Now here’s our crosstalk on the state of play in A.I.…

I’m on record saying the manner in which Hollywood responds to artificial intelligence in the next few years will be the defining move of the next few decades. And there’s been lots of news on that front over the past few weeks. So much, in fact, that I needed to assemble Puck’s new A.I. correspondent, Ian Krietzberg (sign up for his email here), and our legal expert, Eriq Gardner, for a tag-team cross-talk on the state of play when it comes to generative A.I. and the professional content industry…

Matt: Ian, welcome! You talk to generative A.I. companies, just like I talk to the traditional entertainment people, and my people are profoundly scared about where this is all going—even as many of them are using A.I. and the studios are all counting on the technology to lower costs. I’m curious how the top A.I. players see Hollywood and its concerns about protecting intellectual property, amid the push for better and better A.I. models. Is there any chance that this isn’t a zero-sum game? Is there even a desire to work together, or is that all a Trojan horse?

A MESSAGE FROM OUR SPONSOR

This is Atlassian Williams Racing

Two powerhouse teams are joining forces for one epic goal. Atlassian is thrilled to become the Official Title Partner and Official Technology Partner of Atlassian Williams Racing.

This is a partnership for the ages. Formula 1 is defined by high-stakes teamwork and innovation – and both Atlassian and Williams Racing were founded on those same principles.

For 20+ years, Atlassian has worked alongside high-performing teams to develop a System of Work that helps them work more effectively together. Now, in partnership with Williams and a global community of F1 fans, we’re turbo-charging teamwork on the race track.

Ian: There are two elements to this. I’ll start with Hollywood’s copyright concerns, which have been echoing across just about every facet of the media industry since ChatGPT went live in November of 2022. The dominant view of the A.I. companies is that the use of copyrighted content for model training purposes is both fair and necessary; that said, these companies have also recognized that this perspective might not exactly hold up in court, and so they’ve taken to securing licensing agreements with I.P. holders, when possible. Still, the impression is that, if you put something “out there,” it’s free for the taking.

Matt: Nice. You can’t see me but I’m shaking my head.

Ian: The other element—the Trojan horse question, as you put it—has to do with what exactly these players are attempting to accomplish in the media space. The goal goes far beyond subtle augmentation; instead, it’s a dramatic transformation of what media is, and how it’s consumed. The vision is a world of hyper-personalized, live, interactive, and on-demand film, TV, and video games. You’ll frequently hear the word democratization, which serves as the basis of A.I. video–generation platform Pika’s entire organization. The company’s tagline is, “We’re for anyone with an imagination.”

Now, it’s not at all clear that such a vision will be technically feasible. The models that underpin these efforts so far remain limited in a number of ways that make such a vision lofty, to say the least. But if they were able to design a system that could reliably (and cost-efficiently) bring this mission to fruition, we might see creators, as a source recently suggested, doing less creating, and more establishing overall visions—or guardrails—for these generative media experiences. And if something like that rolled out at scale, you have to imagine there’d be even fewer jobs in Hollywood.

Matt: How long do you think until a fully A.I. film or TV series breaks through and finds an audience comparable to traditional movies or shows?

Ian: Frankly, I would be really surprised if something like this were to happen anywhere in the near or mid-term future. I actually can’t really imagine an environment where we’d see something like this occur, and the reason has less to do with A.I., and more to do with humans and storytelling.

There was a study conducted by psychologists at Duke in 2023 that aimed to address the open-ended possibility of machines replacing humans in the world of art-making. For the study, the researchers showed their subjects a series of A.I.-generated images, randomly assigning “human-created” or “A.I.-created” labels to the images. They found that “people tend to be negatively biased against A.I.-created artworks relative to purportedly human-created artwork,” due to perceptions around effort and narrative storytelling.

That validates my own personal feeling: that the act of storytelling is uniquely human, and requires a human community. A.I. lacks intention and emotion. When it generates art, the bot isn’t trying to communicate anything at all. It’s all statistics, not soul. ChatGPT today can generate a novel—thanks to all the books in its training data—and publish it on Amazon, but it won’t be any good, and likely no one’s going to read it. And so I find it hard to envision a world where large groups of people can fully separate humanity from the arts, no matter how realistic things might look. But maybe I’m just old-fashioned.

Matt: You’re saying A.I. can’t fully replace the human touch. But it’s not going to be long before it can replace everything except the artist’s intention, right? How long until 80 percent or 90 percent of the hundreds of people who work to create every movie or television show will be thought of as, say, hand-drawn animators in the years after Pixar made them obsolete?

Ian: It depends. One precedent might be Industrial Light & Magic, the incredible special effects team behind Star Wars, among others, which underwent a long and rocky transition from using scale models to primarily computer animators. But while the physical effects department went away, and that was a difficult moment for many people who worked at ILM, people themselves never went away. ILM just hired fewer and fewer model folks, and hired more and more computer animators.

With generative A.I., you’re not going to have the same number of people working on different tasks. You’re also almost certainly going to need fewer people, full stop. But it might take longer than you think to get there.

Right now, we have A.I. systems that can help enable better real-world simulation, or face-swapping animation, or de-aging. But we aren’t close to having systems capable of full automation. Corporations outside of Hollywood are dealing with this now, where A.I. projects are often far too expensive, and, due to consistent reliability problems, aren’t really worth the expense—Gartner expects 40 percent of so-called agentic projects to be canceled by 2027 as the hype simmers down. 

To get to the point where 90 percent of the people who work on a film become obsolete would require a couple of things: major technical breakthroughs; time for those innovations to be adapted (which could be as much as 20 years); and an audience that supports the final product. So I’m pretty skeptical. A world in which it’s possible to automate 90 percent of Hollywood—scouting locations, generating truly unique scores and scripts, convincing performances, smart editing—is also a world in which 90 percent of all work is fully automated. At that point, we’d have a lot of other things to worry about.

Matt: Trust me, nobody in Hollywood cares about anything other than Hollywood.

Matt: The Trump administration is about to issue its A.I. “action plan,” which will suggest a framework for how U.S. copyright rules are applied to training models. Disney’s Bob Iger and his top lawyer just went to the White House to argue for greater protections. Meta’s Mark Zuckerberg, who is spending billions to recruit an A.I. “superintelligence” team, was also spotted in the Oval Office this week. Given that the head of the Copyright Office was fired just as she put out a report advocating for protection of works like movies and TV shows, what’s your sense of how the Trump team will come down on this issue?

Ian: Everything that we’ve seen from this administration so far suggests that it is very much in the corner of large tech companies. Where Iger and his Hollywood and media peers want stronger protections, the major A.I. companies want copyright exemptions. In its policy proposal, OpenAI called it a “matter of national security” that the fair use doctrine, which traditionally gets decided on a case-by-case basis by the courts, apply to A.I. training. Their argument is simple: A.I. labs in China “will enjoy unfettered access to‬ data—including copyrighted data—that will improve their models.‬‭ If the P.R.C.’s‬ ‭developers have unfettered access to data and American companies are left without fair‬‭ use access, the race for AI is effectively over.” (That letter is from OpenAI policy chief Chris Lehane, the former Clinton guy who readers might know as a longtime lobbyist in California. He also wrote the 2012 movie Knife Fight.)

Matt: Yes, I just read that China is aggressively using A.I. to turn old kung fu movies with Jackie Chan and Bruce Lee into new animated features for a younger generation. That certainly wouldn’t fly with U.S. copyright law or the Directors Guild.

Ian: Google is a little less dire in its framing, though it wants the same thing—a legal exemption for training on copyrighted material. I would expect Trump to land on the side of the tech companies here, though it’s not really clear what that would look like in practice.

A MESSAGE FROM OUR SPONSOR

This is Atlassian Williams Racing

Two powerhouse teams are joining forces for one epic goal. Atlassian is thrilled to become the Official Title Partner and Official Technology Partner of Atlassian Williams Racing.

This is a partnership for the ages. Formula 1 is defined by high-stakes teamwork and innovation – and both Atlassian and Williams Racing were founded on those same principles.

For 20+ years, Atlassian has worked alongside high-performing teams to develop a System of Work that helps them work more effectively together. Now, in partnership with Williams and a global community of F1 fans, we’re turbo-charging teamwork on the race track.

Matt: Isn’t this like General Motors arguing they need free engine parts from vendors or they will lose ground to cheaper Chinese carmakers? Why should global competition be an excuse for theft?

Ian: It shouldn’t be. But this is one of those arguments that has become a massive Silicon Valley talking point. The interesting thing is that specifics, details, and definitions are so vague. We’re supposedly in this race with China (and the rest of the world) to win on A.I., but it’s not clear what exactly we’re racing to achieve.

If we’re in a race to build the kinds of models that essentially disrupt the media industry, that feels like a business venture, rather than an urgent national security interest. And if we were in a race to own the absolute cutting edge of scientific advancement and research, through and with A.I., we don’t need Disney movies to do that. Medical A.I. systems, for instance, are generally trained on a combination of medical images and patient records—pictures of Mufasa aren’t going to help you there.

And if we were trying to build an artificial superintelligence—a scientifically dubious hypothetical—I would expect to see systems that can flexibly and reliably generalize based on a very small amount of data. So, again, Disney isn’t going to help you there.

The loose idea behind this impression has to do with scale. There has been an observation (so far) in this field that, the more data and compute you throw at models, the better they perform. Most of the major A.I. labs today have been aggressively pushing on scale for the past couple of years, generally to relative success. And yet scale has not solved A.I.’s reliability problem, and we are talking about corporations spending billions annually on talent, chips, and electricity. Most would rather not spend more on licensing their training data, too. That’s more what it comes down to.

Matt: Two recent legal decisions involving Anthropic and Meta seem to have tipped in favor of A.I. companies. Keith Kupferschmid of the Copyright Alliance told the Journal that the cases “may create this copyright chaos for A.I. companies and copyright owners.” Eriq, what is your takeaway from the rulings?

Eriq: First of all, I’m sure Mark Zuckerberg is grinning ear to ear that you’re calling Meta an A.I. company. As for the rulings, I don’t find them especially definitive, or all that instructive about how other judges might rule. They were nominal wins for Anthropic and Meta, sure, but with some pretty significant caveats.

In the Anthropic case, Judge William Alsup signaled the outcome could be different if the plaintiffs were targeting infringing outputs rather than just the A.I. training data itself. And in the Meta case, Judge Vince Chhabria went further, taking issue with Alsup’s analogy that A.I. training is like using books to teach children how to write. Chhabria said that’s a flawed comparison that glosses over the potential harm A.I. might cause to the market for copyrighted works. The only reason he ruled the way he did was because the book authors hadn’t put forward much evidence at all showing, or at least speculating about, that harm. He was openly critical of the lawyering.

Bottom line: These are important early rulings on A.I. and fair use, absolutely, but there’s going to be plenty more ahead. The plaintiff lawyers in the Meta case have already filed a new suit, and appeals are certainly coming.

Matt: And how might these early precedents apply to the big Disney and Comcast case filed last month against Midjourney—a smaller generative A.I. company?

Eriq: I actually think Disney and Comcast are quietly pleased by the rulings. As I pointed out a few weeks ago, their case isn’t focused on A.I. training. It’s very deliberately aiming at those outputs, especially visual outputs—images that look too much like Mufasa, Darth Vader, and other iconic characters. So to the extent these judges have been flagging weaknesses in purely training-focused copyright claims that don’t draw a good connection to the impact on the marketplace, Disney and Comcast can feel good that they’ve addressed those issues.

Also, just because Disney and Comcast have finally sued over A.I. doesn’t mean they should be lumped with Hollywood’s loudest A.I. critics. Take the White House A.I. action plan, which you mentioned. Did you see the Motion Picture Association’s comment?

Matt: Don’t get me started on that. I think it’s hard for the M.P.A. to take a forceful position on A.I., because Amazon and Netflix are now members alongside the traditional studios.

Eriq: The M.P.A. cautioned that “sweeping generalizations that training is always, or is never, lawful under the fair use doctrine are neither helpful nor correct.” It even acknowledged the possibility, ahem, that content creators might use A.I. tools on their own content. Alas, to the extent courts are leaning into nuance right now, that actually lines up quite well with the studios’ own stated philosophy.

Matt: James Cameron and others have agreed that the best legal path is to focus on that finished product. Not using Darth Vader, but mimicking Vader. So far, is that argument any more effective?

Eriq: It’s too early to tell. There’s that case where Tesla asked an A.I. tool to create “Elon Musk in trench coat looking into the city” and it spat out an image that looked too much like Blade Runner 2049. A judge has allowed Alcon Entertainment’s copyright claim against Tesla and Elon to move forward there.

That said, my hunch is that the Midjourney case, and others like it, are going to wrestle with another question: Are we talking about direct infringement, or are these Homer Simpson copycats happening at the hands of A.I. users? And if it’s really the latter, should the A.I. companies be on the hook for contributory copyright infringement?

Disney and Comcast have pled both theories in their suit, although I suspect they’ll end up having to proceed on contributory infringement claims. But those can be tricky territory. There’s of course the father of all modern cases—the Sony Betamax opinion, where the justices ruled that the manufacturer of the VCR wasn’t liable because it was capable of “substantial non-infringing uses.” And then we get into stuff like whether mere knowledge of misbehavior by users is enough, or whether it’s necessary to show culpable intent on the part of the tech company. Just this week, the Supreme Court agreed to hear Cox Communications v. Sony Music, a case about secondary copyright liability in the context of I.S.P.s and user piracy. Keep an eye on it. It may have ripple effects for Disney v. Midjourney, and the like.

Matt: Actress Natasha Lyonne and others are part of a new A.I. studio called Asteria that uses only generative A.I. that trains on data and images used with permission and, presumably, payment. I seem to get a press release about one of these companies every week, and I even had one of them on The Town. Do these companies have any shot at success?

Ian: Even if audiences evolve to a point where they can truly and fully separate humanity from the arts, I don’t really know if these more ethically minded companies stand a real shot. It’s a kind of funny problem. Consumers who care about the human ethics behind the scenes presumably wouldn’t be interested in viewing A.I.-generated “art” at all, whether or not it comes from licensed models (there are a lot more ethics and morals at play with generative A.I. than just I.P. theft.). Conversely, consumers who don’t care about the human element in the content they consume presumably wouldn’t really care too much about the licensing—or lack thereof—of the models responsible for that content.

I suppose the audience Lyonne is going for is some sort of middle ground—people who are down with A.I. content, but want to feel a little better about it. Think of the three categories as vegan (anti A.I.), organic (ethical-ish A.I.), and conventional (straight-up A.I., no fancy labels.)

Matt: Well, I think the companies want to attract artists to participate in A.I.-driven projects, and the artists definitely care about the licensing part.

Ian: Sure, but as of right now, I see a lot of vegans in the audience, and I see a small but strong base of folks who would be down for straight-up A.I. When it comes specifically to A.I. in the media space, I don’t really see too much of that middle ground.

See you Monday,
Matt

Got a question, comment, complaint, or the perfect watermelon salad recipe? Email me at Matt@puck.news or call/text me at 310-804-3198.

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St, New York, NY 10006