A metronome ticks time. Not for the student, but for the teacher, who plays a short piano melody. Without missing a measure, the student follows with an improvised, yet derivative, cello run. The student plays the same run again, and then again. “I have it looping, actually, so you can hear the response over and over again,” says the teacher, Jesse Engel, a computer scientist with Google Brain. “And you can hear some similarities with what I played, but it’s not doing the job of trying to replicate what I played. It’s trying to continue it in a meaningful way.”
The student here is an artificial intelligence algorithm; the instrument, a synthesizer. And the real lesson is teaching an audience of hundreds how computers might someday become capable of producing real works of art. Engels is onstage at NYU’s Skirball Center for the Performing Arts as part of the 2017 World Science Festival, along with three likeminded experts. Each of them is there to showcase how they nurture creativity in computers.
Which begs the question: What is creativity? The broadest definition is any nonlinear solution to a problem. Music is a creative way of making noises that sound pleasant. Language is creative communication. Airplanes are a creative solution to the problem of flight. “But the fact that we can build airplanes that fly faster and higher than birds does not necessarily explain how birds fly, or how they evolved to fly,” says Peter Ulric Tse, a neuroscientist at Dartmouth College. Tse is onstage with Engel, but rather than using AI to tackle a creative endeavor, such as music, he believes they are a vehicle for understanding the nature of creativity itself.
In humans, creativity evolved mysteriously. Homo sapiens became a distinct species around 200,000 years ago. Our ancestors’ characteristic (or, sapient, if you will) feature was their huge foreheads: the site of the frontal cortex, where high-level reasoning occurs. But the earliest indications of creativity in humans didn’t appear until relatively recently. A sculpture of a human with a lion’s head—one of the earliest examples—dates to around 40,000 years ago. That, and other archaeological evidence from the same time period means we Homo sapiens likely spent most of our evolutionary history with unrealized creative potential. However, no physical evidence exists to explain what flipped the switch. “Thoughts don’t leave fossils, neurocircuits don’t leave fossils,” says Tse. “All we have are bones and skulls and artifacts.”
Artificial intelligence’s path towards creativity probably won’t ever fully explain how it evolved in humans. At most, it will give neuroscientists like Tse ways to examine the problem laterally. But it could help scientists understand creativity’s theoretical limits. Lav Varshney, another member of the onstage panel, is working on a mathematical theory of creativity. “The way I’ve been defining it is things that are both novel, and of high quality in their domain,” says Varshney, an engineering theorist at the University of Illinois Urbana-Champaign. For example, a new kind of food.
In the case of cuisine, Varshney says he trains his AI to measure “goodness” based on things like hedonic psychophysics—a branch of research that studies the molecular properties of human flavor perception. He does similar work in fashion, feeding his algorithm information on color matching, and so on. And according to his research, creativity has theoretical limits. Varshney says that as you increase the value of both quality and novelty, you get more and more noise. That is, it becomes harder and harder to distinguish the newness, and the goodness, of a thing. This probably explains why the avant garde is so … well, avant garde.
Like Engel, Varshney is also teaching algorithms to compose music. On stage, he demonstrates one that is learning to compose in the style of Bach. But, he points out, this is not pure creativity. The computer learns by having another algorithm—a teacher—progressively introduce constraints—here are different available instruments, these are chords, this what it means to sing in soprano. In essence, the algorithm is replicating Bach’s creativity based, not evolving its own creative genius. As such, AI algorithms are best suited to be creative collaborators.
Which is exactly what Sougwen Chung displays next. Chung is a visual artist, currently an in residence at Bell Labs, who draws with a robotic arm assistant. “I’ve had a lot of human collaborators, and thought it was time to switch it up a little bit,” she says. Watching the pair—woman and machine—work together is mesmerizing. At first it looks like the arm is mirroring her strokes. But as a piece progresses, you see that the arm has its own style. Yes, a style that is derivative of Chung’s—but still not the same.
When Chung first started using the robotic arm—called DOUG—she thought the collaboration itself might be part of the artistic performance. However, she now believes the arm is pushing her to consider new creative frontiers. “When I collaborate with this algorithm, there’s a real randomness and sense of unpredictability to it, and a lack of understanding that’s kind of exciting,” she says.
If that kind of freedom is at the heart of creativity, the next logical question is whether algorithms could ever eclipse human creativity. Engel, who has settled back into his seat after his performance, seems to think the answer is no. “The intentionality is human on both ends of the spectrum,” he says. That is, humans are both the input and the consumer for anything a computer creates. “You can treat it more like a garden,” he says. You control the garden at a high level: planting seeds, watering it, pruning as necessary. But the garden grows on its own.