Quantized Art

The Fate of the Animals by Franz Marc

Quantized art. The idea came about while reading how the music industry assembles top-liners, producers, artists, performers, etc. to create top 40 hits. For example, there has been a recent trend in pop music to use 'drops', when the song builds to a crescendo and then a crazy, catchy bass line is released that causes everyone to dance. This has been perfected to the point where even an okay song can become popular because the producers know when to build, at what moment to intersperse catchy, meaningless lyrics and how to end the song on a high. I like the idea that art (as in paintings, drawings, etc.) can be dissected and quantified. While the two dimensional nature of artwork increases the algorithmic complexity, considering the advances in pattern recognition and other techniques, creating an program, that quantifies essential aspects of art (form, symmetry, color, allusions, etc.) and then produces its own artwork should be possible. But whether this art would be accepted is another issue.

Talking to a friend recently, I broached this idea. The response was a bit negative. Art is different (surprise!) it was argued and can't be broken down, at least, not good art. There are two issues here: can we produce popular art and is that art 'good'? I think the former question is easy to answer; designers and artists have known for a long time what forms, ratios (e.g. golden ratio), colors and other rules are pleasing to people (think Gestalt principles). Whether devising a computer algorithm that incorporates all these rules can lead to 'great' art is another matter. To do this, we would have to define 'great' art, is it how it immediately impacts people? Its historical context and cultural significance? Should the artists story, their experiences leading up to the event, impact it's worth? These are questions for another discussion, but they are worth raising if one wants to quantize art.

Some people have attempted to quantify art, using different criteria. For example, David Galenson uses the number of times a work is referenced in textbooks to rank their importance. But this method is both flawed (it is self justifying since new textbooks largely rely on old textbooks) and not predictive. In other cases people try to quantify art based on color, expression, composition, etc. But again, these are vague, imprecise measures. And they don't prescribe why and how one can reproduce great art. On the more quantitative side, Hughes, et al. use quantitative means to differentiate between true and imitator works of Bruegel. But again, this is less to do with identifying key artistic elements and more to do with pattern recognition.

To implement quantized art, it would be necessary to simulate the creation of art work given a set of constraints (knowledge about the world, tools available, rules of composition, etc.). This could be further enhanced by allowing random injections of noise that give undue weight to certain parameters (say the use of colors and geometric shapes) that would simulate 'revolutionary' artwork. It

The Fate of the Animals by Franz Marc (see above) is by all measures a classic. While it satisfies several rules, e.g. contrasting colors to grab attention, in other areas it baffles and defies conventional wisdom. Without prior knowledge, the subject matter is not clear and it is difficult to discern actual forms in the madness. However, there is a palpable sense of fear and confusion; it is possible to understand that some great sadness or hurt is trying to be communicated. At first, replicating this with a computer may seem absurd, but consider this painting, but with the colors brighten. Would it still have the same effect? To a large degree, we associate certain colors and forms with particular feelings. This extends to how things are interpreted, consider if a government pamphlet was red on green versus gold on white. Thus, given a large enough database of color:association pairings, an algorithm could be feed news about a war, then random tweaks to a parameter associated with what type of feelings should be evoked varied and the results observed. This could be augmented by using Twitter, Facebook or other social networks to analyze posts associated with an event to determine the general mood and then work from their (either going with popular consensus or against it). Whether the chaotic or delicate strokes of a true master can be imitated may just be a matter of raw processing power.

Quantized art would help us better understand what factors might be needed to induce new revolutions in art or better understand what 'geniuses' are seeing as they create their works. With the increased capacity to access and expand large databases, it might be possible to produce unexpected behaviors by having an algorithm combine current and past events, knowledge of its past artistic output, and the rules of composition. The only issue is how to develop an algorithm to understand the abstract or the concept behind a drawing. Because it has no 'experiences', the program wouldn't be able to create a work like 'The Fate of the Animals', if you view the essence of a work as both the physical and spiritual. However, it might be able to produce eye-catching or thought-provoking art for the masses that quickly help convey the mood of the times. This might prove just as valuable as art produced by man, as a way to objectively capture the times in a subjective manner. The implementation will be explored in more detail in a future post.