What Algorithms Can Teach Us About Art

An algorithm has shown that landscape paintings over the past five centuries are surprisingly similar, suggesting selection bias.

AsianScientist (Dec. 8, 2020) – Algorithms developed by scientists from Korea have revealed the evolution of Western landscape paintings over time. The team’s findings, published in the Proceedings of the National Academy of Sciences, suggest a bias in painting composition by art curators.

Whether you prefer classical masters like Vincent Van Gogh or modern visionaries like TeamLab, there’s no question that art forms part of our society’s collective memory. Much like photographs, masterpieces like the Mona Lisa and even Banksy’s graffiti give a snapshot into the dominant way of life across space and time.

“One foundational question among art historians is whether artwork contains organizing principles that transcend culture and time, and if yes, how these principles evolved over time,” explained Professor Hawoong Jeong, a statistical physicist at the Korea Advanced Institute of Science and Technology (KAIST).

To address this question, Jeong, along with collaborators from South Korea, Estonia and the US ran nearly 15,000 landscape paintings through a computer algorithm. The algorithm progressively divided the paintings—which ranged from the Western Renaissance in the 1500s to the contemporary art of today—into horizontal and vertical lines, giving insights into how artists spanning different art styles compose their masterpieces.

The team found that all paintings could be partitioned horizontally or vertically, with up to four different categories, depending on the primary partition. As the name suggests, horizontal-horizontal paintings were overwhelmingly characterized by horizontal components. Meanwhile, horizontal-vertical paintings featured a dominant horizontal component, but with vertical elements scattered throughout the work. Other categories include vertical-vertical and vertical-horizontal paintings.

The paintings were divided into four categories depending on how they were partitioned by the algorithm. Credit: Hawoong Jeong.

Assessing the evolution of the categories over time, Jeong and his team found that before the mid-19th century, horizontal-vertical paintings were common. Following this period, horizontal-horizontal paintings began to eclipse the latter in popularity.

They also looked at how the horizon line, which typically separates the sky from land in paintings, changed over time. They found that the horizon line was typically located in the middle of the canvas in the 16th century. However, it gradually descended to the lower middle of the canvas by the 17th century. Since then, the line has started to rise again.

Surprisingly, the algorithm also revealed that these patterns were reflected across diverse cultures and artistic periods. For Jeong, this may indicate a selection bias in the curated datasets, which were obtained from WikiArt and the Web Gallery of Art.

“In recent decades, art historians have prioritized the argument that there is great diversity in the evolution of artistic expression,” said Jeong. “This study serves as a reminder that the available large-scale datasets might be perpetuating severe biases.”

Moving forward, the team hopes to apply their algorithm to other forms of art—including film, architecture, typography and the like. Diagonal compositions could also be explored, as well as artworks from non-Western sources.

The article can be found at: Lee et al. (2020) Dissecting Landscape Art History With Information Theory.


Source: Korea Advanced Institute of Science and Technology; Photo illustration: Lam Oi Keat/Asian Scientist Magazine.
Disclaimer: This article does not necessarily reflect the views of AsianScientist or its staff.

Asian Scientist Magazine is an award-winning science and technology magazine that highlights R&D news stories from Asia to a global audience. The magazine is published by Singapore-headquartered Wildtype Media Group.

Related Stories from Asian Scientist