The Man Behind AlphaGo Thinks AI Is Taking the Wrong Path
David Silver, celebrated for creating AlphaGo—the groundbreaking AI that mastered the game of Go—now believes the current AI development strategy is heading in the wrong direction. In 2016, AlphaGo astonished the world by teaching itself to outplay top human champions, marking a milestone in artificial intelligence. Silver's new venture, Ineffable Intelligence, is now aiming to push AI even further.
A Different Vision for AI Progress
While most tech companies pursue superintelligence using large language models (LLMs) trained on human data, Silver contends this approach is fundamentally limited. He describes human-sourced data as a "fossil fuel," providing only a temporary shortcut. Instead, he advocates for AI that teaches itself using reinforcement learning—an approach where systems improve by exploring, experimenting, and adapting through trial and error. This, he says, acts as a "renewable" power source for continual advancement.
Building Superlearners
Backed by $1.1 billion in seed funding and a $5.1 billion valuation, Ineffable Intelligence is assembling top talent to create "superlearners"—AI agents capable of outsmarting people across diverse fields. Silver sees developing truly independent AI as a mission for the benefit of humanity; he even pledges to donate all profit from his equity to impactful charities.
Why Reinforcement Learning?
Silver and his team believe reinforcement learning, inspired by human and animal learning, holds the key to AI that can make independent scientific discoveries and innovations, rather than just echo what humans already know. He plans to develop AI agents that learn and interact within complex simulations, allowing them to practice problem-solving, cooperation, and adaptation beyond pre-existing human knowledge.
Tackling Alignment and Safety
Reinforcement learning comes with its own risks, notably the possibility of AI optimizing in ways misaligned with human values. However, Silver hopes that rigorous simulation environments will make it possible to monitor behaviors and ensure alignment. Supporters believe this approach could be safer than LLM-based systems, as the AI's learning isn't strictly imitative.
The Road Ahead
Silver’s unwavering focus on pure research and ethical goals has attracted top scientists and major investment. With rising interest in more general, autonomous AI, Silver’s alternative path could shape the future direction of the field.
For the full story, visit the original article by Will Knight on WIRED.
