The second episode of our podcast, Everyday AI, unpacks what’s going on behind the scenes of creative artificial intelligence (AI). We pose the question: Is creative AI any different from human creativity?
AI in music
Could AI give the popstars of the world a run for their money?
In 2020, Australian band Uncanny Valley won the world’s first AI Eurovision Song Contest with their song, Beautiful the World.
In 2019, the album ‘Chain Tripping’ by LA dance-pop group YACHT was nominated for a Grammy. It was composed, in part, using free AI models available online.
In our podcast, Claire L. Evans from YACHT explains how this works. For the melodies, an AI model was trained to create new melodies using existing riffs by the band.
“We essentially collaged all these different melodies together, structured them into songs like a puzzle,” Claire says. “Then we assigned all those melodies to different instruments, which we then performed live in the studio.”
Another model generated the lyrics. The AI drew from the back catalogue of the band’s 82 songs, plus a huge text file of lyrics from their favourite tunes.
However, much like the melody, human minds were required to arrange the digital outputs into something easy on the ear.
While the use of AI in music may be increasing, it’s still vital to have humans in the mix. As Claire says: “An important part of this process is the human capacity to make decisions and the human capacity to determine what sounds interesting or what sounds good.”
AI in visual art
In 2021, DALL.E, a program that creates artistic images from text descriptions, took the internet by storm.
DALL.E, and its latest iteration DALL.E 2, are built on a model trained on hundreds of millions of images and captions. AI effectively learns to recognise the symbols, images and styles that words represent, and can reproduce them on demand. It does this by breaking down these images and reconstructing them through a process called diffusion.
This raises important questions about the rights of copyright holders of the ingested images. This issue is currently being explored in the legal world.
AI can produce impressive, even beautiful, images. But can it innovate?
AI expert Toby Walsh from University of New South Wales says current systems are limited in their ability to innovate.
“Picasso was technically brilliant. But, equally, what was brilliant about Picasso was his ability to invent new styles,” Toby says.
“He pushed forward the language of imagery and changed completely how we actually depict things. He changed the metaphors, changed the language with which we paint.
“It’s interesting to ask the question if you’re building a system like this, which is just based upon lots of trained data. Will it will actually innovate in those ways and come up with truly new ways of thinking about images and depicting the world?”
The difference between human creativity and generative AI
According to Alison Gopnik, a fundamental difference between humans and AI is the ability to learn new information and expand a knowledge base. Alison is a child psychologist and member of the Berkeley AI research group at the University of California.
“Even tiny babies already seem to have really abstract ideas about how the world works. And yet it’s not as if they just keep those ideas. They revise them, they change them, they go out and do experiments, they observe, they look at the world and they change what they think about the world,” Alison says.
“Now, the question is how can you design an algorithm that could do any of those things? That is still a very challenging question.“
These limitations play out in creative spheres. While AI in art, music and writing is on the rise, its practical use still heavily relies on human prompts, editing and creative direction.
In its current form, AI is perhaps best described as a partner in creativity.
Tune in to the podcast to hear these experts (and more) paint a realistic picture of the power and limitations of AI in creative expression.