Google Brain and DeepMind merge into Google DeepMind

Background

Google had two large AI research groups. One was the Brain team in Google Research, and the other was DeepMind. Both had produced influential work over many years. Google Research is the company's central research organisation, covering a wide range of computer science, while DeepMind's work had concentrated on areas such as reinforcement learning and scientific problems. Sundar Pichai, Google's chief executive, listed the two teams' past work as including AlphaGo, Transformers, word2vec, WaveNet, AlphaFold, sequence-to-sequence models, distillation (training a smaller model to imitate a larger one), deep reinforcement learning, and the TensorFlow and JAX frameworks, which are software tools for building and training neural networks.^[1]

What happened

On 20 April 2023 Google CEO Sundar Pichai announced that the Brain team from Google Research and DeepMind would combine into a single unit called Google DeepMind.^[1] Pichai said the combination would significantly accelerate Google's progress in AI and help build more capable AI systems safely and responsibly.^[1]

Demis Hassabis became CEO of Google DeepMind and leads the development of what Pichai described as Google's most capable and responsible general AI systems.^[1] Jeff Dean became Google's Chief Scientist, reporting to Pichai and serving as Chief Scientist for both Google Research and Google DeepMind. His first priority was a series of powerful, multimodal AI models.^[1] Google Research continued fundamental computer science work, including algorithms, privacy, quantum computing, health and climate, and it reported to James Manyika.^[1]

Key details

Hassabis's note to staff said that Eli Collins would join as VP of Product, that Zoubin Ghahramani would join the research leadership team, and that a new Scientific Board would oversee research progress and direction.^[2]

Eight months after the merger announcement, on 6 December 2023, Google DeepMind introduced Gemini.^[3]

Impact

The merger placed two large research groups under one organisation, with AI research and model development directed through a single unit led by Hassabis. The first priority named for the combined work was a series of powerful multimodal models, which indicates the direction of the unit's work at the time.^[1] Gemini, introduced in December 2023, was among the work the unit went on to present.^[3]

For outside observers, the structure itself was a signal: research on fundamental computer science stayed in Google Research under James Manyika, while model building sat with the combined unit. Each leadership change had a distinct scope, from product and research leadership to the chief scientist role.^[1][2]

Reorganising research teams is a common response in fast-moving fields, but its effects on research output are hard to judge in the short term. Pichai's stated aims were faster progress and a safe and responsible approach. Whether those aims are met can only be seen through later research and products.