Computers have been used to create art for more than five decades. But what started out as relatively straightforward programs – programs that required human direction to achieve specific results – has evolved into a diverse array of sophisticated algorithms. One kind of artificial intelligence, so-called Generative Adversarial Networks, or GANs, is of special interest to both artists and computer scientists, because GANs can learn to create images by learning from examples given to them as “training” data. The advent of GAN-ism and other machine learning applications is opening up new horizons for the ways we create, perceive, and communicate about art.
Those new horizons could be glimpsed at Duke’s Rubenstein Art Center recently, as the venue hosted a reception showcasing entries from Duke’s very first AI for Art Competition. Hosted by the Duke +Data Science initiative and the Vice Provost for the Arts, the reception drew students, faculty, and staff from across the Duke community as well as art enthusiasts local to the Triangle. Intended to explore the intersection of AI and creativity through the lens of GANs, the competition received more than 20 AI-generated entries submitted by Duke students, faculty, and staff. Their distinctive creations were displayed throughout the Rubenstein Art Center and garnered curiosity and appreciation from the many attendees and visitors present in the gallery.
“The AI for Art Competition raised awareness of a topic that is culturally on everyone’s mind – AI and its effects – and grounded it in something we can explore together as a community,” said Victoria Szabo, PhD, Associate Research Professor in the Department of Art, Art History, and Visual Studies at Duke and one of the event’s judges. “At Duke we have an environment uniquely suited to cross-pollination and collaboration, and this competition was only the beginning of what is possible in the future,” said Dr. Szabo.
A reviewer and critic with expertise in digital humanities and media studies and a special interest in spatial and immersive media forms, Dr. Szabo has a distinctive perspective on art created with AI techniques. “I did not see this as art generated by a machine. I saw it as art that was created by humans with a machine as a tool, even as a partner or collaborator.” Dr. Szabo thinks that humans tend to want to ascribe agency to the machine, and with good reason, but the idea of art, in her opinion, does not exist without humans to apprehend it.
“Most of the AI art I have seen in the past has been some form of abstraction and generalization or a form of style transfer. As a judge for the AI Competition, what was inspiring for me was how interesting some of the images proved to be due to the sheer variability and fecundity of digital media,” she said.
The competitors for the AI for Art Competition devised novel ways for AI to recreate art works using learned data, generating unusually captivating images that were a fascinating blend of art and technology. Winning submissions for the competition, judged by faculty from Duke’s Art, Art History & Visual Studies Department, and from the Rhodes Information Initiative at Duke (Rhodes-iiD) on the basis of overall originality and aesthetics and technical contribution, were eligible for monetary awards. Duke’s Zachary Monge, a PhD student studying cognitive neuroscience won the competition for his work titled “Abstract Forests.” A team comprising Duke students Alina Jade Barnett, James Hoctor, Chaofan Chen, and Oscar Li, with their faculty mentor Cynthia Rudin, PhD, earned second place for “This Looks Like Art,” while Angier B. Duke Merit Scholar Sachit Menon took third place for his work titled “The Reception.”
“I was struck by the range of approaches we saw to the prompt,” said Dr. Szabo about the winning entries. “The top three each had a distinctive take on what was possible. “The ‘Abstract Forest’ images had a dreamy quality produced by the sketchiness of the outlines in combination with the intense color resulting in almost archetypal moments. ‘The Reception’ looked like a post-impressionist painting, while the Birds from ‘This Looks Like Art’ were something else. What was amazing about them was the sheer quantity, the undifferentiated grid containing plausible, distorted, and outright monstrous results alongside one another. The artistry here is in juxtaposition and execution of an idea,” she added.
Barnett and colleagues have since created higher resolution works and prints of the Birds from “This Looks Like Art” and have donated some pieces to Preservation Chapel Hill for their "May Member Art Exhibit and Sale" opening May 5th.
Fellow judges Scott Lindroth, Vice Provost for the Arts, and Robert Calderbank, Director of the Rhodes Information Initiative, opened the evening and introduced the competition. Lindroth has worked closely with Duke art departments, Duke Performances and the Nasher Museum of Art to expand and coordinate art programming on campus. Matt Kenney, an instructor from the Department of Computational Media Arts and Cultures who was also a judge for the event, introduced the winners and their work.
“We need more bridges between the arts and the sciences, where neither side becomes merely instrumental to the other, and where we continue to heal the rift of the 19th century and explore the art-science connection,” said Dr. Szabo. She thinks Duke has succeeded with the former through the AI for Art Competition and hopes the University will continue to bring together students from diverse subject orientations to explore the artistic potential of AI for their critical and creative practice.
“What I would love to see in next iteration of the competition would be majors from all over campus digging into the challenge, whether individually or in interdisciplinary teams, and maybe gearing up for it through a class or other extended engagements with these systems,” said Dr. Szabo.
“We were very pleased to see the creativity from across all parts of Duke’s campus, and we look forward to building on this foundation,” commented Larry Carin, PhD, Duke University’s Vice Provost for Research, who convened the collaboration of both arts and science for the event.
+DataScience is a Duke-wide program, operating in partnership with departments, schools, and institutes to enable faculty, students, and staff to employ data science at a level tailored to their needs, level of expertise, and interests.