Jeff Dean Leaves Google to Found Discovery Loop

Jeff Dean, the engineer who spent 27 years building much of the software that runs Google, is leaving to start a company. He will co-found Discovery Loop, a firm set up to use artificial intelligence to speed up scientific and engineering research, according to reporting from TechCrunch and Bloomberg.

He is not going alone. Three other Google veterans are joining him: Sanjay Ghemawat, a senior fellow who co-designed the systems behind Google search and storage; Quoc Le, one of the founders of Google Brain; and Oriol Vinyals, a vice president at DeepMind. Between them they account for a large share of the research that made Google a leader in machine learning.

What Discovery Loop wants to build

The pitch is narrow to start and broad in ambition. Discovery Loop is set up as an independent public benefit corporation, and its first goal is to automate the work of machine-learning research itself: the slow loop of forming a hypothesis, running an experiment, reading the result, and trying again. If that works, the founders want to point the same approach at harder problems such as chip design, drug discovery, and clean energy.

The name captures the idea. Research is a loop, and the bet is that a capable AI system can run that loop faster than a team of people can, then hand the promising results back to human scientists to judge.

Google is backing the people leaving it

The unusual part is that Google is not simply losing these researchers. It is investing in them. Google will be a founding investor and cloud partner for Discovery Loop, and Radical Ventures and Khosla Ventures are co-leading the seed round, as GeekWire reported. The round has not closed, and the company declined to share a valuation.

The departures land at an awkward moment for Google. The company recently reshuffled its AI leadership, with Demis Hassabis moving from DeepMind chief executive to a chair role. Losing Dean and three names of his stature in the same stretch is a real dent, even with Google keeping a financial foot in the new venture.

This is a familiar pattern in AI by now. Senior researchers leave a big lab, take a slice of its funding with them, and set out to prove that a small focused team can move faster on the problems that matter to them. Whether Discovery Loop can actually automate science is an open question. The people trying it are about as credentialed as the field gets.