Alphabet is combining its DeepMind and Google Brain AI research units, ending a long-running internal rivalry between the London and Silicon Valley-based groups as it tries to make up lost ground in generative AI against Microsoft and OpenAI.
The shake-up will leave Demis Hassabis, the British head of DeepMind, in charge of an expanded group that can lay claim to many of the research breakthroughs that have turned AI into the industry’s most important new technology since the arrival of the internet.
He will also take on a more direct role in developing advanced AI systems that are central to Google’s future business, after nine years working on areas of advanced research that seldom related directly to the internet company’s operations.
The move to focus Alphabet’s most advanced research more directly on AI models with a direct application to Google’s internet business follows the breakout success of OpenAI’s ChatGPT, which has presented the biggest challenge to Google’s internet search business since Microsoft mounted an unsuccessful challenge two decades ago.
Alphabet chief executive Sundar Pichai said the shake-up was designed to “ensure the bold and responsible development of general AI”, and would leave Google with “a unit that will help us build more capable systems more safely and responsibly.”
DeepMind’s biggest breakthroughs have included its AlphaGo system, which beat world Go champion Lee Sedol in 2016, and AlphaFold, a system with potentially far-reaching impacts on medicine that predicts the structure of proteins.
By contrast, Google Brain, which was founded to apply the latest advances in neural networks to Google’s business, scored early successes in areas like machine translation and learning to identify images without human supervision. Its researchers also devised a new technique for “understanding” language called transformers, which led directly to the rise of so-called large language models race and services like ChatGPT.
Jeff Dean, a veteran Google engineer who has headed Google Brain, will move to a new position as Google’s chief scientist, as well as acting as chief scientist of its merged AI research division. He will continue to report directly to Pichai rather than Hassebis, who was named chief executive of the new Google DeepMind.
The AI reorganisation completes the slow assimilation of DeepMind into Google’s operations more than nine years after the London-based research group was acquired for $500mn. DeepMind’s founders fought early on to be kept at arm’s length from Google’s other research activities, leaving them free to pursue a research agenda aimed at developing advanced general intelligence.
Google took steps to make more practical uses of DeepMind’s work several years ago, transferring its research into uses of AI in healthcare over to a Google unit and setting up a team to try to apply DeepMind’s research breakthroughs to its own business.
In a sign that Alphabet was trying to harness DeepMind’s work to speed up its wider AI efforts, while also reducing its independence as a standalone unit, Alphabet said earlier this year that it would no longer treat DeepMind as a separate business inside its “Other Bets” division. Instead, the AI unit’s costs are now being treated as general corporate expenses for Alphabet, reflecting its role in developing key technology for the wider group.