GenAI at GDC: A Conversation About Reframing Gaming
What role will generative AI play in the gaming industry of the future? That is the question. Well, it was the central question everyone was asking back in March, at the Game Developers Conference (GDC) in San Francisco, California.
But there were also a few answers on offer at GDC 24. Many of these were formulated at an EPAM-led panel called “The AI Revolution: An Expert-led Discussion on Generative AI's Potential in Gaming.” Vitalii Vashchuk, Head of Gaming Solutions at EPAM, served up the questions, and a team of gaming all-stars returned them with insight and the occasional hint of wit. Our panelists included:
- Stef Corazza, Head of Roblox Studio at Roblox
- Chris Benjaminsen, Founder and Head of AI at FRVR
- Nico Perony, Director of AI Research at Unity
- Luis Villegas, Chief Technology Officer at Bungie
- Jeff Skelton, Head of Technology Partnerships at Electronic Arts (EA)
Gaming’s New Influencer: GenAI
The discussion kicked off with Vashchuk asking: How has GenAI influenced the creative and developer process?
Skelton remarked on the “create-edit-review loop” of creators and how GenAI gives them the ability to make faster, higher-quality iterations of their work. “The more swings at the bat you get at looking at something, the higher quality you can make,” he said, adding: “It's all about creative velocity.”
GenAI, we learned, is about creating partnerships with the tech. “Rather than replacing artists and replacing game devs and programmers, we’re trying to augment them. We’re building AI tools that can be injected in existing user flows to simply facilitate and accelerate game development,” Perony said.
Benjaminsen noted the “new paradigms of how teams are working together.” He said: “With GenAI, you can try 100,000 things almost free.” Those sitting in the directorship chair will see their role evolve into one about “choosing which of the infinite amount of choices [they] have,” and selecting the best one for the product that they’re trying to build.
Players: Empowering Them. Moderating Them.
But this tech won’t just affect directors. Users will also have skin, lots of it, in the creation game. Looking forward at the user-generated content (UGC) of the future, Villegas said, “It should be as easy in a few years to create a game or content… as it is today to create a TikTok video.” No longer will they need to fight “that thing called technology barriers;” instead they can devote serious time to creation.
Of course, users are also interested in things other than creation. Sometimes, in fact many times, toxicity can be a problem. Perony said that toxicity (which he defines as “people being horrible to each other”) is “the leading cause of churn, and 92% of players online report toxicity as a huge problem.” To address this, his company Unity is employing AI and machine learning to locate toxic comments in gaming environments. “We’re using AI to augment human moderators, not only to signal bad behavior, but also and more importantly, to encourage positive behavior and positive play.”
Corazza agreed, and talked about how GenAI can oversee the civility space. Roblox, he said, features real-time translation and moderation of chat “to make sure that the content is kosher and non-toxic.” Such work is not a task for people. Why? “Because if you have to detect the bad behavior in 100 milliseconds, there’s no way a human can do it.”
Ten Years is a Long Time
Maybe the most interesting moment was when the panelists tried to look forward a decade.
Villegas speculated that AAA games could be produced 10 years from now “with 25% less time and 25% less talent.” He admitted that it “sounds super controversial, and I understand that. But keep in mind that every time we had hit [previous] accelerations, the result [was] not smaller teams. The result [was] higher quality content, every single time.”
“Ten years is a long time,” Benjaminsen reminded everyone. “If you look at Moore's Law around hardware and acceleration data and you combine it with the pace of productivity increase from algorithms, we have a doubling rate of capability around seven months, which means that in 10 years, AI will be 100,000 times faster.”
Some Back-and-Forth about Data
But before fast-forwarding completely, the conversation paused on the topic of data.
Corazza said: “We should just give the control to creators and developers to decide if their content is going to be used for training AI or not.”
“We have to be realistic there: The cat is out of the bag, right?” said Benjaminsen “And it’s not going back in.” He mentioned the scenario of training on images that are not considered “fair use,” which is only open to “ginormous platforms, and then nobody else gets to participate.”
Said Perony: “This is the first time I disagree with Chris,” and he segued into Unity’s ethics of model training. “We have a responsibility to do right by our creators. And that means… using only content to train models that we license or own.”
“I actually think we agree,” responded Benjaminsen. “What you’re saying is that it's very expensive. Which means that only the largest companies in the world… can reasonably train a model that can actually be used in an IP setting.” He added that people who want to create with these tools in the future can “only do so by the good graces of some of the companies that already [own] most of the virtual IP in the world.”
Laying the Foundation
The situation, as you can surely hear, is complicated and that was reflected in the conversation. When it comes to GenAI and gaming, there’s much potential, but also much uncertainty.
It’s impossible, even for our experts, to know exactly how things will play out. But the industry needs to exercise all its energies, right now, to ensure it heads in the proper direction. As Villegas put it: “We have to really focus on how we build the right business model, the right ethical model, the right legal model, to create a strong foundation for the future.”
