David Silver’s $1.1B Bet on an AI That Doesn’t Need Us

David Silver just made one of the biggest bets in AI history—and he’s doing it without human data.

His new lab, Ineffable Intelligence, raised $1.1 billion at a $5.1 billion valuation. The company is barely a few months old. That’s a lot of trust in a guy who spent years at DeepMind pushing reinforcement learning into the mainstream.

Silver was the lead behind AlphaGo, the system that beat Lee Sedol at Go in 2016. He also co-created AlphaZero, which taught itself to play chess, shogi, and Go from scratch—no human games, no human strategies. Just self-play and a reward function. That’s the core idea here: build an AI that learns entirely on its own, without relying on human-generated data.

This is a sharp departure from the current AI orthodoxy. Most of today’s big models—GPT-4, Claude, Gemini—are trained on massive piles of human text, code, and images. They’re mirrors of our collective knowledge, biases, and mistakes. Silver wants to build something that doesn’t need that mirror.

Ineffable Intelligence’s approach is pure reinforcement learning at scale. The idea is that an agent can learn complex behaviors through trial and error, guided only by a reward signal. No curated datasets, no human feedback loops, no supervised fine-tuning. Just an environment, a goal, and a lot of compute.

The $1.1B valuation is aggressive for a company with no product, no revenue, and a research direction that’s still unproven at this scale. But Silver has credibility. His work at DeepMind showed that RL can produce superhuman performance in games, robotics, and even protein folding (via AlphaFold, though that used some human data). The question is whether that approach generalizes to the messy, open-ended problems we actually care about.

I’m skeptical but intrigued. RL has a terrible sample efficiency problem. AlphaZero needed millions of games of self-play to master chess. That’s fine for a closed environment with a clear reward function. But teaching an AI to navigate the real world—where rewards are sparse, ambiguous, or delayed—is a whole different beast. Silver’s team will need breakthroughs in exploration, credit assignment, and safety.

The name “Ineffable Intelligence” is a bit pretentious, but it signals ambition. They’re not trying to build a better chatbot. They’re trying to build something that discovers knowledge on its own, without being told what to learn. That’s the dream of AGI researchers since the 1950s.

This funding round is a statement. It says that some investors believe the next leap in AI won’t come from bigger datasets or more parameters, but from algorithms that can generate their own training signal. Whether that bet pays off is anyone’s guess, but it’s refreshing to see a high-stakes alternative to the data-hungry status quo.