Quick Study: June 20, 2024
Jo Reed: Welcome to Quick Study, the monthly podcast from the National Endowment for the Arts. This is where we'll share stats and stories to help us better understand the value of art in everyday life. I'm Josephine Reed. Sunil Iyengar is the pilot of Quick Study. He's the Director of Research and Analysis here at the Arts Endowment. Good morning, Sunil.
Sunil Iyengar: Good morning, Jo.
Jo Reed: Okay, I'm ready.
Sunil Iyengar: Well, Jo one of my favorite things is watching great artists evolve over time, to see how they have different phases of work, whether it's called their early, middle, or late period.
Jo Reed: Yes. I really like watching artists develop.
Sunil Iyengar: Yeah, and part of what fascinates me about these protean figures is how, not only are they so prolific, but how the market and the critical establishment seem to go along with them when they make these changes. There are any number of visual artists for whom this is true. In literature, you have poets like Yeats or Auden that led long lives and practically mold new sensibilities with each sudden shift of idiom. Or in music, the Beatles or Bob Dylan, or more recently, someone like Beyonce, especially with her country turn. I could go on.
Jo Reed: Yes, you could. I mean, I would yell Beethoven at the drop of a hat.
Sunil Iyengar: So what I'm really trying to say, Jo, is it's an interesting research question to know how an artist who has a following, but who makes some bold innovations, how far the market will go along with this to follow the artist, and whether the artist will, in effect, get rewarded for moving away from their peers or from their own previous work.
Jo Reed: Okay, I feel a study coming on.
Sunil Iyengar: Well, wouldn't you know? As it happens, there is a research team that has studied this exact question. They published a paper last year, and I only got to hear about it the other day in a virtual research symposium led by one of our NEA research labs, the Arts, Entrepreneurship and Innovation Lab at Indiana University. You see, the symposium was called Arts Engagement in an AI World.
Jo Reed: Okay, back this one up. What does AI have to do with this? Is this about AI now?
Sunil Iyengar: Not exactly, or not completely. At the symposium, which still can be viewed at the AEI lab website, one of the speakers, Mitali Banerjee from McGill University, she's the lead author on that paper I mentioned, shared her work, and described how her team used AI to help answer that research question. So to be specific, Banerjee and her team wanted to understand to what extent market pressures, or alternatively, the pressure to conform to the styles of one's peers, would inhibit artists from producing works that might be regarded as innovative in light of what went before. Put another way, are artists rewarded for creating art that is distinctive from that of the past, that of their peers, or even from their own previous work, and do the results depend on how established the artist already is, both in terms of a peer network and also market recognition?
Jo Reed: So what the art actually conveys is kind of meaningless in this equation. The point is how the market responds to it.
Sunil Iyengar: Yeah, okay, so it has that kind of angle on it. It's not so much about what the arts are conveying to people. It's more about how artworks differ from each other when they're produced by the same artist over time, say in that artist's career, and whether innovations that are made in the way an artist produces that work have any market consequences, or are they relating to the audiences in a more profound way, perhaps, than in the past?
Jo Reed: But how would somebody even begin to tackle that through research? But then also, how does this account for mold-breaking and innovative young artists?
Sunil Iyengar: Right. So let me answer your second question first. The study tells us absolutely nothing about the role of an artist's age, whether young, middle-aged, or older, but as you'll see, the study does account for whether the artist was more or less established in artist circles and, or in the marketplace. So I'll get to that in a bit. But for your first question, how would you even begin to research something like this, here's exactly where AI steps in. The researchers decided to focus on a unique period of modern art from 1905 to 1916, a period that saw the emergence of cubism, fauvism, and other sort of movements. They looked at more than 12,000 paintings from 153 artists working in this period. When I say they looked at, what I really mean is they had a machine look at them. The team used a deep learning tool called a convolutional neural net. It's designed for computational image analysis. So this tool has the ability to extract what are called feature vectors from visual images and compute how close or distant they are from the feature vectors of other images. In other words, how distinctive they are. So it's all very complicated. I'm sure it's totally out of my league in terms of my being able to parse the technical details, but the main point is they used the tool to figure out when and how the 153 artists produced work that differed significantly from that of the past, that of their peers, and that differing from their own previous work. So when looking at how the artworks differed from those in the past, for example, the researchers examined 2,000 works from the 19th century by way of a reference group. Oh, and one more thing, Jo. In training the model, the researchers didn't feed it any new information about the artists or artworks. Instead, they trained it on more than a million non-art images from everyday life. They did this because in the researchers' words, “This kind of training better reflects how humans perceive differences in visual stimuli in their visual environment. As such, the algorithm is able to detect nuanced distinctions between images,” the researchers write.
Jo Reed: I have questions.
Sunil Iyengar: Yes.
Jo Reed: So one, it's also taking the art completely out of context because between 1905 and 1916, the Great War happened, and that was a huge marker in terms of society, in terms of the arts, economically. It was a paradigm shift.
Sunil Iyengar: That is so astute, Jo, and you're right. I mean, all these kinds of contextual factors, sociopolitical, other things maybe having to do with the life of the artist, or that are unique to the artist's milieu, those kinds of things are not really the focus of the study, and therefore constitute some of the limitations. I think what they're trying to do here is basically, you have a group of researchers who are trying to understand using this novel technology, how innovation is rewarded within the arts, and so their premise is, let's use this tool to figure out how images differ from how they would have been constructed. I know this sounds very paint-by-numbers the way I'm describing it, but how the artist's oeuvre changes over time, and how it relates to their past works, and also how it relates to what their contemporaries are doing, and so you can actually use some of these mathematical models apparently, to kind of, if you will, objectively identify changes in the way this is manifested visually across time. So that's what I'm trying to indicate.
Jo Reed: Okay. So this is going to sound snootier than I mean it to but that's frankly, something any good art historian could do, but it seems like what they're then going on to your original question: how those innovations end up being rewarded in the marketplace, and that's really what the big difference is.
Sunil Iyengar: Right, but I will say yes, but also yes, and. It's true that art historians can do this, but you're talking about a volume of empirical data, really. I know that sounds like a horrible way to talk about paint strokes and vectors of what they call feature vectors, but essentially, they're using the artworks as data to try to analyze patterns that can be identified, and then use those patterns to compute departures from the past, and so in this very formulaic way, they're essentially saying, these are these mathematical differences between the ways artists portrayed things in the past versus how they're portraying them now, and how the artist themselves has changed in their own works, and then you're right, then they're connecting it to whether the market is rewarding them for this, and also rewarding them for where they sit in relation to other artists.
Jo Reed: So tell me what grabbed you about this particular study.
Sunil Iyengar: So what got me about this is actually their conclusions. So just to kind of complete the circle here. H aving gone through this exercise of identifying the points of departure of works of modern art from earlier works, or from works by their contemporaries, or the past, the researchers looked at how many exhibition opportunities were provided for each artist, and in which geographic locations. T his is how they were able to gauge the market rewards for the artists to see whether those opportunities grew or shrank based on the degree to which the artist departed from conventions in their own work, or compared to works in the past, or works of their peers. What's cool about the study then, to answer that question, is not only this technological approach, but also how it seems to be the first one to look at how rewards change depending not only on the degree of innovation in the producer, but also based on the producer or artist's status in a hierarchy, either in relation to their peers, or in relation to what their market status is. So putting it really simply, if an artist doesn't have much of a following to begin with, you wouldn't expect the fact that they make a sudden departure from their past work, or the work of their peers, you wouldn't expect that to have a very big impact on their market, would you? By contrast, if an artist already has a big following, like some of the artists I mentioned at the top of this episode, you might expect the market to follow their changes from work to work.
Jo Reed: Is that what they found?
Sunil Iyengar: Well, so yeah, that's essentially what the study found. To quote the study's abstract, “We find that artists are, quote, rewarded for distinctiveness from prior and current competitors and their past selves up to a point. However, artists' autonomy to differentiate themselves depends on their position in the social structure.”
Jo Reed: Can you sort of break that down? Thank you.
Sunil Iyengar: Sure. So what they mean is, if an artist has strong ties within a network of peers, and therefore has one kind of status within a network of other artists, they tend to reap higher rewards for departing from their competitors, for coloring outside the lines, as it were. This is reflected in their opportunities to exhibit work, but what's striking is that even artists who have weak ties within a network of peers are rewarded when their output is rated distinct from their competitors. At the risk of oversimplification, even artists who are on the margins of the establishment can create a following by doing something different from what their peers are doing. It's really those artists who fall within the middle, who don't have strong or weak ties with their peers who don't seem rewarded for innovation. At the same time, artists who are already doing very well in the market also reap higher rewards when their work departs from their own past creations, but they get fewer rewards when they stray from the styles of their current competitors. So overall, the researchers suggest that while markets can either reward or punish artists for innovating from a previous reference point, it does seem that peers, quote, strive to constrain each other to conform.
Jo Reed: So is there any sense of what the implications are from this study for either the art world or other sectors? Because honestly, it seems like an awful motivator to make art.
Sunil Iyengar: Believe me, Jo, I couldn't agree more. As a motivation to make art, absolutely, but I think the analytical methods the researchers developed do have long-term value. I mean, this deep learning technology, the researchers point out, can be applied to other types of visual analysis involving fields as varied as fashion, architecture, or advertising, but in a broader sense, the study shows, if you will, the parameters for innovation when it comes to market recognition and the place of the artist within a social hierarchy. So I can imagine there'd be all sorts of lessons here, or at least future research questions for firms that are trying to decide when to pursue product differentiation, for example, and when to conform to the tried and true, but those are just my hunches. It is quite an unusual study, which I thought was worthy of attention.
Jo Reed: Yes, I can see why you would bring it up. Thank you, Sunil.
Sunil Iyengar: Thanks, Jo.
Jo Reed: That was Sunil Iyengar. He's the director of research and analysis here at the National Endowment for the Arts. You've been listening to Quick Study. The music is We Are One, from Scott Holmes Music. It's licensed through Creative Commons. Until next month, I'm Josephine Reed. Thanks for listening.
In this episode of Quick Study, we ponder the use of computational image analysis to learn how differentiation in art translates to market rewards.