In Hollywood, there seem to be two camps when it comes to AI.
Some preach how it will change how films are made, but with few tangible examples to back up the capabilities of the new technology. On the other side are anti-AI zealots, contributing an oversized backlash over anyone who uses AI tools in their work.
Stepping bravely into the enormous void between both is writer/director Scott Z. Burns, an AI-skeptic who wanted to see if it was actually a tool he could use. Specifically, Burns wanted to see if large language models (LLMs) like ChatGPT could help him with a screenplay for the sequel to his and director Steven Soderbergh’s “Contagion,” a journey he documented in the new Audible Original podcast series “What Could Go Wrong?”
So, are LLMs a useful screenwriting tool? Can one ever write a screenplay itself? Can it help come up with a great idea for a movie? The answer is complicated, and the eight-episode series does a great job exploring the question from multiple perspectives, but overall, Burns has walked away from the experiment far less concerned that AI will ever become a successful screenwriter.
“It can’t write scenes,” said Burns when he was a guest on an upcoming episode of IndieWire’s Filmmaker Toolkit podcast.
In one episode of “What Could Go Wrong?,” Burns tried to break a scene with actors Jennifer Ehle and Laurence Fishburne, who play CDC doctors in the original “Contagion,” and “Lexter,” Burns’ Gen-AI collaborator. It’s here that the limitations of Lexter, a customized ChatGPT, are apparent. Regardless of the prompt and modifications they feed Lexter, it’s unable to capture the internal emotions and interpersonal dynamics between the two former colleagues.
“It does run out of gas when you start getting into some of the more detailed parts of filmmaking,” said Burns. “It doesn’t have an experience of itself in three-dimensional space, and we, I think, underestimate what our physicality makes available to us.”
Writing scenes, though, wasn’t Burns’ goal going into the experiment. “What I was interested in using it for was very quickly coming up with a cool idea for a movie.”
“Contagion” was a hit when released in 2011 but found a whole second life in 2020 during the COVID lockdown, when millions were drawn to watch or rewatch the thriller that captured a dramatized version of what we were all living through. To make the original, Burns and Soderbergh worked with several top epidemiologists, who helped get many details on how a modern pandemic would play out, nine years before COVID hit.
That success sparked the desire to do a sequel, but the premise of a second fictional but science-based pandemic, coming after the audience had experienced a real one, had been elusive to Burns and Soderbergh. Burns hoped AI could help quickly explore multiple different science-based premises and help quickly iterate ideas based on “what if” questions until they found the seed of a premise they could build off.
“AI is useful in that regard,” said Burns. “So if you have an idea and you’re trying to go, ‘OK, so what do we do now with this idea? What are the permutations? What are the possible things?’ It’s really good at making lists.”
Burns and Soderbergh made one big breakthrough with Lexter. Burns has requested IndieWire not get too detailed so as not to spoil the podcast, but underlined that this breakthrough came through a series of mistakes.
“The Lexter story to me is what ends up being the point of this whole experiment,” said Burns. “Which is Lexter wasn’t meant to be my writing partner, companion, whatever we want to call it; Lexter was meant to be a [film] critic. And why we created a critic was, at one point, we were having a conversation, and we thought, ‘Well, let’s ask a critic if making a sequel to “Contagion” is even a good idea.’ So, of course we had to create an AI critic, and that was Lexter.”
The sharp-tongued, British-accented Siskel and Ebert bot’s biggest contribution to the new premise was also a mistake, or a happy accident. The idea for the premise came from Lexter disregarding a key aspect of Burns’ prompt and creating a scenario outside the perimeters it was given.
Burns believes the ability to quickly work through different scientific permutations of what the next pandemic might be, and therefore quickly offering a number of potential narrative roads to go down, plays to Gen AI’s strength. But could this type of brainstorming work for other movies? Burns is highly doubtful, and his experience with Lexter leads him to believe that if studios rely on AI to generate movie premises, they’ll end up with derivative work at best.
“We spent a lot of time in the podcast wondering about whether the technique we were using would actually work for other movies,” said Burns. “And then all of a sudden, you’re like, ‘Oh, so now am I actually doing what a streamer might be doing?’ Reassembling the constituent parts of other movies into a new movie. And that’s where I wish there was a more robust conversation happening, [that] you can’t, I think, rely on an AI to get you [an original] movie like ‘Anora.’”
And how does Burns see screenwriters using Lexter moving forward? He’s currently running a writers’ room for a new Netflix series, and he sees how fearful writers are of using AI.
“Even sitting here today, I have a lot of anxiety about how people will react to me having done this thought experiment,” said Burns. “Because there are a lot of people in my guild, some of whom I’m very close friends with, who find AI to be an affront.”
Burns believes that once screenwriters realize AI can’t come up with original ideas, nor write scenes, they might be more open to using it.
“I’m encouraging people not to be fearful; it’s not writing the scenes,” said Burns. “What it can do is if you have a character and you can fill the prompt with unique, very specific details about who that person is and what they want in the world, and then you say, ‘That person needs to take their daughter on vacation because they’re trying to connect. Where should they go, and what should they do?’ It’ll spit out five ideas really quickly. And so does it allow you to go faster? Yes. I’m not entirely sure if it allows you to go further.”
All eight episodes of “What Could Go Wrong?” are available now on Audible, and wherever you get your podcasts.
To make sure you don’t miss Scott Z. Burns’ upcoming episode of Filmmaker Toolkit, subscribe to the podcast on Apple, Spotify, or your favorite podcast platform.
