David Silver’s new AI lab raises $1.1 billion to train systems without human data
A major bet on reinforcement learning
Ineffable Intelligence, a British AI lab founded only a few months ago by former DeepMind researcher David Silver, has raised $1.1 billion at a $5.1 billion valuation to pursue a very different kind of AI model: one that learns without human data.
The company says it wants to build a “superlearner” that can discover knowledge and skills through reinforcement learning, a method in which systems improve through trial and error rather than by studying human-generated examples. That approach is closely tied to Silver’s background. He spent more than a decade at Google-owned DeepMind, where he led reinforcement learning work before leaving to start the new venture.
From AlphaZero to a broader ambition
Silver’s work at DeepMind helped produce systems that learned to beat top human players in chess and Go without being trained on human game records or strategies. The best-known example was AlphaZero, which learned purely from experience and went on to defeat the world’s strongest computer programs in those games.
Ineffable Intelligence is now aiming to take that idea far beyond board games. According to its newly launched site, the company believes its superlearner could uncover “all knowledge” from its own experience, rather than relying on the vast datasets that power most current AI systems.
The lab’s language is sweeping. Its website says that if the effort succeeds, it would amount to “a scientific breakthrough of comparable magnitude to Darwin,” adding that its law would explain and build all intelligence.
Silver, who is also a professor at University College London, described Ineffable Intelligence as “his life’s work” in a personal note later published on the company’s blog. He also told Wired that any money he makes from the company will go to high-impact charities.
For now, the company’s finances appear to be the least mysterious part of the story. The $1.1 billion raise shows investors are willing to back a long-shot attempt to move AI beyond the human data pipelines that define much of the industry today.
Sources: