Back to selection
The Gotham Pages: AI and the Usefulness Question
Amid the perpetual torrent of news about rising investment in artificial intelligence tools within the entertainment industry, one recent item garnered particular attention online. Included in Lila Shapiro’s sweeping look at the current state of AI usage in Hollywood for Vulture were snippets about Asteria, a new studio co-founded by entrepreneur/producer Bryn Mooser and actor/director/producer Natasha Lyonne. Lyonne, who is also a three-time Gotham Award nominee, has even announced that she plans to use AI in the production of her debut feature. The news that the popular actor had thrown her weight behind the controversial technology caused consternation on social media, characteristic of the negative response viewers often have to AI in film or television. Similar backlashes occurred when True Detective: Night Country used AI-generated band posters and Late Night with the Devil used AI to create 70s-style television interstitial cards.
These anxieties are understandable. In a precarious climate for the industry, AI is seen as an unprecedented threat to the job security of artists in nearly every discipline, from production design to animation to editing. On the audience side, many experience an instinctual revulsion when encountering the often-uncanny images AI tends to create. But AI also has its boosters. Mooser insists that this is the wave of the future, and Lyonne is adamant that Asteria will be an “ethical” purveyor of the tech. In conjunction with parent company Moonvalley, Asteria has developed the image- and video-generation tool Marey, which they claim has been trained entirely on material shot in-house, licensed from media companies either directly or via third-party data brokers, or in the public domain. All of this raises a question: What uses are there for AI in production that don’t threaten anyone’s job, aren’t based on scraping copyrighted material, and actually look good?
In the Vulture article, Mooser avers to Shapiro that “The story of Hollywood is the story of technology,” comparing Asteria to the innovations of Walt Disney and George Lucas. While this is true, the sentiment belies that for every technological revolution in filmmaking, there are just as many (possibly more) technological dead ends. This history is littered with hyped developments that went nowhere—like the theatergoing gimmicks and unusual shooting formats of the 1950s—or that surged before petering out, like the multiple 3D fads. In fact, Mooser himself was involved with an earlier such trend, having co-founded the immersive media company RYOT in 2012. While there have been intriguing virtual reality and augmented reality projects, the format has not lived up to the grander promises that surrounded it more than a decade ago. Instead, streaming, social media, and short-form video overtook the entertainment landscape. Meanwhile, a great deal of the oxygen and venture capital that once went to VR has been redirected toward AI.
Writing about artificial intelligence is difficult because the subject is such a moving target. Advocates insist that the technology will continue to get more complex, smoothing out the hallucinations and uncanniness that plague the output of generative models, while skeptics believe we might already be hitting the limits of their sophistication. At the time of writing, the biggest story in the field concerns the rollout of the newest ChatGPT model, with complaints about everything from workflows being broken to chatbot “personalities” being destroyed by the changes. By the time of publication, some other story may have supplanted it. Concerns persist about the technology’s gargantuan energy and water usage. Additionally, the whole concept of using AI in creative work has a Sword of Damocles hanging over it in the form of high-profile legal actions concerning the unlicensed use of copyrighted material to train these programs, from Disney and Universal taking on Midjourney to a massive class-action lawsuit against Anthropic AI. To date, AI companies have worked mostly ahead of regulation, but anything catching up to inhibit them could dramatically upend the utility of their tools in production.
Asteria is theoretically ahead of this curve, thanks to Marey’s “ethical” training. In various interviews and profiles, representatives from Moonvalley and Asteria have been cagey about the specifics of this process—Moonvalley did not share “details about where and how exactly the company had managed to pay for and acquire a sufficient trove of data” with Shapiro, calling such information confidential. (At the time of writing, neither Moonvalley nor Asteria granted an interview or comment to The Gotham Pages.) In a later interview with TIME, CEO Naeem Talukdar said 80 percent of the training footage came from B-roll owned by the licensees. More recently, CAA, which maintains a digital database of clients’ likenesses to protect them from unlicensed use, became a backer of Moonvalley. If all training materials for Marey are indeed aboveboard, this spares any filmmakers who use its tools from any fear of legal reproach.
The current fervor makes it easy to forget that this technology isn’t new to the industry. The distinction is that AI tools are much more accessible and prevalent now; before, this was mainly an option only for big-budget films. One prominent early example is the de-aging effect in 2008’s The Curious Case of Benjamin Button, which used machine learning as part of the process to turn Brad Pitt into both an old and young man without the use of makeup. But this example is also instructive in how the tenor of the conversation has changed. Looking back at contemporary coverage of Benjamin Button’s special effects, like articles in Wired and MIT Technology Review, the phrase “AI” isn’t used at all, as they instead focus on the innovations in performance capture and facial replication. Today, the film’s VFX executive producer, Ed Ulbrich, eagerly highlights the AI aspects in an interview with Deadline about how he’s recently joined Moonvalley as head of strategic growth and partnerships.
Complicating matters is the way “artificial intelligence” is understood and discussed, both by the public and the industry. It’s not a single technology but an conceptual umbrella for multiple discrete ones, thanks to disparate companies hoping to profit from surging interest in the field. University of Washington linguistics professor Emily M. Bender argues that it’s more of a marketing term for automation than anything else. An AI tool might be a machine learning model like DALL-E. Or it might be dressing up preexisting tech as something else, like Cluely, the AI assistant that supposedly lets users “cheat on everything” but which apparently just pings ChatGPT for answers to queries. Or it might be a Potemkin setup built on underpaid human labor, like the fraudulent e-commerce site Nate, which secretly used remote workers in the Philippines to fill in customer information, which was supposedly done with AI.
The prevalence of black boxes throughout this subject poses an obstacle to discussing how these tools are used in filmmaking. People understand “AI” as one all-encompassing entity. This is to the
advantage of businesses hoping to cash in on the trend by slapping the label on their services, but it muddies the waters for anyone assessing how AI is being used in filmmaking. One of the biggest recent AI-related film controversies concerned the use of Respeecher to tweak Hungarian-language dialogue in The Brutalist. But Respeecher is fairly tight-lipped on how exactly AI is used in its voice cloning process, making it hard to judge whether this was really an artistic “cheat,” as some claimed. Many seemed to reflexively balk at the presence of any AI at all.
Such an attitude contributes to a climate that stifles open discussion of this subject. Public outcry has been effective at shaming nearly any use of AI in film, to the point that workers on the recent film The Legend of Ochi had to clarify that they had not done so in response to social media rumors around its special effects. Despite how much Shapiro delves into the topic, her Vulture article lacks specificity on how, per its title, “everyone” in Hollywood is supposedly using AI. It cites mostly anonymous creatives confessing to using these tools for small tasks without saying too much about what they were. Some of this reticence can be attributed to fears of either professional reprisal for skirting labor rules concerning AI established in the wake of the recent strikes, or of legal consequences for any association with the tech, in light of the aforementioned lawsuits over intellectual property.
The grand rhetoric around AI—both to hype it and to fearmonger around it—can obfuscate its actual use cases. Looking at quotes from Moonvalley staff, there’s a lot of talk about how its tools could help filmmakers, but not many concrete examples of how it has. Mooser and others claim Marey can do things like provide alternate angles on test footage or turn concept sketches into full-fledged artwork. In an article for TechCrunch, director Ángel Manuel Soto tells the writer that Marey “helped him cut production costs by 20% to 40% and work more freely” but doesn’t explain how. Whether it can truly fulfill such roles—or, more important, whether it can meaningfully replace the workflows established in the industry—very much remains to be seen.
One tangible example in the Vulture article is a VFX artist on Everything Everywhere All at Once using an AI tool from Runway to remove greenscreen elements. When looking at confirmed uses of AI across disparate films and TV shows, a distinct pattern emerges. In Benjamin Button, EEAAO, and perhaps The Brutalist, AI was employed to help speed through some of the drudgery of effects work, make tweaks to smooth over little inconsistencies, and generally act effectively as an aid to human creativity rather than a replacement for it. Whenever AI is employed as a production shortcut, as in True Detective or Late Night with the Devil, it seems to quickly raise hackles in viewers who suss its uncanniness. Per Bender, it may be most helpful for filmmakers to think of AI through the lens of what can be conveniently automated in their work, rather than a miraculous panacea that can be resorted to in any circumstance. If the wilder promises made around AI come to pass and it actually obviates the need for writers, artists, and directors, then a new conversation will have to take place. But as much as some executives might hunger for a world in which they can bypass all these voices, betting on technocratic moonshots over humanity has never paid off in the long run.
