Sleepwalkers Podcast: What Happens When Machines Find Their Creative Muse

In March 2018, an eerie portrait created by an artificial intelligence program sold at Christie’s Auction House for almost half a million dollars. A few months later, a movie written and directed by an AI algorithm was released amid much hype. And this March, a record company signed an AI artist for the first time.

Artificial creativity is the subject of the second episode of the Sleepwalkers podcast, an ongoing series exploring the implications of AI.

Machine-made art has flourished in recent years, thanks to advances in AI, and some examples are both impressive and unnerving. After all, creativity is something we like to think of as uniquely human.

The episode passes a critical eye over AI art and asks what it means for us to live among machines that seem not only intelligent but, increasingly, creative too.

Filmmaker Oscar Sharp and data artist Russ Goodwin created an AI program called Benjamin that produced the AI movie Zone Out, starring Thomas Middleditch of Silicon Valley. Sharp spent years seeking a technology that would let a machine write creatively, and Goodwin showed him how AI programs can now produce passable poetry and prose.

Take the verse below, for example:

The dream is like a shiny black hair and the sun is like a dream.

I stand up and watch the sun shine on a single day,

and the sun is a chance to accomplish from the springs of my own delight.

It reads like something a tortured beatnik might dream up. Like other “creative” AI algorithms, its creator learns by feeding on example data (thousands of actual poems), before regurgitating something statistically similar.

“Recently, something magical happened,” says Sebastian Thrun, a prominent AI researcher, in the episode. “The field has discovered something called machine learning. With AI, computers can now find their own rules. You just give them examples.”

Of course, AI creativity is quite different than human inspiration. Machines can only capture and reproduce our inventiveness as reflected in training data.

To some, that’s an invitation to be creative. Janelle Shane, a research scientist and author of the blog AI Weirdness, uses AI to create everything from weird pick-up lines to bizarre recipes with ingredients like “chopped whipping cream” and instructions including “fold water and roll it into cubes.”

Shane’s creations show how dumb and imitative AI programs actually are. Her recipes are created by an algorithm, which is itself a form of recipe. It “takes something that’s very ordinary and mixes it up into this sort of surrealistic thing that sounds like the original, but the meaning has been completely changed,” Shane says.

So, while AI won’t make human creativity obsolete anytime soon, it can help us understand the technology’s limits. According to Thrun, the AI scientist, this is worth remembering if we want to avoid creating AI systems that reflect our worst selves.

“I can promise you, whatever you get out reflects the data you put in,” he says. “It’s up to us, the people, to make responsible decisions, and eradicate certain biases in society that exist today. I promise you, if you work hard on this, AI technologies will reflect that.”

More Great WIRED Stories

Read More