EA says control is generative AI's next frontier
EA's Vaidya argues generative models produce convincing pixels but lack the deterministic control games need, pointing instead to neuro-symbolic systems that pair emergent generation with reliable rules and physics.
Electronic Arts' Vaidya laid out how the publisher thinks about generative AI in games, in remarks published by Runway from its 2026 AI Summit.
Vaidya pushed back on what he called the industry's "orientation towards neural everything." Some parts of a game can't be probabilistic: "You don't want the physics of an F1 car to be hallucinated." Games have to respond to players "60 times per second, endlessly," and stay in sync across thousands of concurrent players, so he sees deployment as "a complex systems coordination problem," not just a content-generation one. The frontier he is watching is "a neuro-symbolic architecture" that pairs emergent generative output with determinism and control.
He said today's models "can often produce very, very pretty pixels. They just don't feel like the right pixels," because they don't capture a studio's intent or style. Computer graphics built controllable primitives first and added fidelity later. Generative AI did it the other way around. Closing that gap, he said, will take deep customization and workflow collaboration between studios and model builders.
EA groups AI's impact into three horizons: efficiency, expansion and transformation. On expansion, he pointed to The Sims, which has engaged more than 500 million players and offers "almost a trillion different permutations" of play. His concrete example of efficiency was using AI to convert decades of assets from franchises such as FC, Battlefield, Madden and Skate into physically-based rendering pipelines. He said that work "would have taken armies of folks" in the past.
Runway's excerpt ends before he finishes describing that example. It doesn't name specific tools or partners.