. But Chatterjee says the brain pathways are likely similar, even if a person’s taste in art differs significantly. “This is not the whole story, just a small variance we can explain,” Iigaya says.