Google Brain and Deepmind form Google Deepmind, with a new focus on developing large-scale multimodal models.
Deepmind was founded in 2010 by Demis Hassabis and Shane Legg. The company was acquired by Google in 2014, but has remained largely independent of Google.
In March, The Information reported that Google’s in-house AI team, Google Brain, was joining forces with Deepmind for the secret Gemini project.
Google Brain + Deepmind = Google Deepmind
Now Alphabet and Google CEO Sundar Pichai and Deepmind CEO Hassabis have announced the merger of Google Brain and Deepmind to form Google Deepmind. It marks the end of an era.
Hassabis will become CEO of Google Deepmind, and Jeff Dean, previously head of Google Brain, will become chief scientist of the new company, reporting directly to Pichai.
“Working alongside Demis, Jeff will help set the future direction of our AI research and head up our most critical and strategic technical projects related to AI, the first of which will be a series of powerful, multimodal AI models,” Pichai said.
Together, the Google Brain and Deepmind teams developed many influential AI technologies over the years, including Transformer and Deep Reinforcement Learning. However, many of the scientists involved at the time have since left the companies, with some setting up their own AI startups.
“Get to that future faster”
“We have a real opportunity to deliver AI research and products that dramatically improve the lives of billions of people, transform industries, advance science, and serve diverse communities,” Hassabis said. “By creating Google DeepMind, I believe we can get to that future faster.”
Combining the team with Google’s computing resources will significantly accelerate its own AI development, Pichai said. This AI development is obviously about scale and multimodality.
The move is part of a larger realignment for the company, which seemed blindsided by the AI successes of OpenAI and Microsoft, and has since developed its own chatbot, Bard, offered its own AI models in the cloud and in Workspaces, and is looking to revolutionize search with its ‘Magi’ AI project.
Pathways and their consequences
The focus on multimodal models is in line with current trends being followed by other AI companies such as OpenAI or Germany’s Aleph Alpha. Dean presented this multimodal AI future of Google a few years ago with Pathways. From Pichai’s point of view, this makes economic sense, because even if it is not yet clear whether problems with large (multimodal) language models such as hallucinations are teething troubles or insurmountable obstacles, OpenAI, Nvidia, AI21Labs, Amazon, or Aleph Alpha show that numerous industries are very interested in the technology.
Pichai himself believes that Google can solve the problems of Bard & Co, as he said in a recent interview with 60 Minutes. However, with the German data protection authorities’ administrative proceedings against OpenAI, Google is likely to face another challenge apart from AI hallucinations.
With so much on its plate, it remains to be seen whether Google Deepmind will have the resources to continue to support open-source projects such as AlphaFold 2 and push research projects away from multimodal models.