Deni Ellis Béchard on writing an AI-focused novel at the dawn of ChatGPT

Yahoo

28-01-2025

TORONTO — When Deni Ellis Béchard started writing his new novel five years ago, the concept seemed far-fetched — even fanciful.
These days, he said, the society he created for "We Are Dreams in the Eternal Machine" — one that turns to artificial intelligence for salvation following a second U.S. civil war — doesn't feel quite so preposterous.
"It's mildly nerve-racking to release it in a moment where American politics and technology seem really a hair's breadth away from some of the things happening in the book," said the British Columbia-born author.
"It's something I thought was going to seem a lot more speculative that doesn't feel nearly as speculative as I expected."
He came up with the idea while researching an article for Stanford University's alumni magazine that focused on how China used artificial intelligence to censor journalists.
At the time, AI was far less sophisticated than it is today, but the experts he spoke to explained where the technology was heading.
"I was being told: This is a long ways off; we're not going to really see any real advancement any time soon," Béchard said.
Those predictions inspired the technology at the centre of his novel, known only as "the machine."
The machine is mandated to protect humans and does so forcefully by separating people into individual AI-fuelled rooms, resulting in lives spent entirely in hyperrealistic virtual reality, albeit ones where each person can author their own digital fate. They can use the technology to relive moments from their lives with facsimiles of their loved ones or create altogether new worlds — but ultimately, they have to do it alone.
Béchard's premise comes from a sort of thought experiment, based on Isaac Asimov's first law of robotics: "A robot may not injure a human being, or through inaction, allow a human being to come to harm."
The only way that is possible, Béchard contends, is to keep people apart.
"You can't really put humans together and expect no one ever to get harmed," he said.
But how would people respond when thrown into such a situation? What life would they choose, if they could have anything at all? Those are the questions he sought to explore.
The sort of technology that would make that possible is still a long way off — but it was even more distant when Béchard wrote the bulk of his book.
Then, in March 2023, OpenAI released GPT-4, the fourth generation of its flagship large language model and basis for the company's famous chatbot.
Generative AI — the kind that creates content rather than just interpreting it — became the topic du jour, and thanks to ChatGPT and a litany of competitors, far more accessible.
And the tech just keeps getting better, Béchard said, pointing to AI video generators such as Google's Veo 2 and OpenAI's Sora.
"More than anything, that gave me a sense of urgency. It sort of just lit a fire under me," Béchard said.
He also felt like he had to engage with generative AI in some way. With the book already mostly written, he decided to use it both to fact-check what was already in the story and to source quotes from real-life authors he could sprinkle throughout the book.
That process, he said, confirmed one of the central themes in his book: that artificial intelligence is more concerned with doing what it's told than it is with the truth.
"GPT has a tendency to tell us what we want to hear," Béchard said.
"If you really want a quotation from somebody...and it can't find it, it may just make it up."
Many authors — including some of the biggest in the business — have come out against generative AI, accusing its creators of essentially stealing their copyrighted materials to feed the large language models it's based on.
Some have also expressed concerns about a future in which AI-written books flood the market, though Béchard isn't too worried about that — in part because there's not very much money to be made in literature.
Further, he said, AI is "trained on too much bad prose."
"It's also the lowest common denominator in the sense that it defaults to the most likely syntactical structures based on statistics, which gives you a bland sort of writing in general," he said.
But he said AI can be very good at quickly combing through piles of data to find a piece of information — just so long as you fact-check what it produces.
"With any technology, we kind of have to embrace the good and try to steer it away from the bad and try to encourage people, let's not use it to put artists out of work."
This report by The Canadian Press was first published Jan. 28, 2025.
Nicole Thompson, The Canadian Press