Tinseltown isn’t known as the most ethical industry. Back when I was in film school, my hardboiled professor wanted us to know just what we were getting into. (Hollywood had apparently chewed him up and spit him out.) Our assignment was to read the book What Makes Sammy Run by Budd Schulberg.
Set in the 1930s, it tells the tale of Sammy Glick, a ruthless young man from New York who comes to the west coast with big dreams. Driven to succeed, he doesn’t care who he steps over to get to the top of the heap. Possessing bottomless ambition and zero scruples, Sammy manipulates his way into the screenwriting world. His duplicity isn’t a hindrance. It’s a virtue amongst the cutthroat. Betraying mentors, lovers, even friends to get what he wants, he becomes a vaunted and feared figure in the film industry.
Nearly 100 years after the fictional events of this novel took place, a new ethical quandary is again rattling the Hollywood power structure: AI. Screenwriters are once more being squeezed by powerful market forces. As the Los Angeles Times reported in February: “John Rogers, a 58-year-old screenwriter in L.A., has spent years co-creating the world of TV drama series Leverage. After experimenting with ChatGPT, Rogers said he and the show’s creative team suspected that 77 episodes of the series—or five years’ worth of work—had been ripped off and used to fuel AI.”
This prompts a question—one that wouldn’t have kept Sammy Glick up at night: What happens when the tools we use to create threaten the very creativity they’re meant to serve? Clearly, generative AI has erupted onto the scene like some antihero straight out of central casting—equal parts hero and threat. For the moment, studios are cautiously optimistic. As for writers, directors, and other creatives? They’re wary. Very wary.
This is because in the breathless rush to adopt AI into more and more business applications, many people have looked the other way as ChatGPT and other models scrape large swathes of copyrighted content for training. As MSN explains, “These generative artificial intelligence platforms have hoovered up 19th century novels, beat poetry, draft contracts, movie scripts, photo essays, millions of songs and everything in between on the way to becoming the most disruptive technological force since the invention of the internet.”
All that hoovering led a man named Naeem Talukdar to ask: “What kind of future are we building when our tools are trained on stolen voices?” Co-Founder and CEO of Moonvalley, the startup he leads is an AI-powered film platform “with generative models trained on fully licensed content and precise, intuitive tools for creators.” Privately funded, it contains a team of “best-in-class engineers, researchers, product designers, and filmmakers.”
I recently sat down with Talukdar to better understand how his vision compares with other generative AI platforms unconcerned with the ethical implications of wholesale data lifting to complete in a Hollywood increasingly powered by artificial intelligence. “At the heart of the generative AI debate lies a truth most tech companies would rather ignore,” says Talukdar. “Many of today’s most powerful models are built on ‘dirty data.’”
This content is scraped indiscriminately from the internet—YouTube clips, entire film libraries, digital art portfolios, and more. Most are acquired without consent or compensation. And though this business practice has led to lawsuits from authors, artists, and entertainment guilds, it remains to be seen what a glacially-moving judicial system can do in the face of a technology blasting ahead at hypersonic speed. Coupled with this concern is what invariably goes unsaid outside writers’ room throughout Hollywood: Sure, all this may be questionable in the long-run. But for now? It works.
That’s because bigger data sets—especially the unfettered, pilfered variety tend to garner better creative results. To put it another way, imagine you were training to be a painter several hundred years ago. Would you learn your craft better by observing as many paintings as possible—both good and bad? Or would you be better off siloing yourself from outside influences as you personally honed your craft?
Moonvalley is betting on the second option. Building on the expertise of researchers and engineers hailing from the hallowed ranks of DeepMind, Meta, Microsoft, and TikTok, it produced Marey, a generative AI video model for Hollywood studios and filmmakers preferring “clean” datasets unencumbered by potential copyright violations. “To get here, we made a bold decision: every single pixel used to train our model has been licensed directly from the original creator. No scraping. No shortcuts. No gray areas,” says Talukdar.
Because Moonvalley paid for the rights to train on this data directly from owners it can indemnify its clients in ways others competing platforms cannot. Such a legal shield might be the gamechanger Moonvalley needs to set itself apart from so many competitors. Already, brands and studios are facing increasing pressure to act ethically. In this regard, Moonvalley may have the edge—especially as audiences and creators demand more accountability from what they view as intellectual property theft.
For now, it’s worth returning to the topic of efficacy. One might assume training only on “clean data” would hinder the creative ability of an AI platform. After all, if you’re not feeding it vast amounts of content to learn from, won’t it be at a disadvantage? This pertains to a personal experience. Though I watched hundreds of movies in film school, my real cinematic education occurred when I worked for Creative Artists Agency as a reader. Over the course of two years I must have read over a thousand pieces of content—TV scripts, film scripts, books, treatments, to produce coverage. Exposing myself to the good, the bad, and everything in between was a godsend—informing my writing and critical thinking.
Talukdar admits this was a concern with his own company. However, Moonvalley sought to address the issue by relying on a diverse network of data partners, including indie creators and international studios to data collectives representing underrepresented voices. “The result is a model that not only delivers state-of-the-art results but also does so without parroting copyrighted material. In fact, Moonvalley’s model doesn’t even ‘know’ what Star Wars is.”
As I look around at a media landscape quickly transforming in the blink of an eye it’s comforting to know there are alternatives to doing business. Returning to my professor, he assigned us that novel because he wished to show his students there’s another way to build a Hollywood career, one that honors goodwill to others. Naturally, there will always be professional paths offering the easy route, ones that trade integrity for efficiency. But here’s to a brighter future for creatives and audiences alike, one that doesn’t just respect what came before, but honors those contributions as a win-win for all.