David Silver raised $1.1B to build an AI that doesn’t need human data

David Silver, the DeepMind alum who helped shape modern reinforcement learning, just pulled off one of the biggest AI funding rounds I’ve seen in a while. His new lab, Ineffable Intelligence, raised $1.1 billion at a $5.1 billion valuation. Not bad for a company that was founded only a few months ago.

The headline number is impressive, sure. But what’s more interesting is what Silver plans to do with it.

He wants to build an AI that learns entirely without human data. No scraping the web for text, no feeding it labeled images, no human feedback loops. Just raw, self-directed learning from scratch.

This is a radical bet. Most of today’s AI, from GPT-4 to Gemini, relies on massive amounts of human-generated data. We’ve been training models on the entire internet, then fine-tuning them with human preferences. Silver’s approach skips all that.

If you know his work, this makes sense. Silver was the lead researcher behind AlphaGo and AlphaZero, systems that taught themselves to play Go and chess at superhuman levels using only the rules of the game and self-play. No human games, no expert moves. Just reinforcement learning and a lot of compute.

He’s essentially trying to scale that philosophy to general intelligence. Instead of learning from human knowledge, the AI would discover its own strategies and representations through interaction with environments, simulation, or whatever Silver’s team cooks up.

$1.1 billion is a lot of conviction. The valuation suggests investors believe this isn’t just a research project but a potential business. I’m skeptical about the timeline – building a generally capable AI without human data is harder than it sounds. Language, for instance, is deeply human. How do you learn to communicate without examples?

But Silver has earned the benefit of the doubt. AlphaZero was a breakthrough precisely because it discovered strategies humans never thought of. If Ineffable Intelligence can pull off something similar at scale, we might see an AI that doesn’t just mimic us but genuinely thinks differently.

The name “Ineffable” is telling. It suggests something beyond description, beyond what can be captured in data. That’s either ambitious or arrogant, depending on your mood. I’m leaning toward ambitious.

This is a space worth watching. If Silver succeeds, it could change how we think about training AI entirely. If he fails, well, at least we’ll learn something about what human data actually contributes.

Either way, $1.1 billion says the bet is serious.