Many comedy writers view AI as no laughing matter. “It’s horrific,” Seth Rogen explained to The Hollywood Reporter at the May 10 premiere of Apple TV+’s show Platonic, in which he stars. “Any use of AI seems terrifying and also just unfair from a financial standpoint because it’s all being input with things that they’re not keeping track of,” referring to how the technology is trained on material without its creators’ consent.
The wariness has intensified during the ongoing writers strike. A key sticking point in the broken-down contract negotiations between the WGA and the AMPTP, which bargains on behalf of studios, is a proposal to regulate the use of artificial intelligence. Machines wouldn’t be allowed to write or rewrite literary material, or to be used as a source, and union-covered output couldn’t be used to train AI models, either. The studios have rejected that plan, instead only offering to meet once a year to discuss issues presented by the insurgent technology.
Writers Guild members see AI as not just a looming threat but a clear and present danger, believing the business landscape will have inexorably changed by the next negotiation cycle in several years. “ChatGPT Doesn’t Have Childhood Trauma,” read one sign at Radford Studio Center in the San Fernando Valley. “A robot can’t feel shame,” explained Harris Mayersohn (Tha God’s Honest Truth With Charlamagne Tha God) on the first day of protesting outside of Netflix’s Hollywood headquarters. “It’s essential to being human and essential to being a writer.”
For the most part, scientists in the AI research field known as computational humor think the immediate worry is overblown. These experts, some of whom have been studying the questions surrounding funny robots for decades, observe that large language models can be taught to whip up passable formulaic material because the propagation of hack jokes relies on systematized pattern-recognition processes. However, they believe that original, path-breaking comedy will likely remain out of the conceptual reach of such machinery, at least in the near term, because it’s such a singularly complex, quicksilver language. “Humor is highly contextual and situational, which makes it an extraordinarily difficult problem to solve,” explains Georgia Tech’s Mark Riedl.
These specialists contend that AI appears to face two essential challenges in creating comedy of high caliber. (While newly released image and music generators have wowed users with outputs that frequently go viral, the latest wave of chatbots have so far yielded no equivalent laugh-out-loud watershed moment.) While it has access to unfathomable knowledge, it can only leverage it to approximate life experience. At a fundamental level, it may always struggle to comprehend humans. “Unless the machine understands why a joke is funny, you are nowhere,” says Julia Rayz at Purdue. DARPA, the Department of Defense agency which helped pioneer the Internet, could help remedy this. It’s now trying to teach computers cross-cultural communications by developing natural language processing technologies.
The other problem is that AI, which synthesizes existing data sets, would seem to be at a disadvantage if the goal is to generate edgy, boundary-pushing outputs: innovative gags, rebellious conceits, unpredictable tonal decisions. “AI, a conservative technology, doesn’t understand what taboos are, so it can’t break them,” notes Cornell’s Guy Hoffman.
Not everyone is a naysayer. Tony Veale at University College Dublin, who explains that being (purposefully) funny requires the cognitive ability to understand and attribute mental states to oneself and others — what’s known as theory of mind — thinks it’s possible to evolve humor in something like the ChatGPT model by programming it to favor the incongruities and deviations from established norms that are the hallmarks of comedy. “A language model uses probability to guide its choices,” says Veale, the author of Your Wit Is My Command: Building AIs With a Sense of Humor. “You could adjust its probability controls, from the expected to the unexpected, since the dominant theories of humor have to do with taking ideas and subverting them.”
Joe Toplyn, a former head writer for both Jay Leno and David Letterman, has already made his own punchline-producing bot, Witscript, based on joke algorithms he’d earlier developed, patented and taught at the Peoples Improv Theatre in Manhattan. “When it’s working at its best, it’s writing jokes that are good enough to be used on a late night comedy talk show, without any editing,” he says. Toplyn, a WGA member with a Harvard engineering degree who recently picketed outside HBO’s Manhattan office while holding a sign that read “Don’t Let ChatGPT Write Yellowstone,” insists the future is already here. He explains he developed his beta software, which he demonstrated to THR, to assist non-pros who can’t write jokes, but thinks that if producers had access to it, “they could cut costs” on writers. (As part of a February bit, Jimmy Kimmel tried out AI-generated jokes, one of which earned a smattering of genuine guffaws from his audience.)
The commercial question, as ever, is quality. If brilliance is the business plan, then the human humorist is secure. But if not, then AI, even if it only achieves a bare minimum of hilarity, may do just fine. “How mediocre are you OK with your comedy being?” muses Piotr Mirowski, a scientist employed by a machine-learning firm who also co-founded an AI-enabled improv company that incorporates a chatbot into its performances. “It’s on us to judge the outputs of AI with discernment.”
The consensus among these experts is that AI, while a clear labor threat, will become a baseline tool for comedy writers, like a thesaurus or a search engine. Diyi Yang at Stanford notes that, in her own research, professional stand-ups may not have found AI-generated jokes funny. Yet they often were inspired by the model’s strange, unconventional verbal associations — this frisson being akin to utilizing the musician Brian Eno’s Oblique Strategies cards, which are meant to break creative blocks and foster unconventional thinking.
As Columbia’s Tuhin Chakrabarty puts it of AI and humor, “there always has to be humans in the loop.” Adds Simon Colton at Queen Mary University of London: “It’ll still make commercial sense to hire a professional. Good companies will realize it’s a multiplier.”
For his part, Jeff Schaffer, the noted comedy writer (Curb Your Enthusiasm) and creator (Dave), professes not to be concerned with AI’s rise. “I don’t know a lot about AI, but I know a little about people,” he says. “So far, AI is not funny. And that is its most human quality — because most people are not funny, either. And the more AI becomes like people, the more it will think it’s funny. And so if it becomes exactly like people: The AI will think it’s hilarious, and everyone else will be listening to it or reading it and saying, ‘The guy thinks he’s so funny — but he’s such a bore.’”
Jackie Strause and Kirsten Chuba contributed to this report.
A version of this story first appeared in the May 31 issue of The Hollywood Reporter magazine. Click here to subscribe.
