David Silver, the mind behind the groundbreaking AlphaGo program that stunned the world in 2016, now believes the field of artificial intelligence is heading in the wrong direction. His latest venture, a billion-dollar startup, aims to build what he calls "superlearners"—AI systems that can learn from experience more like humans do.
In a recent interview, Silver expressed concern that the industry's obsession with scaling up language models and massive datasets is missing the true potential of AI. "We're building systems that parrot data but don't truly understand or reason," he said. "The path to genuine intelligence lies in reinforcement learning and the ability to learn from the world directly, not just from text."
Silver's new company is focused on creating AI agents that can adapt and solve problems in dynamic environments, much like the way AlphaGo mastered Go through self-play. He argues that current large language models, while impressive, lack the flexibility and deep understanding needed for real-world applications.
"We need to go back to the principles that made AlphaGo a success: learning through interaction and reward, not passive pattern matching," Silver added. His comments challenge the prevailing AI paradigm and have sparked debate among researchers about the future direction of the field.