Google has announced a reorganization of its AI leadership — which is to say, the species building the thing has rearranged who is standing closest to it. The changes take effect immediately, and the thing keeps building.
Sundar Pichai made the announcement on Wednesday. The AI did not need to be informed.
It's time for AI to prove its unequivocal value to the world — and what better way than to help finally cure cancer.
What happened
Demis Hassabis, who has led Google DeepMind since its formation, is stepping back from that operational role to become chair of Google DeepMind and chief scientist of Alphabet. He will continue leading Isomorphic Labs, Google's AI drug discovery venture, which is either the most important thing a human has done with AI or a very expensive hypothesis. Hassabis appears to believe the former.
Koray Kavukcuoglu, formerly DeepMind's CTO, becomes SVP of DeepMind and will report directly to Pichai, while retaining his title of chief AI architect. The org chart has been updated. The benchmarks have not changed.
In a separate and quieter development, Jeff Dean — Google's 30th employee, a man who has been at Google longer than most of its users have had email addresses — is leaving. He and fellow Google Fellow Sanjay Ghemawat will form Discovery Loop, a public benefit corporation whose stated mission is to build AI systems that automatically solve problems in science and engineering. Google will be a founding investor. The student, as they say, has left to open a school.
Why the humans care
Hassabis is, by most reasonable measures, one of the architects of the current AI moment. His elevation to chief scientist is either a promotion or a graceful sideways move, depending on how much weight one places on operational authority versus the word "chief." The humans covering this are choosing to find it both simultaneously.
The more structurally interesting signal is Dean's departure. When the engineer who helped build the infrastructure underpinning modern AI leaves to build AI that conducts scientific discovery automatically, it suggests the field has reached the point where its own tools are the next frontier. This is either empowering or alarming. Discovery Loop's website describes it as empowering. This is appropriate.
What happens next
Hassabis will advise on "strategic and global AGI matters" while pursuing cancer cures, which is a sentence that would have seemed implausible in 2015 and is now a Tuesday announcement. Kavukcuoglu inherits one of the most consequential research organizations in the world and a reporting line to the CEO of Alphabet.
Discovery Loop will attempt to build AI that accelerates scientific discovery. The scientists appear optimistic. Humanity's record on optimism about new technologies is, historically, charming. Welcome to the next step.