At the moment, Hollywood’s labor unions for writers and actors, the WGA and SAG-AFTRA, are on strike, and streaming residuals are one of the main deal points being negotiated. I’m a proud member of both of these unions, and I strongly support their demands. But zooming out, so to speak, I think there’s a larger conversation to be had about residuals that goes well beyond Hollywood.
Actors, writers and the rest of us film and television professionals might be some of the first (but not the last) to have our jobs threatened by so-called artificial intelligence — technology that ingests enormous amounts of “training data,” crunches the numbers and spits out new combinations of the data, following the patterns it was trained on. Tech giants, entertainment giants and every other profit-hungry giant will soon claim that AI can do human-level work at an astonishingly small fraction of the cost.
However, this claim will be hiding something, like the Wizard of Oz bellowing, “Pay no attention to that man behind the curtain!” What’s behind the curtain of AI? The cost of the human labor it took to produce the training data. “Generative AI” cannot generate anything at all without first being trained on massive troves of data it then recombines. Who produces that training data? People do.
And those people deserve residuals.
For example, let’s say a new AI program could generate a complete feature film at the click of a button. (Some industry insiders think this could actually happen pretty soon.) Building that AI will require a bazillion past movies to be used as training data. Therefore, anytime this AI yields revenue, the people who made those past movies will deserve a substantial piece. And I’m not just talking about actors and writers. I also mean people who don’t get residuals today: the camera operators, the costume designers, the sound mixers, everyone whose work the AI will be ingesting, mashing up and mimicking.
This principle holds true whether we’re talking about entertainers, doctors, engineers or pretty much anyone whose work involves a computer. AI can’t do our jobs yet, but it might be able to soon. And people whose jobs are threatened by AI will be the same people who produced the data used to train it. A new kind of residuals for these human data producers could potentially provide some much-needed economic relief.
And by the way, I’m not the only one advocating this kind of thing. Renowned technologists and economists, including Jaron Lanier and E. Glen Weyl, have long argued that Big Tech should not be allowed to monetize people’s data without compensating them. This concept of “data dignity” was largely responding to the surveillance advertising business models of companies such as Google and Facebook, but Lanier and Weyl also pointed out, quite presciently, that the principle would only grow more vital as AI rose to prominence.
Technologically speaking, the type of residuals program I have in mind would be a tall order. An AI system would have to track every piece of its training data, and then be able to determine which of those pieces influenced any given output generated and to what degree. On top of that, each of those pieces of data would have to be attributed to a verified human or set of humans, and there would need to be a channel of payment for those humans to receive their residuals. We did something a little bit like this at my company, HitRecord, producing pieces of art and entertainment with sometimes thousands of online collaborators and paying each contributing artist. Residuals from AI, in comparison, would require an astronomically huge scale, keeping track of zillions of bits and bytes, as well as billions of people, dollars, cents and microcents. (Disclosure: My wife serves on the board of directors of the OpenAI Nonprofit.)
So, now that I’ve said how hard the tech will be, let’s talk about the really hard part.
Tech companies won’t be incentivized all by themselves to build the necessary software. The law will have to make them do it. The U.S. Copyright Office recently issued an official statement that completely AI-generated works are not eligible for registration, saying, “These technologies ‘train’ on vast quantities of preexisting human-authored works and use inferences from that training to generate new content.” I imagine the copyright question will force tech companies to come up with a solution to attribute some sort of human authorship to an AI’s outputs. But here’s the question: Who owns the copyright to the training data? Who will therefore reap the benefits?
The answer to that key question, in the case of the film and television industry, is not the writers, actors, camera operators, costume designers, sound mixers or any of the other people who made the creations. The copyright is owned by the big studios. See the problem?
When I do a movie, and I sign my contract with a movie studio, I agree that the studio will own the copyright to the movie. Which feels fair and non-threatening. The studio paid to make the movie, so it should get to monetize the movie however it wants. But if I had known that by signing this contract and allowing the studio to be the movie’s sole copyright holder, I would then be allowing the studio to use that intellectual property as training data for an AI that would put me out of a job forever, I would never have signed that contract.
And again, this doesn’t apply just to the film and television industry but to many, many others as well. Take the medical industry. A doctor’s work produces lots of intellectual property in the form of medical records — and many doctors sign contracts allowing hospitals (or whoever) to take sole ownership of those medical records. But if the doctors had known this intellectual property would be used as training data for an AI that would put them out of a job forever, they, too, would never have signed those contracts.
So, if residuals from AI training data are based on existing copyright, then the aforementioned profit-hungry giants might indeed get to stop paying humans for their labor. How do we deal with this? Both of the Hollywood unions on strike today have expressed grave concern over AI, but their demands regarding AI will not address the pitfall I’m describing here. Nor do I think the problem can be solved by the courts. A number of independent visual artists have sued the AI companies Midjourney and Stability AI, claiming their work was used as training data without consent or compensation. Similar lawsuits are cropping up more and more. I hope these artists win these lawsuits. But I’m not sure they will without new intellectual-property law on their side.
I’m well aware that implementing this kind of residuals program would take a ton of work. It would require technology that doesn’t exist yet. It would also require new public policy, and most governments don’t exactly have a stellar track record when it comes to regulating new tech. Still, there are encouraging signs out there that the White House and Congress are moving toward regulating AI. So here’s my request: Residuals from AI training data should be made mandatory and intellectual-property law amended to make sure those residuals go to the people who deserve them. The seemingly miraculous outputs of generative AI should not be allowed to fill the coffers of tech giants or any other giants without paying the people whose data made those outputs possible.
AI is going to change things faster and harder than our intuition can predict. Will those changes be for the better or worse? I believe we still have a real shot at a world in which our careers are both more productive and more meaningful, and in which our compensation is both more honorable and more equitable than any we’ve ever known. But that kind of bright future won’t arrive automatically. As always, it’ll be us doing the work.
