A computer can predict if you prefer Rothko or Monet. Here’s how.
Your art preferences may inform what you buy in the future.
From towering, color-blocked Rothkos, to the soft brushstroke of Monet’s landscapes, one’s taste in art seems like a deeply personal choice. What moves you is a purely human reaction.
Or is it?
A recent study published in the journal Nature Human Behavior has shown it’s possible to accurately predict art preferences, using a deep-learning neural network that did not include any previous art training.
And while you might think of your own personal art style as boundary-defying and genre-bending, the study found that most participants’ art preferences can be grouped into just three categories.
If you aren’t sure which group you fall into, this new A.I. just might, Kiyohito Iigaya, a postdoctoral scholar at California Institute of Technology and first author on the study, tells Inverse.
Close to 80 percent of participants preferred more realistic paintings, like landscapes.
Understanding the human mind
This new work is much more than a “gotcha” to art critics worldwide. John O'Doherty, a senior author on the study and professor of psychology at California Institute of Technology, instead says in a statement that it’s a step toward helping researchers — and their algorithms — better understand forms of abstract human thinking.
“The main point is that we are gaining an insight into the mechanism that people use to make aesthetic judgments," says O'Doherty. "That is, that people appear to use elementary image features and combine over them. That's a first step to understanding how the process works."
Right now the research team has restricted its focus to paintings and photographs, Iigaya says, but in the future the applications for such a technology might be even wider. If the eruption of intelligent marketing algorithms in recent years is any indication, an algorithm like this may even one day help companies predict what kind of packaging or products you’re most aesthetically drawn to.
The second most popular “cluster” or art was abstract works of art such as Rothko’s color fields.
How does it work — Before they could set their artistic algorithm free, the team had to first collect data from human beings on how they felt about the aesthetics of different artworks. Iigaya says that a random selection of art spanning styles like cubism, impressionism, color fields were downloaded from Wikiart.org and presented to groups of both 1,359 digital volunteers and seven in-person volunteers.
When presented with the artwork the volunteers were asked to rate:
- How much they liked the work on a three-point scale (where zero meant they didn’t like it at all and a three meant they liked it a lot)
- How the work aligned with characteristics like being dynamic or still
In total, the seven in-person volunteers rated 1,001 paintings while the online group rated about 60 each.
The paintings were also evaluated by a machine vision algorithm that looked for “low-level” aesthetic patterns (e.g. color, saturation, blurred edges ) that might inform the human participants’ “high-level” judgments.
The least popular cluster of artwork were dynamic pieces like these which were both more abstract than the impressionists but more dynamic than some of the abstract work.
By combining this data, the neural network was then able to predict whether or not a volunteer would like a new, previously unseen painting with high accuracy.
“I used to think the evaluation of art was personal and subjective, so I was surprised by this result," said Iigaya in a statement.
Where do you belong — As if breezily predicting your bespoke artistic fancy wasn’t demoralizing enough, the team also found that people’s art preferences could be clustered into just three groups:
- 78 percent of volunteers preferred realistic paintings of people or scenery (including impressionist works)
- 15 percent of people preferred less concrete works, like abstract art of color fields (e.g. Rothko)
- And only 7 percent of volunteers preferred more dynamic images, like Picasso's cubism
But before you bemoan the mundanity of your art interests — or those of others — the authors write in their study that there are still many other human considerations about art preference that their neural network can’t yet capture. Not to mention, the works used in this study barely scratch the surface of all known masterpieces.
“We note that this is by no means a complete enumeration of the features used by humans [to judge art],” write the authors. “For instance, the semantic meaning of a painting, its historical importance and memories of past experiences elicited by the painting are also likely to play important roles.”
For example, maybe you hate the look of the Old Dutch Masters paintings, but your grandmother had a print of “Girl with a Pearl Earring” hanging in her living room.
“Thus, rather than offering a feature catalog, our findings shed light on the general principles by which feature integration yields aesthetic valuation,” the authors conclude.
Even if you’re not aesthetically drawn to a piece of art, you may still have a less tangible emotional connection to it.
As for Iigaya’s art preferences, he tells Inverse that he’s not yet subjected himself to the neural network’s keen eye, but says it would be a “good idea.”
Iigaya says the next step for him and his colleagues is to continue refining their algorithm such that it “actually captures what's going on in our brain when viewing paintings.”
The Inverse analysis — According to Iigaya, the ultimate purpose of this work is to “understand how humans compute aesthetic values.” However, if their work can achieve this successfully and adaptably, then it’s possible such an algorithm could be used beyond communities of docents.
So-called “Big Data” knows an awful lot about us — from our work schedules to our shampoo preferences — but this data is gleaned from the outside looking in. Understanding how the human brain evaluates art or even products could be a huge breakthrough when it comes to targeted marketing.
Whether or not this will constitute legitimate mind-reading is too soon to say, but it certainly could come alarmingly close.
Abstract: It is an open question whether preferences for visual art can be lawfully predicted from the basic constituent elements of a visual image. Here, we developed and tested a computational framework to investigate how aesthetic values are formed. We show that it is possible to explain human preferences for a visual art piece based on a mixture of low- and high-level features of the image. Subjective value ratings could be predicted not only within but also across individuals, using a regression model with a common set of interpretable features. We also show that the features predicting aesthetic preference can emerge hierarchically within a deep convolutional neural network trained only for object recognition. Our findings suggest that human preferences for art can be explained at least in part as a systematic integration over the underlying visual features of an image.
