AI & Show Business: A Fake South Park Episode Sparks Controversy
Last week, San Francisco startup Fable Studio announced an ‘AI Showrunner’ capable of generating a full-length, fake South Park episode written, animated, directed, and voiced based on a single prompt. It sparked the necessary controversy.
It sparked controversy because we’re currently in the midst of one of the biggest strikes in showbiz history. A strike partly ignited by the fear of AI. The last thing you want to hear is that there’s this company claiming it has managed to outsource every meaningful part of creating a TV-show to a computer program. To add insult to injury, the fake episode they released featured Cartman who, after finding out about the Screen Actors Guild strike, pitches a new business idea to his friends: a deep fake streaming service called Queepi.
A bit of a slap in the face, if you ask me.
Having said that, it is an incredible feat. The opening scene stars Cartman, Kenny, Stan, and Kyle. Although their voices lack liveliness, they match the original characters and talk the way they talk. Cartman is being his boastful self, Kenny is mumbling. The storyline is mildly entertaining but surprisingly coherent, and features celebrity cameos like Meryl Streep and Elon Musk.
So how did the researchers pull this off? And is this the final nail in Hollywood’s coffin?
The power of generative agents
Let me start by saying that none of this was possible without the arrival of generative AI and in particular something called ‘generative agents’.
Beginning of April this year, Stanford and Google researchers created an AI town called Smallville to see what if they could get a number of LLM-powered agents to produce “believable simulacra of human behavior”.
And producing believable simulacra they did. They would talk, go for coffee, take a walk in the park and participate in all sorts of other innocent activities with each other — all powered through a novel architecture made up of 25 instances of ChatGPT ran in parallel, allowing the agents to “to remember, retrieve, reflect, interact with other agents, and plan through dynamically evolving circumstances”. They effectively ran a simulation.
Fable’s AI Showrunner can be viewed as an iteration on the same concept. Instead of a running a virtual town, the system is grounded in an existing TV show leveraging all of its imagery and existing characters including their backstories, personality traits, voices, et cetera. This grounding principle provides the system with a constant contextual feedback loop, which is what makes it so powerful.
The intention isn’t to plagarize. The ability to create your own South Park episodes wasn’t made publicly available by the researchers; besides the fact that it would get them sued, the ambitions of the company lie elsewhere.
In an interview with VentureBeat, CEO Edward Saatchi says:
“Imagine a world where fans can put themselves in their favorite shows, create new episodes and compete to create the best episodes ever made or a world where creators can use the AI Showrunner to make their own original shows.”
He sees a future where TV shows are uniquely created by you, for you and with you.
“We used South Park only so that people have a comparison point between a super high-quality human show and an AI show. We will be announcing several original IP Simulations with attached AI TV shows later this year — a space exploration sci-fi simulation, a satire of Silicon Valley simulation and a detective simulation.”
The dramatic fingerprint of a show
It’s a bold vision, I must admit. But having reviewed the paper more closely, I also feel like there are some caveats worth pointing out. Not to criticize or cast doubt, but because it’s good to put things in perspective.
All things considered, it really looks like the Fable’s AI Showrunner is powerful enough create a full episode with as little as a single-sentence prompt.
The paper covers this process in quite some detail:
“In order to generate a full south park episode, we prompt the story system with a high level idea, usually in the form of a title, synopsis and major events we want to see happen over the course of 1 week in simulation time (=roughly 3 hours of play time).
From this, the story system automatically extrapolates up to 14 scenes by making use of simulation data as part of a prompt chain. The showrunner system takes care of casting the characters for each scene and how the story should progress through a plot pattern. Each scene is associated with a plot letter (e.g. A, B, C) which is then used by the showrunner to alternate between different character groups and follow their individual storylines over the course of an episode to keep the user engaged.”
What remains unclear, however, is if the researchers selected the best out of hundreds of terrible episode generations or if the model generates high-quality output every time.
Another important footnote is the level of customization that is required. Every TV show is different. It’s something the researchers themselves describe as ‘the dramatic fingerprint of a show’ (it’s basically what makes South Park South Park) and it’s used to train the model on. Because every show is unique, every show would require its own AI Showrunner. And how much time is required to set this up and fine tune it to the degree that it can produce high-quality output remains an open question.
Also, when you think about, South Park lends itself extremely well; in many ways it’s the perfect show to be automated due to a highly predictable format and its simplistic animation style.
More complex TV shows will most likely pose new challenges along the way and TV shows with real actors seem wholly unfeasible for now.
Dreams of grandiosity
I do see how the showcased technology could pose a threat to writers, directors, and voice actors of animated TV shows. The quality might not be there yet, but it’s getting close. To pose a real threat, however, the generated output would have to be either on par with human-created content or better, since no one is going to watch a crappier version of the same TV show they are already watching.
Meanwhile, Fable Studio isn’t concerned with any of the disruption it might cause. They dream of different dreams.
It has high hopes of creating a new genre of content entirely: a platform that allows you to be the director, main character, or extra in your own show or that of someone else. ‘The Simulation’, they call it. They launched an eponymous website, including waitlist.
On homepage it reads:
At Simulation Inc, we're redefining the contours of existence, conjuring a universe where the line between the physical and the virtual blurs into oblivion. Our mission, as audacious as it is intriguing, is to birth a new kind of life: the world's first genuinely intelligent AI virtual beings. Each one, a mirror of the human psyche, navigating the tumultuous seas of emotions and experiences in a digital cosmos of our creation.
If you think that all sounds rather grandiose, wait until you hear this. Simulation Inc. isn’t a registered company and the founding team is completely fake: the portraits on the website aren’t just AI-generated, they aren’t real people to begin with. The company history is fabricated, too.
It’s all part of the concept, explains CEO Saatchi, when asked about it by a TechCrunch reporter. It gets even weirder and more grandiose, when he adds that the company’s ultimate goal is not to change Hollywood, it’s about achieving AGI. Forever transforming the way we consume film and TV as human beings is just a byproduct of that — an inevitable side effect.
AGI, short for artificial general intelligence, is every Silicon Valley founder’s favorite three letter-acronym nowadays, it’s all they ever talk about and search for: the modern holy grail.
If you ask me, someone has watched a little bit too much Westworld, but who am I to judge? It’s not a crime to dream big.
Jurgen Gravestein is a writer, consultant, and conversation designer. Five years ago, he stumbled into the world of chatbots and voice assistants. He was employee no. 1 at Conversation Design Institute and now works for the strategy and delivery branch CDI Services to help companies drive more value with conversational AI.
Reach out if you’d like him as a guest on your panel or podcast.
Appreciate the content? Leave a like, comment, or share with a friend or colleague!
Great piece. I'm also glad you reminded me about Smallville. I've had that in my post queue for months but it was delayed and deserves more attention. I'm glad you gave it a nod. A very interesting and important piece of work.
As to Fable, I've interviewed Edward for the podcast and have interacted with him a few times in the past two years. He is intelligent, creative, and unafraid of bold undertakings. He is the type of person that could make this work.
Granted, Edward and I disagree about the achievability of AGI, which I believe includes, at minimum, humanlike reasoning, agency, emergent abilities, and a sense of wisdom (even if an Alien wisdom) that goes beyond mere intelligence. No need to discuss consciousness which is even farther afield. The listed elements are not strictly computational challenges which provides a tall bar to clear for computational systems. Of course, everyone keeps changing the definition of AGI depending on whether they want to invoke fear, inspiration, or set an achievable milestone for their VC pitch.
I see AI Showrunner as one more copilot for the professional class. While I expect it to require more than a single prompt to create high-quality output, it certainly can help enhance efficiency and creativity. If it works, it will also reduce one more barrier to content creation which I generally see as a good thing.
The striking screenwriters using a tool like this may now be able to create their own shows without reliance on the studio system. That could be a very positive outcome.
I tend to agree: this is much ado about very little, at least for now. I can certainly see businesses onramping this as a service to folks for a fee, but it'll probably start with a very small user base, and might just be nonviable for a few more years. It'll get there, though.