Computer scientists have developed an algorithm to analyze paintings and other art to determine which pieces are the most creative. In essence, the algorithm scans images for novelty relative to the art that went before, and then for influence on the art that follows [You can read their paper, Quantifying Creativity in Art Networks, here.]
Below is a graph showing what the algorithm determined were the most and least creative paintings from a set of 1,710 images culled from wikiart.
Behold! The most creative painting in the history of humankind!
I can conclude from the algorithm’s flawless analysis that my own tastes are abysmal, and the best art is well beyond my grasp. Or I can conclude that the algorithm is blind, misguided, and doesn’t even know it is alive, because it isn’t. I’m going to go with the later. The Lichtenstein painting above would make an ugly postage stamp – the horizontal stripes give me a headache – and the algorithm seems to have a strong preference for ugly, minimalist paintings with awkward compositions. [Note: there are a few pieces above the Lichtenstein, but we don’t know what they are.]
The graph shows a spike in creativity in the late 19th century, and this is just one of several dramatic flaws in the whole project. The algorithm equates difference from what went before as equal to degree of creativity: the more different something is, the more creative it is. Consider why this is flawed. If we apply this to music, a composition of single beeps would be about as far as we could get from Beethoven’s late string quartets. The musical creativity algorithm would determine that Beethoven’s quartets were similar to Mozart’s, which were based on Haydn’s, whereas the beeps were completely different, hence more unique and creative than any of the quartets by the three composers. In reality Beethoven’s string quartets would require far more creativity than the beep music. Difference alone does not equal creativity. Because of the onset of geometric abstraction and minimal art in the early 20th century, the algorithm mistakenly identifies several artists working in similar styles as the most creative in comparison to earlier art, but doesn’t register that they aren’t dissimilar to one another.
The ugly Lichtenstein bowl of fruit uses just three colors (assuming you count black and white), has an absolutely ghastly composition, and thus is completely unlike beautifully painted images with subject matter, a lot of colors, and complex compositions. In this instance, the computer interpreted a canvas as great because it was radically different from prior art, when it was different because it was bad. In fact the piece in question is a parody!
“His work defined the basic premise of pop art through parody.” Arnason, H., History of Modern Art: Painting, Sculpture, Architecture
The algorithm doesn’t even attempt to identify or measure beauty or quality [an enormous oversight], in which case the most beautiful image couldn’t possibly be the most creative. It only measures divergence from a shared core, and because it is comparing the art of highly acclaimed artists who are historically striving for beauty, it will inadvertently select ugly art as the most innovative.
Let’s take a look at what the algorithm finds most creative in 20th century art.
And the undisputed champion of the world is…
#121, by Fernando Calhau, 1998.
Sadly the human species hasn’t really caught on to how brilliant this work is, because I can’t even find a reproduction of it larger than a few hundred pixels wide. And it also seems that the algorithm registered it as a painting, when it is a sculpture with a white background. Oops! No wonder it was so different from the other paintings. It also has the same problem as the Lichtenstein – it’s been overvalued because it is minimalist, and in stark contrast to centuries of elaborately beautiful, complex, and subtle images. A few of the runners-up look like ties I bought at a thrift store for “tacky tie day” when I had an office job.
Let’s just look at the others in the top 6 to see if we can find any similarities among the most original and creative art of the 20th century, according to science.
Here we have “Untitled” by Piero Dorazio, and another winner that’s really pushing the limits of creativity. It also might do OK as a pillow case, or a napkin. By me it’s decoration, but fairly well done, especially in the overlapping regions. Creative? Mmmm. Nah. Not really.
Next up another tiny image, this one from Ronnie Landfield. It really is odd that the computer is choosing the most creative images from tiny Jpegs. I know this because I got the image from the same source as the good people who ran the algorithm – Wikiart.
This one has straight diagonals, whereas the previous image had wavy ones. This one would look good in the lobby of a bank, perhaps. Again, it’s decorative, and safe. I like it, but, not especially.
What’s next? Another Piero Dorazio! If we believe the algorithm, at this point we need to rewrite the art history books to include this unsung master.
This one would make the best pillow case, and I can’t say I mind it. It’s kind of like visual muzak, but I think perhaps more tasteful. It’s nice, but, it does seem more than a bit like grandma tampered with the computer and superimposed her own favorite paintings as a prank.
Next up, a pretty painting by Leon Berkowitz. This one has no diagonals. It makes me think of what else the prized paintings have in common, and I might want to designate them as “lyrical abstraction“, which is a kind of kinder, gentler abstraction than the harder geometric abstraction, or the more vigorous work of Abstract Expressionists like Jackson Pollock, or Willem de Kooning. I like the way it glows.
For my tastes these are fairly light pieces that, if they really are extremely creative, it’s not in a way I find at all unsettling, and you could probably sell prints of these as “wall art” at Walmart. But the more I look at them the more I think the algorithm mistook decorative art for novelty in fine art. I guess it hadn’t encountered much decorative abstraction until the latter part of the 20th century, and thus thought it was a radical new departure.
Just one more example:
Here’s another lyrical abstraction with strong diagonals. I’d probably be being a bit rude to call this one a pillow case, except that, yeah, it really is a pillow case. This makes me curious about how the algorithm would process pillow case design.
I’d really like to put the algorithm to test, but to make it more fun, I’d like to create my own deliberately bad piece of minimalist art that hurts my eyes to look at, and then have the computer take a crack at evaluating it. Be back in a few minutes. Below is my submission.
I’ve now shared this in a comment with one of the authors of the paper. I kinda’ doubt they’ll entertain my little experiment, because it could undermine the credibility of their project. But, I did submit it to them to try, and theoretically they should welcome a challenge to the experiment. This also gets at what is probably foremost on our minds when faced with a creativity-measuring algorithm – what will the algorithm make of our own art or art that we are interested in…?
Besides the oversite of the algorithm not being designed to recognize beauty or quality at all, and the critical flaw of it equating degree of difference from a standard with degree of creativity, there are other outstanding problems with the whole project.
First among them is the notion of creativity it uses. Novelty is not synonymous with creativity, and neither is influence. Consider misuses of English, such as using “literal” to mean “figurative”, as in “It was literally raining cats and dogs”; or using “random” to mean “unspecified”, as in “I went into some random store”. They were novel uses of words at one point, and they caught on to the point where you can hardly use them in their proper context anymore. If I say, “I was literally chased by a pack of dogs” (I was, recently), people will assume a couple dogs behind a fence yapped at me from across the street. However, these lazy misunderstandings and misappropriations of language are not creative. The algorithm can’t distinguish between a positive development and an aberration, a good and bad trend. If it were applied to human biology, diabetes might be considered as valuable an evolutionary development as consciousness. One might also look at cultural developments historically. The Inquisition was a new development that lasted centuries, but it wasn’t a positive one.
Next we have the problem of how the algorithm establishes novelty. It does it purely by reducing all images to abstract compositions, and then analyzing them in terms of similarities and differences, preferring in all instances differences. The differences are interpreted necessarily as innovations, and are considered creative if they recur in subsequent art. Content is not registered. Subject matter is not registered. Feeling is not registered. I’m quite sure it would make little or no difference if all the images were submitted upside down. The computer running the algorithm also doesn’t ever “see” actual pieces, but bases all of its calculations on small reproductions collected from wikiart.
Beyond that, the experiment unquestioningly falls completely within the highly contestable paradigm of a linear history of art in which one movement leads to another. This is a terrible starting point to work from because it only considers art valuable, in retrospect, if it is viewed as leading to present styles. To what end? Art is not like science, and styles only build upon one another in a linear progression if artists believe they must, and deliberately attempt to do so, creating a false lineage. When artists fall into this trap they attempt not to make art, but to make art history. Ascribing the most importance and creativity to art which seems to break with the past could easily be countered by only appreciating as relevant art which substantively integrates and builds on prior art. Both those paradigms are probably false, and I believe it is better to look at art for its intrinsic merit independent of one or another grand narrative.
It’s disappointing that people can think that visual art is so simple that it can be reduced to purely formal qualities, that its relevance can be established entirely by its placement within a [skewed and artificial] linear progression of art history, and that it can be analyzed for creativity by a computer using an algorithm. This is an excellent example of “scientism“, which is the misapplication of science to answer questions that are outside of its scope. This kind of reductionism tries to find correlations and come up with conclusions, but seemingly ignores obvious lines of attack, or possible inquiry that might subvert its validity. Did they not even think to deliberately run crappy art by the algorithm to see if it misidentified it as extremely creative?
Finally there is the enormous mistake of presuming that something that is not even conscious can in any way appreciate art. The algorithm is severely misguided, but most importantly doesn’t even know that it exists. You actually have to be conscious to appreciate something produced in consciousness for other consciousnesses. It’s shocking that people could miss the point of art by such a distance.