A pair of computer scientists at Rutgers University has developed an algorithm that they claim is capable of discerning true artistic creativity from the stuff that just happens to be in a museum. Good news! It’s totally fine for you to love Munch’s Scream.
Arguing that creativity is really just a combination of “the originality of the product and its influential value,” according to the study, the researchers developed a computational logic that could compare and rank 2-D art against its peers in Western art history. Paintings are compared to one another within their historical context, which takes into account both their aesthetic novelty and influence. The researchers published their results in a formula-laden paper on the open journal arXiv. In their included visualizations, the high dots are the most creative works, and the lowest are the most derivative.
In this overall snapshot of the project, containing the online web-museum Artchive’s 1,710 paintings, you can get a feel for the scope of the project. Pieces by da Vinci, Goya, and Munch top the charts, while the algorithm deems works by Durer and Ingres unoriginal.
Curiously, the pop artist Roy Lichtenstein dominates both here and in other charts broken out by the study (charts that test the algorithm on different sample sets). This even though his work was modeled after comic book art (unique to high art, sure, but also intentionally derivative of an entire genre that this data set doesn’t take into account). The research’s intent was to judge art through the lens of art only, without critics involved. But as a result, I’d argue that Lichtenstein was able to game the system, at least a bit. He borrowed his aesthetic from a place that wasn’t accounted for by the computer.
In this 1850-1950 zoom-in of the above graph, we see that the algorithm preferred Picasso’s Maquette for Guitar over all other cubist pieces it analyzed. Mondrian’s gridded works score high across multiple data sets, but in this graph, researchers actually spotted an error. His piece Composition en blanc, rouge et jaune was mislabeled in Artchive as having been made in 1910 rather than 1936, so the algorithm gave it extraordinary ranking–a red flag for the researchers. The researchers believe that their algorithm may be useful in fact-checking large artistic data sets.
In this final graph, researchers tried something a bit different. They removed the timeline idea altogether, and just had their algorithm judge work by aesthetics alone. What you should see is the most creative work of 12,310 portrait paintings from the Wikiart archive, painted between 1420-2011. But what you end up seeing is a very shallow snapshot of 20th-century work that uses wilder approaches to color, texture, and figure. In other words, when a computer is left to judge creativity, it has no respect for the masters, or the analog world filled with the light and nuance that they were painting. It prefers the cutting edge of abstraction, leveraging pigments and forms that have very little to do with human reality.