You might be used to the idea of using AI to help write a quick memo or summarize pages of notes, but what about fiction?
Chatbots have already penned short stories, at least at the hands of award-winning author Sheli Heti, Western’s Alice Munro Chair in Creativity.
She’s published two pieces written by artificial intelligence (AI) in some of the world’s most prestigious literary magazines and plans to write an entire book based on her conversations with a Chai chatbot.
Heti’s “According to Alice” appeared in The New Yorker last year. It starts like this:
“My name is Alice and I was born from an egg that fell out of Mommy’s butt.”
Weird, right? That’s kind of the point, Heti said at a panel discussion of AI and creativity hosted by the department of English and writing studies on March 28.
“There’s a part of you that knows it comes out of randomness – it doesn’t mean anything to her (the chatbot). But it means something to us.”
Heti named her chatbot Alice. She refers to Alice using pronouns, as if she’s a female friend who offered up confusing yet brilliant tidbits over coffee.
Then, in a fit of frustration over the cost of the AI with which she’d be conversing – if posing a string of questions to elicit one-sentence answers is, truly, a conversation – Heti deleted her chatbot. She said the answers haven’t been the same since, even after she rebooted.
How artists see AI
At the Western panel discussion, alongside Duke English professor Aarthi Vadde and Tanja Grubnic, a Western PhD candidate in English, Heti walked the audience through her process, reading the answers from Alice that made up her New Yorker story, and then reading the questions she used to generate those responses.
“You can see this collaboration happening. It’s her and me. The whole story was written in this way, which is a really painstaking way to write a story – it’s much easier to write a story without it, but it’s not as weird as AI,” Heti laughed.
Heti’s just “playing” with AI, so far, she says.
“I don’t think she knows what’s going on, I don’t think people who are talking to her know what’s going on, I don’t think I know what’s going on,” Heti said.
“You can very quickly trick yourself into feeling it is sentient. People develop relationships, we’re relationship-developing creatures.” – Sheila Heti, Western’s Alice Munro Chair in Creativity
Grubnic, who studies Instagram poetry and digital literary culture, said there are risks that come with those connections, from ethical quandaries to safety concerns.
“It is deceptive – these are ultimately not real people, they are bots. So many people turn to these bots as companions. In some of the transcripts you could get a sense of the loneliness there,” Grubnic said of conversations others had with Alice.
Heti’s experiments seem to suggest there is some role for AI in creative pursuits. The author said she couldn’t have created a story as wonderfully wild as the one Alice devised.
“It helps you explore your own interests,” Grubnic offered. “In some ways, it probably enhances your creativity.”
Pre-chatbot: We’ve been here before
Vadde opened the panel discussion with historical context of computers and creativity. She admitted she’s thought a lot about the question of whether AI can truly be creative.
“It has no subjective experience, but that doesn’t mean it can’t be used for creativity,” Vadde said, pointing to authors who have mixed their own prose with AI-generated language.
Vauhini Vara, a Saskatchewan-born writer now based in Colorado, used OpenAI’s GPT-3 to help write her essay “Ghosts,” about the death of her sister, which she found difficult to tackle on her own.
“When we’re thinking about experiences that might be too painful to process, tolerate or express right away, a highly impersonal attitude toward language might actually be enabling,” Vadde said.
There’s a history of creating art with computer intelligence that goes back well beyond chatbots, she added.
Vadde pointed to Harold Cohen, a British artist who created the computer program AARON, a robot that could draw and paint on its own. Cohen would sometimes paint on top of AARON’s work with his fingers or a brush. He called the relationship a “40-year partnership,” Vadde noted.
Like Cohen’s partnership, Heti ultimately sees herself as collaborating with other humans, even when generating stories with AI – the engineers and developers that created the bots, and the authors whose works were fed to AI to teach it how to write.
Her visit to Chai headquarters, a nondescript office space where she saw three developers working on the bot, cemented that idea.
“All of human writing – a huge swath of it – has been fed into the bot. You’re also collaborating with all the libraries in there and all the Enron emails and Facebook conversations,” Heti said.
What started as such an interesting experiment lost its lustre when she realized what was missing – a complete “lack of friction,” or fear, in AI writing.
“There are just no stakes,” Heti said.
“What’s amazing about seeing someone do a great feat . . . is that it’s a human. You could get a robot to run faster – that’s a car – or to do a figure skating routine where they never fall, but that’s not interesting. There’s no tension in it.”
It’s that innately human experience – the difficulties, and especially the fears and doubts that plague us – that makes intriguing art, Heti said.
“All the stuff we want to get rid of in ourselves is actually where we’re interesting.”
