Haven wants to clarify machine learning's fuzzy future

Yesterday, we ran a story about a panel in which Haven Studios CEO Jade Raymond and chief technology officer Leon O'Reilly talked with PlayStation 4 and 5 hardware architect Mark Cerny about the studio's experiments in artificial intelligence and machine learning, and how such technologies might shape the industry in the future.

After the panel, we are given a chance to speak with O'Reilly and Raymond about the Montreal studio's hopes for the new tech. Both been around the games industry for a while, and they've seen a variety of heavily hyped new technologies come down the pipe – digital distribution, stereoscopic 3D, AR, VR, cloud streaming – and succeed or fail to widely varying degrees. Wtih that in mind, we ask if their attitude toward each "next big thing" has changed over the years.

"I'm still very optimistic," O'Reilly says. "We're very focused on game development for a lot of these technologies, and we're very clear on what problems we're trying to solve and how these will more efficiently solve them. When you have consumer-facing technologies, you don't always know whether there's going to be a market there for your technology or not, and that's often why they fail.

"I think we want to position Haven as a place where we do pioneer technology. And we will try things, and not everything is going to succeed. And that's OK, so long as the things that don't work, we fail quickly, then we iterate, we try something, we pivot, we change, and keep going. But I think fundamentally these technologies we're looking at have a very bright future."

Raymond adds, "The key here is we're trying to solve real development problems. The real development problems have to do with us wanting to make bigger and bigger games, the teams getting bigger, and the types of games players actually want changing. We've seen that people want more social games – even moreso since the pandemic – where people expect a certain amount of content updates. Obviously, graphics have gotten higher fidelity. And in order to deliver those kinds of experiences players really want, we need to re-think things within the dev team so we can create those player-centric, community-centric experiences."

Dozens of founding Haven employees came over from Google Stadia, where Raymond had been spearheading first-party software development for the cloud-streaming effort. And while the consumer-facing side of streaming technology has yet to pan out (Google itself announced Stadia's shutdown late last month), Raymond says cloud technology was a perfect solution for developers looking to tackle the very specific problem of how to work remotely during a pandemic.

"We see the promise of a lot of this tech, but we're really looking at very concrete things we're trying to do to solve specific problems as a dev team trying to deliver new types of experiences players want," she says.

During the panel, Mark Cerny had mentioned the potential of machine learning to enable teams of "a moderate size" to make bigger games. We ask if ultimately, that means fewer jobs for artists.

"We don't believe that's going to be the case, to be honest," Raymond says. "Are there fewer people that make a blockbuster movie than make a game? There are also directors, art directors, costume designers… we've replicated a lot of those roles in digital. Maybe at some point we're going to find new roles that don't even exist. And that's been the case for game development since we started making games.

"Back in the day, people used to look at this one programmer who did everything, right? And then we created these amazing games and now we've created needs for voice directors and composers who didn't have a role in games originally. So I think dev teams will continue to evolve, but I think it's going to be more based on what our imaginations will enable and where we want to take games ourselves."

"Maybe at some point we're going to find new roles that don't even exist. And that's been the case for game development since we started making games"

Jade Raymond

O'Reilly stresses that they do not want to shrink the development team in any way.

"But if we can, with that same dev team, create more interesting content, try more ideas, iterate more, iterate faster… this is what the tools allow us to do," he says. "It's about building those muscles for the creative team, enhancing and empowering them."

While some of the more splashy use cases for machine learning in development are still not ready for prime time, others are already implemented and impacting the pipeline, such as one system O'Reilly describes as "almost a texture compression system" that takes a grayscale texture with four colors and interpolates between them to avoid having an artist create a full-color version from scratch.

"It's not a particularly creative endeavor in any way, so instead of them spending their time doing that, we had a machine learning algorithm figure out the optimal way to pack the data," O'Reilly explains. "And it actually did better than the artists would do by hand. We're seeing significant memory and disc size savings from this, as well as artist time savings. That's just one example of where we can put these things into the production pipeline for artists, save the mundane work, and allow them to be more creative with their time."

And given the team's current emphasis on using these technologies for dev tools rather than user-facing parts of the game, they say there's less concern about any "black box" qualities of machine learning, where the model produces unexpected or undesirable outputs.

"The types of problems we're trying to solve are quite analog in a way. When you're trying to enhance your creativity, you're not looking for a precise result"

Leon O'Reilly

"The types of problems we're trying to solve are quite analog in a way," O'Reilly says. "When you're trying to enhance your creativity, you're not looking for a precise result. You're seeing what comes out and tweaking or iterating on that. It doesn't really matter what's going on in internally. If we were making, I don't know, a traffic light system that had to work precisely, that would be an issue. But if you're using it to enhance your artists' creativity and concepts, it's actually I think an advantage to be a little fuzzy."

During the panel talk, Raymond said Haven's north star was "creating a new IP that can be owned by the fans, where the fans are really driving that." Given the topic of new technology and the nature of many recent start-ups that have invoked the concept of ownership in games, we ask for their thoughts on blockchain technology and the metaverse.

On the latter, at least, Raymond describes herself as "a huge fan."

"I don't know what it means for our industry, but everyone seems to care about it now," Raymond says. "As a geek that loves sci-fi books, I love these little things coming up and people trying to make them reality."

"As a geek that loves sci-fi books, I love these little things coming up and people trying to make them reality"

Jade Raymond

O'Reilly adds, "It's great to be able to dream. Whether it comes through or not, it's exciting. I think we're living in a very exciting time."

Not having heard them touch on blockchain, we press specifically on the subject.

"Video games have forever had digital items," Raymond says. "We've had NFT-like things in video games forever. It's like, you do a certain thing and you get a collectible or a trophy, something like that.

"One thing we are pretty excited about is the concept of rarity and items in games – not tied to blockchain, per se – but I find it interesting how with blockchain and tied to crypto, things we've had in games take on a new form. [Items and rarity] is more something interesting we're talking about and thinking about."

O'Reilly offers a more definitive statement.

"Blockchain as a technology is not something we're engaged with at all," he says. "At all."