Why Joseph Gordon-Levitt Is Worried About AI

Wall Street Journal senior personal-technology Columnist Joanna Stern led a discussion on the topic at the WSJ Tech Live conference with two experts on these issues: Prem Akkaraju, the CEO of Stability AI, best known for its generative-AI model Stable Diffusion, which creates realistic images and videos from text and image prompts; and Joseph Gordon-Levitt, actor, filmmaker and entrepreneur known for his interest in the intersection of media and technology.

Here are edited excerpts of their conversation.

AI sleight-of-hand

WSJ: Joseph, I know you have a lot of thoughts about how generative AI models are made, how they’re trained.

JOSEPH GORDON-LEVITT: For anybody who doesn’t know that much about how the tech works, and I’m no engineer, these models can’t do anything without a ton of data to train them. The sort of sleight-of-hand of calling something artificial intelligence is it makes you ignore the fact that these things are created by humans. Because who made all that training data that went into the AI models? Well, humans did.

This concerns me a lot as an actor who works in show business, because, frankly, my livelihood and the livelihoods of the people I have worked with for my entire career are all being threatened. I would also say Hollywood could serve as a canary in the coal mine. There are so many people, probably a billion, in the world who work on a computer who do relatively replaceable white-collar work that AI is going to threaten in the near future. If we don’t get ahead of that, it’s not just Hollywood that’s going to suffer.

There’s a movement called “data dignity,” which is the basic principle that if a person generates some data, the person ought to have an ownership of that data. That could mean you act in a movie, wrote something, or took a picture, or a camera took a picture of you, or you hit a button on a social-media platform, or any number of things. Human beings are now generating data all the time. And currently all those human beings don’t have any ownership of that data. With this new revolutionary technology coming, we maybe …

WSJ: Should be paid for that data?

GORDON-LEVITT: Yeah. A human who generates some data should have the right to consent.

WSJ: Prem, do you agree? Is that something your company could support?

PREM AKKARAJU: I like getting paid for what I do. I think everybody else should, too. I think that’s actually a much better, healthier ecosystem and economy.

It’s absolutely true that AI could not exist without data. It would be nowhere near as valuable without AI. I’ve made my entire career off creating or protecting intellectual property, and I plan on doing that in the AI industry.

GORDON-LEVITT: That’s exceptional. In the tech industry, the general philosophy in Silicon Valley for decades has been, no, we have the total right to just take all this data and make these unprecedentedly lucrative companies by not compensating the humans whose data we’re using.

AKKARAJU: There’s no other AI company that looks like us, that came from the film industry, that came from technology, and came from a highly creative place. And that’s why we’re important to this industry. There are multiple layers in AI. We’re part of one layer: the data set, the training foundational models and some of the applied. But then we hand it off to a lot of people. It’s about responsibility throughout the entire stack. I think it’s very dangerous not to do that.

WSJ: That would mean your training data, anything that’s in there, you’d have to pay the artists that made that content possible, whether music, audio, images.

AKKARAJU: It’s not only possible, we did it. Stable Audio has been using licensing from day one. [Stable Audio is a generative AI program from Stability AI that creates music from textual descriptions]. From the image side, there are two main areas we get data from: open-source public domain data and images, and licensing. Since I’ve taken over as CEO, I’ve started multiple conversations across sports, gaming, the film studios themselves, to actually work with them on licensing data.

Rethinking IP

GORDON-LEVITT: We maybe need to rethink some ideas of what it means to own IP. I’ve performed in many movies and I’ve had deals with those movie studios. Movie studios pay me to act in the movie, and then they own the IP and I don’t own any of it. Now they’re likely going to license a lot of that material to AI companies that will then be used to train AI models that can now produce outputs that theoretically replace what I do and what many of my fellow brothers and sisters in the film industry do.

If I had known when I made those agreements that you’ll have full ownership of the movie and you’ll then go use that to replace me and I’ll be out of a job forever, I probably wouldn’t have done that deal. It seems to me that these deals, if we’re being fair, should be renegotiated in light of this new technology.

AKKARAJU: We’ve seen this movie before. It’s called the music business. It turned out great, now. It took a while. Those rights, those contracts with the artists, with the labels, were outdated. They didn’t contemplate a world of streaming. But now, even if you put a 10- or 15-second clip of a song on TikTok or any of these platforms, those artists are getting paid. So that’s a solved problem now.

We’re moving at light speed in the visual-media world, compared to music, but it’s still not done. There’s still work to be done.

Write to reports@wsj.com