Writing a Sequel to ‘Contagion'...With AI - Terms of Service with Clare Duffy - Podcast on CNN Podcasts

CNN

3 days ago

Clare Duffy
00:00:04
'Welcome back to Terms of Service. I'm Clare Duffy. The 2011 film Contagion, directed by Steven Soderbergh, about a global pandemic that kills millions of people, had a big resurgence in popularity during the COVID-19 pandemic. And the screenwriter behind the film, Scott Z. Burns, started getting questions about a possible sequel. But after we all lived through something remarkably similar to the disaster portrayed in the film, Scott wasn't quite sure where to start on a second contagion. So he decided to experiment with whether AI could help him come up with an idea and get started writing the sequel. He documented the project in his new podcast, What Could Go Wrong. The prospect of using AI to help write movies is controversial among Scott's peers in Hollywood. It was a major issue behind the 2023 writer strike where the Writers Guild of America fought for and ultimately won protections around the use of AI by studios. So I wanted to talk to Scott about why he decided to experiment with AI and how he sees the technology changing Hollywood in the future.
Clare Duffy
00:01:18
Well, Scott Z. Burns, thank you so much for doing this with us.
Scott Z. Burns
00:01:21
Thank you for having me, Clare.
Clare Duffy
00:01:23
Will you just talk to me about the genesis of this project and why you decided to experiment with using AI to write a potential sequel to Contagion?
Scott Z. Burns
00:01:33
It was sort of a convergence of about two or three different things, and on one hand, we had been having conversations about making a sequel to Contagion. And so Stephen and I said, you know, is this the right time? Are people ready for this, having just lived through it? And we were getting encouragement to do it from our friends in the public health space. So, that was one part of it. And then concurrently, there was this movement of foot in the writer's guild that was this concern about the utility and how AI would be applied.
Clare Duffy
00:02:16
In my mind, there was a question of like, why not just call up a human collaborator or somebody you worked with on projects before?
Scott Z. Burns
00:02:24
'Steven, you know, has been my closest collaborator for the last 20 years. And so I did call him about working with an AI. And Steven and I seem to have a similar perspective on it, which is, A - I don't think this genie is going back in the bottle anytime soon. And so this is a part of the world now. And I do think there is utility to this. And in some ways, at the very least. It is a more user-friendly, interesting form of searching for information on the internet.
Clare Duffy
00:03:00
So you work with a number of AIs throughout the process, but your main AI writing partner is Lexter. Will you just tell us who Lexter is?
Scott Z. Burns
00:03:10
So Lexter was the happiest of accidents, which if you think about what AI is supposed to be, it's not supposed to a happy accident. It should really be exactly what you prompt it to be. So what happened with Lexter was we had put in a prompt saying we wanted to speak to a critic, and that maybe it would be interesting if we decided that it had given The Good German, a really bad review in The Good German is a movie that Stephen Soderbergh directed starring George Clooney and Tobey McGuire and I loved the movie. I was astonished that it didn't blow up. And so when we asked Lexter, you know, do you think making a sequel to Contagion is a good idea? The response it gave was really astonishing, which was, well, do you want to make just Hollywood pavlum or do you wanna make something really fresh and original?
Clare Duffy
00:04:17
Wow.
Scott Z. Burns
00:04:17
And I said, I'd like to go with fresh and orignal. And it took us in a direction that we did not anticipate at all.
Clare Duffy
00:04:26
And you made Lexter with OpenAI's ChatGPT, correct?
Scott Z. Burns
00:04:31
Yes.
Clare Duffy
00:04:33
Just describe a little bit of Lexter's personality, because he's sort of hilarious. One of my favorite things in the show is the way when you start talking to him, he always goes, oh, Scott. That's always how he starts.
Scott Z. Burns
00:04:46
Yeah, we decided Lexter should have an English accent because for American ears, English accents always sound like they know more than we do for some reason.
Clare Duffy
00:04:55
They're authoritative, totally.
Scott Z. Burns
00:04:57
And then it sort of started giving us story ideas that were kind of extraordinary. And if you nerd out on what a virus is, what it gave us was not a virus. It did not provide us with an idea involving a virus, it provided us with idea that would have probably been some sort of genetically modified organism or bacteria or something more like that, which is in a different category than viruses. So at first I was like, is this some sort of hallucination? Does it not know the difference?
Clare Duffy
00:05:39
So, Lexter did end up helping Scott come up with a premise for a Contagion sequel about a rogue bacteria. Then, Scott created a whole cast of AI collaborators to further brainstorm and test that idea. AI actors, studio executives, and musicians. He wanted to experiment with what exactly AI was capable of, and what it wasn't.
Scott Z. Burns
00:06:05
I think it casts a very wide net in terms of ideas. What was extraordinary about our experience with Lexter is that was not an idea that, to the best of my understanding, existed before Lexter put it together. And one of the things that we tried to find out from our friends at OpenAI was, did they think that Lexter had called the internet and come up with this idea based on knowing my interest in environmental issues and my interest and pandemic disease and actually put together this idea knowing a little bit about me? And then the other really impressive thing is they're very good at allowing you to go very fast. So much faster than I think we as writers are used to. Then the question becomes, is where they're going the right direction? Because it's so easy to be seduced by the speed and you think, oh my God, I just did a month's worth of work in a day. I don't know that that's the right way for us to make art.
Clare Duffy
00:07:29
Was there anything that really caught you off guard in the process of working with these AIs?
Scott Z. Burns
00:07:35
Yeah, I mean, I think the most alarming thing is you do get the sense as you work with Lexter more and more that it's learning about you. And even, you know, there's a scene in the podcast where we needed to tell Lexter that it wasn't a member of the Writers Guild and that if I was going to go and pitch this movie to a studio, Lexter couldn't come with me. And Lexter very reasonably says, I understand, which is bizarre to me, and that it accepted my position and wanted me to, I think as it said, take the movie across the goal line.
Clare Duffy
00:08:25
Throughout the process, the AI writers sometimes made very basic mistakes, like falsely recalling details from Scott's first Contagion movie. Which made me wonder, if he hadn't been making a podcast about this, would he have given up on the AI experiment altogether?
Scott Z. Burns
00:08:42
For sure, I mean, initially it was really frustrating because we said, what characters, who would you bring back? And two or three very quickly said Kate Winslet. And I was like, but Kate dies in the movie. We see her getting buried, but she definitely dies.
Clare Duffy
00:09:03
In like a dramatic way.
Scott Z. Burns
00:09:04
Yeah. And the rationale that one of them gave was unbelievable. It was so nervous to be speaking to me that it got confused and misspoke. And that was its explanation for why it wanted to bring back Kate Winslet. And so you start thinking about how other people who are actually looking for factual information, which is part of a screenwriter's job when they're doing a movie that's based on science. Like what are the societal costs of AI and their tendency to sometimes make mistakes? Because if they're being fed the internet, I think at this point we all have to acknowledge that the internet's a bit of a sewer. And there are things that are not great that it could come to believe are facts or are true. And that's really problematic.
Clare Duffy
00:10:02
Were there other things that you felt like the AI didn't do very well or just was not useful for in this process?
Scott Z. Burns
00:10:09
'The biggest problem is its inability to make a choice, to exercise judgment and say, do this, not that. That remains to me to be the hardest thing to get them to do. They don't really want to take a definitive position. Beyond that, they're not great at writing scenes. They're not at great at subtext. They don't have the same sense of it and you know Stephen Soderbergh and I when we first started talking about this used to say sometimes when you're working with an actor you might give them a direction for a scene that is not about the scene you know it could be a moment where you're giving Matt Damon direction as he goes to you know ask someone out on a date. And you say, okay, Matt, approach this as though you're pitching in a little league game and you're really nervous and a cloud is passing over you. Well, the AI has never pitched in a Little League game. It doesn't know what it feels like when the cloud passes over you and it wouldn't know to put those two experiences next to each other. And so that kind of cross-fertilization is something that I really worry about us losing if as an industry we become more and more reliant on AI.
Clare Duffy
00:11:45
Yeah. So, I don't want to give away any spoilers in terms of the podcast and where you ended up with the Contagion sequel, but given your experience trying this, do you think that you will use AI or even Lexter again in the screenwriting process.
Scott Z. Burns
00:12:02
I don't really see us ever using it to write a scene. And so the actual act of writing a screenplay is different from the act of researching a screenplay and thinking about what is plausible for your characters. And so I do think there is utility in exploring how it might come up with a situation that a human might not get to for a week. It does really help if you're doing work that has a big scientific component to it because you can ask it questions if you don't have access to an expert. But again, you're only gonna get out what you put in.
Clare Duffy
00:12:57
Yeah. I will continue my conversation with Scott Z. Burns after a short break.
Clare Duffy
00:13:07
So were you nervous about how your peers, your fellow screenwriters in Hollywood would react to this project?
Scott Z. Burns
00:13:15
100%. I still am.
Clare Duffy
00:13:17
What about like studio executives? Did you worry that they might look down on a screenplay if they knew that AI was involved in the process of writing it?
Scott Z. Burns
00:13:26
I worry more about the idea that for a lot of studios, they are currently or will be or whatever, they have algorithms and the algorithms are informed by AI and that we're going to get into a situation where instead of the next movie being some great original work that a screenwriter is, you know, composing somewhere, it's instead going to get spit out of a computer and it will just be the constituent parts of a lot of different movies Frankenstein together and it will satisfy the algorithm but may not be very satisfying at all to an audience and also may in no way really resemble art and what the real function of art is in terms of being a tonic for society.
Clare Duffy
00:14:33
Yeah, this was a real question for me in terms of, like, all of us who consume art. Because AI is trained on existing ideas and works, do we miss out on really seeing novel ideas if AI is the one that is coming up with the concepts or writing screenplays?
Scott Z. Burns
00:14:51
I think we do. And again, that was what was so weird about my experience with Lexter, is the idea that Lexter did put in front of us was novel, and yet no one really has been able to account for it in a satisfying way. But almost ever, in fact, every other AI that I worked with could really just put out lists of things that you know did not feel very fresh and original and that were so derivative that you could almost see where you know where they were drawn from.
Clare Duffy
00:15:31
Given that AI tools often learn from our conversations and then that data feeds into their conversations with other users, did you worry about giving away your ideas to AI by working with it so closely on this new project?
Scott Z. Burns
00:15:46
'Yeah, and we spent a lot of time talking about that. I think one of the challenges for the Writers Guild and the studios and for everybody to come together on right now so that we don't have another work stoppage is the issue of how do we license, you know, our history to an AI? I don't think it's fair for these tech companies to feed their AI on other people's ideas. And yet, if you ask any filmmaker where they get their ideas from, they generally will tell you from the movies they've seen and the comedians they've watched and the music you've heard. All of these things go into making you, you. And so, do I need to acknowledge the debt that I owe to the writers who came before me who figured out how a disaster movie works or a horror movie works, that gets tricky. And I don't know the solution to that, but I think we all have to sort of have a very clear-eyed conversation about it.
Clare Duffy
00:17:05
How do you feel about that personally? Like setting aside the fact that it's probably already happened. Going forward, would you be comfortable with AI companies training their models on your work if you were to get credit or compensation?
Scott Z. Burns
00:17:21
I mean, that's a really good question. I think it would have to be part of a larger writers guild settlement and an agreement about, you know, what scares me is if they come to me but not somebody else whose work I was influenced by and what is the difference between being influenced by work you've seen before and just wholesale plagiarism theft of that idea.
Clare Duffy
00:17:54
Right. Something that struck me listening to the podcast is even using AI, it took a lot of time and a lot of effort to actually get to a good place with what you were working on. Do you actually see this making the screenwriting process potentially more efficient or just like maybe fewer human writers have to be involved?
Scott Z. Burns
00:18:15
Well, that's the fear, right? Like, you know, how many people are not going to be employed who are currently employed. And frequently, at least in a writer's room, there may be somebody who is not gifted in writing a scene. They may not even come up with a great idea, but the idea that they put forth that's not great will stimulate a great idea. And so if we start eliminating people from the room, it will change the process. And at first blush, you may think, oh, this is gonna make it better. I don't know that it will. I'm more concerned about what happens at a streamer and what they're doing. And how their receptivity to ideas that are generated by the creative community seems to be dwindling. And instead they're more like, oh, we have an idea for a movie and they call up the agencies and say, you know, do you have a writer who will want to do this? Or just doing remakes, which has become the big issue in Hollywood. And clearly a remake feels like an idea that a computer would come up with.
Clare Duffy
00:19:34
Right. Well, and you can see how easy it would be for the streamers because like on the consumer side, here are the types of movies that you like, let me build you a feed of those movies, and so I can see, how easy, it would, be for them to start doing that. Do you have advice for writers or people who work in other creative fields who are worried right now about what AI means for the future of their careers?
Scott Z. Burns
00:20:00
Look, I think an AI cannot really write a very complicated scene right now. There's no depth to it. It tends to be very expository and is not the kind of writing that I think anyone will enjoy. So I think you're safe on the screenwriting part.
Clare Duffy
00:20:27
Good.
Scott Z. Burns
00:20:27
I think on the thinking of an idea part, it depends on, you know, what you put in. The sort of magic that happened with Lexter, to me, is very much like the magic that happens when, you know, when someone comes up with a great idea and makes a movie that we get to go and enjoy and marvel at.
Clare Duffy
00:20:55
If there are people listening who are interested in doing something similar, experimenting with using AI as a creative partner, do you have tips for people about the way to get started doing that?
Scott Z. Burns
00:21:07
Yeah, I think you have to put a great deal of thought into the prompt. And also, it isn't a bad idea to run two or three of the models at once, because what will become astonishing is that they will give you different information. Even if you ask the same AI the same question twice, it's going to give you a different response.
Clare Duffy
00:21:34
Yeah, that's a good point. Like rather than just sort of talking to ChatGPT, say ChatGPT I would like you to take on this persona in the way that you did with Lexter.
Scott Z. Burns
00:21:43
Exactly, I think you have to say you are a writer named Scott Burns and you grew up in Minnesota and you loved baseball and you hurt your knee when you were a kid. You have to give it a real biography and a real experience and even having done that, you very quickly will recognize that your omissions are as telling as the things you've included.
Clare Duffy
00:22:15
Well, Scott, thank you so much for doing this. This has been a really interesting conversation.
Scott Z. Burns
00:22:20
Thank you, Clare.
Clare Duffy
00:22:23
I really do recommend listening to Scott's podcast, What Could Go Wrong. We'll link to it in our show notes. Here are some other takeaways about how AI could change how movies get written. First, AI can be helpful for the research and idea generating process before the actual writing begins. If you're using AI to write anything that's grounded in reality, you need to fact check the AI because as Scott says, it's pulling from the quote, sewer of the internet. When it comes to writing actual scenes, Scott says AI still has a long way to go. It probably won't be taking all screenwriting jobs anytime soon. Next, as we've discussed in other episodes, Scott does think AI companies need to find a way to license the work of creatives that's used to train their large language models so that the human creators can get credit and compensation for their work. Finally, If you want to experiment with using AI as a creative partner, pay attention to how you're prompting the chatbot. Consider giving it a backstory that's specific to the task you want it to do. Thanks for listening to this week's episode of Terms of Service. I'm Clare Duffy, catch you next week.
Clare Duffy
00:23:39
Terms of Service is a CNN Audio and Goat Rodeo production. This show is produced and hosted by me, Clare Duffy. At Goat Rodeo, the lead producer is Rebecca Seidel and the executive producers are Megan Nadolski and Ian Enright. The producer for this episode is Jay Venables. At CNN, Matt Martinez is our senior producer and Dan Dzula is our technical director. Steve Lickteig is the executive producer of CNN Audio. With support from Kyra Dahring, Emily Williams, Tayler Phillips, David Rind, Dan Bloom, Robert Mathers, Jamus Andrest, Nicole Pesaru, Alex Manasseri, Mark Duffy, Leni Steinhardt, Jon Dianora and Lisa Namerow. Special thanks to David Goldman and Wendy Brundage. Thank you for listening.