AI / Arts / Events

Can ‘intelligent’ machines produce art?

That was the topic of discussion at New Lab Monday night, and it proved to be quite divisive.

Can AI produce a Van Gogh or Picasso? This panel deliberated the question at New Lab. (Courtesy photo)
On the stage at New Lab Monday night, a screen cycled through a slideshow of images, ranging from quirky to psychedelic. Van Gogh or Picasso, they were certainly not — they looked more like the insides of a trippy, pixelated dream. We’re here for that.

The images were produced by Google’s DeepDream, a software program that uses computer vision to enhance hidden patterns within images. (It apparently thinks Marc Andreessen looks like a dog.) DeepDream-generated images have fetched thousands of dollars in the art market. But is this machine-derived work really any good? Can artificial intelligence ever produce great art?

It’s all in the eye of the beholder, of course — making those questions perhaps impossible to answer. But at New Lab’s panel “Every Bot Is a Critic,” four panelists took a stab at doing so, yielding lively discussion along the way. The panel was organized by Nanotronics, one of New Lab’s resident companies, which makes AI-powered industrial microscopes that allow for quicker identification of samples. The event is the first of many, according to New Lab spokesperson Molly Erman, designed to bring the general public into the makerspace and engage with the technology therein.

The slate of speakers included Matthew Putman, the CEO of Nanotronics; Simon DeDeo, a professor of social and decision sciences at Carnegie Mellon UniversityHugo Liu, the CEO and founder of ArtAdvisor, a machine-learning platform that analyzes artists’ bodies of work; and Paddy Johnson, an art critic and the founding editor of Art F City. Justin Stanwix, the chief revenue officer of Nanotronics, moderated the panel.

(Putman previously expounded on the topic of AI and art in an interview with New Lab, which you can read here.)

With the rise of AI, the notion of the singularity — that theoretical point in time when the intelligence of machines will surpass that of us mere humans — has become more prominent in tech discourse. The term was not uttered once during Monday evening’s panel, but the idea of it permeated throughout the discussion.

“Is art the true Turing test?” Stanwix asked at one point.

The panel was evenly divided between those who believed in the potential for machines to create art and those who remained skeptical. Liu was perhaps the greatest proponent of the idea onstage. “Can AI objectively make good art?” he said. “Hell, yes.”

He pointed to one of his AI projects, the “Synesthetic Cookbook,” in which cooks can search for recipes by emotion as well as by ingredient or flavor profile. As a companion project to the cookbook, Liu developed a program that writes original recipes based on similar principles. The resulting work earned him a chef’s profile in the Los Angeles Times, which featured two AI-generated recipes.

One hallmark of AI recipes, as we’ve previously noted, is that they often surface uncommon ingredient combinations. That was the case for Liu’s project: one published recipe, for “demoniacal potato salad,” called for shiitake mushrooms and anchovies. (This Southerner would never dare include either in her potato salad, for the record.) Liu received awed feedback from readers, many of who asked for further recipe tips.

“I had to explain that I’ve never made these recipes in my life,” he said.

Indeed, AI may be more effective than humans at discovering hidden taste patterns among large groups of people. Marketers, as Liu, who previously worked at eBay, pointed out, are already using such tools to predict consumer demand. But Johnson pondered whether art should be defined as having such a utilitarian or consumer-oriented purpose.

“Does that just mean that art is entertainment?” she asked. “It goes back to what we want from art. There’s a certain amount of mythology about art, that it allows us to reach new ways of thinking. Will machines be able to do that? I have no idea.”

Putman argued that art, unlike the examples put forth by Liu, may not have a defined purpose — or even be categorized as art immediately after it’s produced. That, he said, conflicts with how machines go about accomplishing tasks.

“With art, it can take generations to find out what’s good or bad,” he said. “You’re not considering an audience. With AI, you’re optimizing for something all the time. You have to know the question you are asking.”

That observation — of knowing the exact path from input to output — gets to the crux of what makes the possible advent of “intelligent” machines so scary. As DeDeo put it, “machines have no desires,” at least in theory. But in order to produce art, one audience member mused, wouldn’t a computer need to approximate some sort of emotion? DeDeo offered an interesting — and potentially chilling — response.

“One way to define emotion is that it is a thought about something that influences your own self-flourishing,” he said. “What for a machine influences self-flourishing? Somehow the machine has to have values or goals.”

On one hand, those goals could produce work that no human mind could conceive of. On the other hand, those values and goals may well conflict with our own.

Series: Brooklyn

Knowledge is power!

Subscribe for free today and stay up to date with news and tips you need to grow your career and connect with our vibrant tech community.

Technically Media