A YouTuber-turned tech executive is launching an AI media lab that is betting on interactive video as the future of entertainment.

Ben Relles, who channeled a career as an early YouTube creator into an executive role at the video platform, is launching Make Believe, an AI lab that will focus on creating tech that enables videos that can talk back to viewers, betting that it will enable forms of entertainment never previously possible.

“A lot of the conversation in Hollywood is around how AI will make things cheaper and faster for movies and TV, and I would say for us we’re more focused on what formats were impossible before AI existed,” Relles tells The Hollywood Reporter in an interview.

Imagine, for example, a cooking video where you can ask the chef questions as you prepare the dish at home, or a fitness creator who can critique your form, or a guitar coach helping guide you as they teach you a song.

“Some of this we can already do, and some of this we need to build the tools to be able to do it, but I would say that general idea that platforms like YouTube have been such a great place for people to learn to do new things, I think can be really multiplied by what interactive video enables when the video can adapt to you, and it can see what you’re doing,” he explains. “For 100 plus years video has been something you watched. We think part of the next era of video is something that is genuinely interactive.”

The leadership team of Make Believe also includes fellow YouTube veteran Margaret Burris, and Alec Lindsay, who was previously at HeyGen. But Relles, as it happens, knows a thing or two in particular about what it takes to be a creator.

He co-founded the Vsauce network and the Key of Awesome channels on YouTube (you might recall their viral 2007 hit “Crush On Obama,” which has nearly 28 million views), with a combined seven billion views between them. He ultimately joined YouTube, where he led innovation and unscripted programming at YouTube Originals.

It was while at YouTube that Relles connected with Reid Hoffman, the LinkedIn co-founder and AI enthusiast. Hoffman is an investor in Make Believe, with the team’s tech helping to create “Reid AI,” an AI avatar of Hoffman that is trained on his writing and speeches.

THR, in an unusual arrangement, was referred to an AI-powered Reid Hoffman avatar for quotes about his decision to invest in Make Believe. The avatar, without irony, stated of Hoffman’s investment: “It’s about amplifying human agency and connection through AI, not replacing it.”

“A big part of what this technology I think has the potential to do is give you access to people you wouldn’t have access to otherwise,” Relles says of the tech his company is creating. “And the next step was really creating the best in class real time video avatar, an avatar that you can talk to, have a conversation with, it answers in real time, it can pull from the right interview or book or podcast, depending on what your question is, and so we started finding a lot of different use cases for that real-time avatar.”

Make Believe already has a deal lined up with A+E Networks’ The History Channel to develop AI avatars of historic figures, which users would be able to quiz about their experiences and ask questions of.

“People are visual learners, people learn through interactivity. I see a lot of applications through science and math,” Relles says. “Thinking about how you could learn about history by actually having conversations with historical figures about the times that they experienced firsthand, I’d say that’s a general category that I think will be a space that we want to experiment and build, for sure.”

But the current moment of AI video reminds Relles of those early days of YouTube, when it was clear that a new form of entertainment as emerging, but its final form wasn’t entirely clear just yet.

“I always loved the first few years of YouTube, where these new formats would break through that nobody would have expected: Toy unboxings and hair tutorials and gaming play-alongs, and I think it’s going to be similar here,” he says. “I have real convictions around where this will be most valuable, but I also recognize that which specific formats break through in interactive content will be a lot of experimentation and trial and error.”