Hollywood’s Screenwriters Are Right to Fear AI

One of the more harrowing reads for writers concerned about artificial intelligence encroaching on their livelihoods is a study commissioned by OpenAI itself. Published in March, it places writers in the “fully exposed” category. This means that, according to OpenAI, a large language model (LLM) could reduce the time it takes for them to carry out their work by at least 50 percent. AI can already score in the 93rd percentile on SAT reading exams; it can already produce bad stories and poems. Directors are discussing the possibilities of AI-generated scripts.

It’s no wonder, then, that the Writers Guild of America is demanding a greater say in how AI is used in Hollywood.

When WGA members began their picket on May 2, the first such strike in 15 years, focus was on streaming services’ impact on Hollywood and how the residuals paid to writers for streaming projects hadn’t kept pace with those of traditional broadcast shows and theatrical releases. But there was another demand: that the agreement with studios “regulate use of material produced using artificial intelligence or similar technologies.”

The Writers Guild isn’t alone in trying to figure out AI’s place in its industry. Artists, actors, musicians—people in all creative professions—are trying to wrest control of the technology before it is used against them. It’s a smart move. If the history of automation has demonstrated anything, it’s that leaving the implementation of new technologies up to management is a bad idea.

Over the phone, John August, a member of the WGA’s negotiating committee and writer of Charlie's Angels, explains that the proposal “says that material generated by AI or similar technologies is not considered literary material or source material for the purposes of the contract.” In this context, literary material refers to the stuff writers get paid to write: screenplays, treatments, outlines. Provisions over source material, on the other hand, seek to ensure that writers will not be asked to adapt AI-generated scripts—trained on human writers’ work—as they might with a novel. In both of those scenarios, writers’ employment and pay shrinks. The Alliance of Motion Picture and Television Producers rejected this proposal, offering “annual meetings to discuss advancements in technology.”

With emerging tech, things rarely happen as advertised. This is particularly germane to LLMs, where the scope of possible ends still ranges from flash-in-the-pan productivity tools to society-upending sci-fi intelligence. Skeptics often bring up self-driving cars, baptized in hype a decade ago but still not flooding the roads or outmoding truck drivers.

Nonetheless, generative AI’s current trajectory conjures inescapable parallels with the encroachment of streaming services on the Hollywood system. August recalls that at the time of the 2007 WGA strike, the major streaming services hadn’t yet taken off. He had, however, begun to notice the stirrings of deep sea change, portended by innovations like spin-off “webisodes” of The Office being thrown onto the internet.

“We saw that this might be the next way that companies would be able to make money off of our work,” he says. “So that strike was largely about making sure that, no matter what the medium was, our content still was paid for upfront in a fair way and paid for in the back end through residuals. We didn’t know what streaming was going to become. But we knew the internet was going to be the future.”

There’s a long history of management painting automation as something as inevitable as sunrise. It’s an echoing pattern, one the late historian of technology David F. Noble summarized in Forces of Production, his account of the implementation of machine tools in America. “‘Automatic’ or ‘self-acting’ machinery made it possible for management both to eliminate workers altogether and to control more directly the production process,” he wrote. “The machinery, in turn, was used to discipline and pace the operators who attended it, thereby reducing the “labor problem” indirectly via the seeming requirements of the technology of production itself.”

Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity, a book from MIT economists Daron Acemoglu and Simon Johnson that’s due out next month, chronicles a thousand years of elites—from European nobles in the Middle Ages to modern-day tech CEOs—gaining from technological advancements at the expense of workers. Generative AI fits neatly into this historical context. “We argue that this obsession with machine intelligence is not helpful because it’s all about replacing people,” Johnson explains. “Whereas if you focus on making machines useful to people—nurses, doctors, teachers, and so on—that will be much more helpful to productivity and therefore, potentially, to pay.”

Futures range in awfulness. August’s personal dystopia is what he calls the Nora Ephron scenario, where AI learns to mimic cultural titans, eclipsing new human writers. Studios likely won’t employ AI scabs during this strike, not least because having AI tools cross the picket line introduces a host of copyright issues, but it’s not hard to imagine that this could happen at one point. (“You cannot protect studio execs from their bad ideas,” he says.)

And then there is the most likely bad scenario, the one worth getting out in front of right now: a producer requesting that a writer edit a script (which pays less than producing an original work) and not telling them it was generated by a chatbot. “That’s a crisis in our compensation, it’s a crisis in our residuals, and a crisis in our artistic ability to do the things we are put in this industry to do,” says August. “So that’s a fundamental nightmare scenario. And that feels very obvious if we don’t get this resolved.”

More positive outcomes include improved productivity, like moving from a typewriter to a word processor. Commentators are unsure, however, whether that increase in productivity will lead to tangible improvements, like an increased standard of living. ChatGPT is already useful for brainstorming: If you need 15 different names for a Mandarin bagel shop, as August puts it, AI does an alright job. And he sees a possibility that the tech could create opportunities for more diverse writers, improving the scripts of someone for whom English is not their first language, for instance.

Automation and redundancy are not necessarily conjoined, and introducing disruptive technology—like the self-checkout machine—is a choice. There are examples of times when worker perspectives on new technologies, not just those of management, have been successfully taken into account. In their book, Acemoglu and Johnson cite West Coast longshoremen who demanded to be retrained in new technology. They won, leading to a reduction in job losses and an increase in productivity. Katya Klinova, head of Al, labor, and the economy at the Partnership on AI, points to Unite Here, which represents hospitality workers, who in 2018 successfully won the right to negotiate how Marriott plans to bring in new technology, like online services, computers, and even robots.

Digital technologies are inherently isolating: They do not lead people into factories to discuss concerns with their fellow workers. The efforts of a union with the relative power of WGA trying to assert control over AI implementation are instructive for everyone. For the writers, it’s critical: Their contract is only up for negotiation every three years. That’s a long time in tech. “You know, in 2007, streaming wasn’t there yet. But by 2010, you started to see those inklings,” says August. “In 2023, AI is not replacing us—AI is not being used to write exactly what we’re doing. But by 2026, the next time this contract is up, it really feels like that technology will be very refined. We need to make sure that this is addressed.”