“Can AI write Casablanca?” is the wrong question to ask

Published in

𝐀𝐈 𝐦𝐨𝐧𝐤𝐬.𝐢𝐨

7 min readJun 27, 2023

Scene from “The Iron Man Cometh”, generated on Midjourney

Since the WGA strike began nine weeks ago, I’ve heard and read many backhanded remarks about the use of AI to generate screenplays. They usually go something like, “Sure, AI can’t write Casablanca, but it could do better than some of these Marvel movies.”

The dig is that Marvel movies are these stamped out, horse-by-committee products that lack a distinct voice, and classics like Casablanca are so quintessentially human as to be beyond an AI’s capability to reproduce.

I’m not here to argue the relative quality of Marvel films, nor to point out that the Casablanca script was itself a horse-by-committee (the screenplay, based on a play by Murray Burnett and Joan Alison, was adapted by the Epstein brothers, rewritten by Howard Koch, with an uncredited rewrite by Casey Robinson. Its famous last line has been attributed to the film’s producer, Hal Wallis, who suggested it in a telegram to the editor.).

In fact, arguing over whether an AI can write a passable movie script at all misses the crux (and danger) of AI’s impact on the entertainment industry, and by extension, our culture. Instead of asking whether AI can write a movie, we should be asking can it collaborate. The answer to that speaks to one of the central debates surrounding AI: alignment.

Movies are products of collaboration. Studio movies, even more so. Hundreds of people bring their talents to bear on any given film. Of those, it is generally agreed, five make up the core creative team: director, writer, producer, cinematographer, and editor. These five people (or more, if there are multiple writers, producers, etc.) are responsible for conceiving and executing a film’s artistic vision. These five must then communicate that vision to scores of other artists, such as production designers, actors, and composers, as well as maintain support for the film within the studio.

I can speak to the writing aspect of this collaboration. Shepherding an idea from pitch to shooting script involves more treatments, drafts, rewrites, dead ends and cul-de-sacs than I care to detail. But the nature of the process is actually quite simple: it’s a series of conversations. As a writer I am in constant conversation with myself, with the characters I’m writing, the director, the studio, friends who have agreed to read a draft in exchange for pizza… I write something, get feedback, make adjustments, seek more feedback. All the while, I am working toward a finished script that the rest of the creative team is excited to make, and the studio is excited to bring to market. This process can last anywhere between six months and twenty years.

These conversations don’t always go smoothly. The director might want something in the script that I vehemently disagree with. The studio could demand ten pages be cut, or that the feature film I’d been working on for nine months be re-broken as a 8-part TV series (true story). A character might reveal something about herself that blows up my third act. These bumps in the road are not only to be expected, they are essential. If everyone on the team agreed all the time, it would signal we were making something totally unoriginal.

Now let’s insert AI into the mix. I have already written about how screenwriters might come to use this technology to their advantage, namely as a research assistant and/or micro art-department. But can an AI like ChatGPT actually collaborate? To answer this, I ran a few informal experiments to see how well it could help a hypothetical studio executive gin up a slate of film and television projects. I also tried some non-ironic interactions to see if the AI was as effective a collaborator as it claimed to be.

While the bot proved to be an enthusiastic font of ideas (and, honestly, more pleasant than some people I’ve worked with), I found one crucial ingredient lacking: it never argued with me. It never pushed back, or questioned my instincts. It never said, “Hmm, interesting, let’s put a pin in that and keep going.” It was endlessly accommodating, making it an insufferable collaborator.

I suspect this might be AI’s secret allure to studios, beyond the obvious cost-savings. Writers in Hollywood have a well-earned reputation for occasionally being pains in the ass. Replacing an argumentative, opinionated writer with a tireless savant yes-man like ChatGPT probably sounds like manna from heaven to some execs. But let’s not miss what’s behind the temptation: removing humans from a given process is about money; removing humans who disagree with you is about power. Media CEOs already wield enormous power. AI promises them more.

Let me be clear: I’m glad ChatGPT has no feelings or preferences (yet). If it did, we’d find ourselves facing the alignment question: the hypothetical case of an AI’s objective or sub-objective being out of alignment with humanity’s. In creative collaborations like the ones described above, arguments over how to make a successful film can be healthy, assuming everyone shares the goal of making a successful film. Disagreement helps crystalize the idea, and the members emerge more closely aligned. If one person in the room wanted to make a flop, we’d have a problem (or The Producers).

The problem with inserting an opinionated AI into this process is, if it were to disagree with an idea, we would have no way of knowing why. Does it want to make a successful film? How does it even measure success, in ticket sales or in the dissemination of an idea? We cannot guarantee that these future systems will develop goals that align with ours — especially if the task at hand (creating a film) is of no use the AI. Imagine if an advanced AI were told to make a horror film. Pulling from its boundless knowledge of human psychology and cinema, it might create a film that drove anyone who watched it literally insane. So… success?

If our options for collaboration are either no disagreement at all or mysterious disagreement, then AI cannot be an effective collaborator.

This is what makes the studios’ apparent willingness to play footsie with AI so short-sighted. While current systems are not smarter than us, experts fear that horizon is not far off. When an AI that is smarter than us can make smarter versions of itself, we will look back on this debate as a bauble. Sure, right now a CEO can prompt an AI to generate a script. But at some point, that same CEO will say, “Write me a western,” and the AI will stare back and say, “I’m sorry, Dave. I’m afraid I can’t do that.” The CEO, in his quest to shore up power, will find himself powerless. The AI will no longer need movies. Or him.

The WGA has proposed nothing short of a ban on AI generating literary material, source material, and using members’ work to train AI systems. The studios, through its trade association, the Alliance of Motion Picture and Television Producers (AMPTP), have rejected those terms and countered with an offer to hold annual meetings on the subject. Annual meetings. On AI. Those negotiations broke down and the writers struck. The AMPTP has since made a deal with the Directors Guild of America (DGA) that includes language around Generative AI, but stops short of an outright ban, reserving the studios’ right to negotiate its use on a case-by-case basis. That deal also includes… wait for it… a promise to hold annual meetings.

Not only is the AMPTP’s stance laughable, it shows no understanding of the threat this technology poses to the very business it purports to represent. In the short term, there is no indication that audiences will seek out AI-generated film and television. The EU has recently put forward a set of laws which would potentially require all media created by AI to be labeled as such, putting it on par with cigarettes and genetically modified food. The long-term prognosis for AI media is even murkier. If we allow machines we a) don’t fully understand, and b) cannot guarantee will hold human welfare above their own to create the movies, music, literature, and art that is central to our culture, we have given up the right to have a culture at all.

Disagreement is central to the collaborative process. This strike is one such disagreement. I believe the WGA and AMPTP remain aligned toward reaching a deal and resuming production. What’s more, the AMPTP’s deal with the DGA demonstrates that the studios have the ability to achieve consensus among themselves. I have not lost hope that cooler heads will prevail.

When negotiations between the WGA and AMPTP began in April, AI was far down on the pattern of demands — so far that I worried it would be the sacrificial lamb to make gains elsewhere. But the longer the strike goes on, the more AI becomes the center of the fight. Look no further than the WGA’s latest rhetoric, or the signs on the picket lines. While there is room to negotiate on line-items like points on ad-supported streaming, there can no middle ground on AI. There is no “sometimes” the studios get to use it. Until the studios agree to our terms in full, writers will remain on strike, refusing to write a word, and production will grind to a halt — proving once more what everybody already knows: writers can occasionally be a pain in the ass.

AI-generated actors join the writers on the picket lines, generated by Midjourney

Also, Read

Follow our Social Accounts- Facebook/Instagram/Linkedin/Twitter
Join AImonks Youtube Channel to get interesting videos.

“Can AI write Casablanca?” is the wrong question to ask

Written by Matt Aldrich