Embracing a Future of AI Creativity
“A lot of people talk about ‘artificial intelligence.’ But what I’m interested in studying is more like an artificial human.”
By Vikram Nijhawan, Contributor
Over the past decade, there has been no shortage of watershed moments depicting the creative potential of AI, or “artificial intelligence.” This was especially so last year, with the proliferation of programs like the art-generating DALL-E-2, and more recently, the controversial auto-completed prose of OpenAI’s ChatGPT (which deserves partial credit for devising this article’s headline).
But for Sicong, or “Sheldon,” Huang, a Ph.D. student in UofT’s Department of Computer Science, 2019 marked the true turning point. He recalled the emergence of This Person Does Not Exist, a program that could generate original, high-quality headshots of unique human faces, after being trained on only a few photos.
Huang remembers coming across an amusing video clip through Twitter, where a user filmed himself generating a seemingly random face on the photo-generating app. The resulting image was identical to the user’s face. Whether this video was authentic or merely an Internet joke, the metaphor is apt: AI has now become so creatively capable, it can now literally mirror our most defining human aspects.
For Huang, the true data-generating process for a new human face is genetic mutation. If by looking at only images of human faces without any knowledge about genetics, AI can generate new and genuine human faces, the natural next step would be simulating humans by looking at only human behaviours. This made him a believer in “deep learning”, and led him to study this subject in his Ph.D.
“I’m not surprised that AI could get so good,” said Huang. “I’m just surprised they got so good so quickly. I remember thinking this development probably wouldn’t happen for at least ten years – and suddenly, here we are.”
Einstein is thought to have said, “If you want your children to be intelligent, read them fairy tales. If you want them to be more intelligent, read them more fairy tales.” When creatives extol this power of art to contain deep truths, they usually mean it metaphorically. But for Huang, recent developments in creative AI reveal deep truths in a more literal way.
“When we read fiction, we are put in someone else’s shoes, and it allows us to empathize with someone we don’t know,” said Huang. “I think AI could also be a tool for that because eventually, we’ll be able to configure it to simulate any type of human personality. Maybe one day we will be able to simulate ourselves, to gain an understanding of ourselves at a depth that wouldn’t have been possible otherwise.”
As a lifelong consumer of science fiction, Huang realized that the robot-dominated futures sketched out by authors like Isaac Asimov could eventually resemble our awaiting realities. But his understanding of AI was only superficial until in 2016, a friend taught him the intricacies of how AI really worked. In 2017, while taking cognitive science classes as an undergrad, he fell in love with HBO’s sci-fi series Westworld. The show presents an alternate future where sentient androids are used for commercial entertainment by humans – a dystopian premise that coincided with Huang’s newfound interest in charting out a more inspiring future through AI.
He went on to found the University of Toronto Machine Intelligence Student Team (UTMIST), serving as the organization’s first president. Later, with Aidan Gomez, Ivan Zhang, and Bryan Li, he co-founded the non-profit research lab for.ai, which was later acquired and became Cohere for AI.
Huang’s new favourite AI-generated image: a Monet-esque illustration of a dapper man riding an elephant into a faux-oil-painted sunset backdrop. The image was DALL-E-2’s literal interpretation of social psychologist Jonathan Haidt’s famous analogy of ‘The Elephant and the Rider’ – the former represents our impulsive, emotional self, and the latter our rational side struggling to wrest control.
For this AI, the final product may reveal the rider’s control in the composition process. DALL-E-2 gave Haidt’s abstract psychological model an added dimension, as the elephant and rider duo in the picture approach a herd of other elephants. AI may be impenetrable to psychoanalysis, but the generated image does contain some subtext pointing to the Jungian concept of humanity’s “collective unconscious.”
“That’s something that completely surprised me, because I didn’t put that in the prompt,” said Huang. “It’s debatable whether DALL-E actually understood the significance of this decision, but for me, that was my first experience of aesthetic chills from an AI.”
“I think creativity happens when different ideas collide, and when you think ‘outside the box’ – AI is great for that. As humans, we have a lot of cognitive biases, which can make it seem like we’re living in a frame. But these machines can extend our imagination and creativity and take us out of our frames.”
With new concerns being raised about AI’s newfound capability, from disrupting our education systems to more existential threats, those preventative frames may be a blessing. The consensus among Huang’s colleagues is that AI safety should be treated seriously, as humanity might only have one chance at getting it right.
Despite, or perhaps because of such necessary limitations, AI will reflect us more than we would ever imagine outside of our wildest fiction. But Sicong Huang has read the science fiction of the past, and hopes to take those predictions with him into the future.
“A lot of people talk about ‘artificial intelligence’”, said Huang, “But the type of technology we’ll soon have – and what I’m interested in studying – is more like an artificial human.”
Sicong Huang will deliver an in-person talk at TedxUofT’s annual conference on Feb. 5, titled, “When AI is more creative than us, what should we teach in schools?”
