As reported by Gamefile, Google outlined its Genie 3 generative AI technology in a talk at GDC, explaining that memory limits mean the world models it generates start to show inconsistencies after about a minute. Months earlier, the world model could only maintain consistency for a few seconds.
As Gamefile’s Stephen Totilo explains, the world models created by Genie 3 are generated one frame at a time, functioning more like a video that reacts to prompts rather than a traditional 3D game world, anticipating what the viewer will see next.
Alexandre Moufarek, product lead for the Inception team at Google DeepMind, said during the talk (which took place at GDC on Monday) that the DeepMind team is not thinking about replacing traditional video games. Instead, the firm hopes that Genie 3 can help “imagine new game experiences…you couldn’t even imagine without AI.”
Gamefile caught up with Moufarek after the talk to follow up on whether Genie 3 is not primarily about making video games.
Moufarek confirmed that video games were not the primary goal, but said that they might emerge from the tech. Instead, the DeepMind team is focused on the goal of artificial general intelligence (AGI) and creating bespoke virtual worlds for an AI agent to navigate. “We’re not at all in a stage where we can just, say, make a game with it,” said Moufarek.
Nevertheless, he thinks it would be interesting to put the technology in the hands of game developers so they can “play around with it.”
During the GDC talk, Google also outlined its SIMA 2 AI agent, which is designed to be able to play any video game. As a demonstration, the agent was asked to find a spaceship in the game No Man’s Sky. Google DeepMind first revealed SIMA 2 in November.
The reveal of Genie 3 caused stock prices of several video game firms to tumble back in January. But in an interview with GamesIndustry.biz, Pete Lewin, a partner at law firm Wiggin, warned that tech such as Genie 3 could face “greater scrutiny and pushback from IP owners” over copyright concerns.
