Google Cloud, Research Units Collaborate to Speed AI Development

Google said it has seen faster progress in Artificial Intelligence (AI) as a result of closer collaboration between its cloud computing unit and its AI research lab.

Google Cloud struck deals to weave its latest AI models into popular consumer products, sharing more than 75 customer stories, including from the social app Snapchat and the storage company Pods, the company said as part of its virtual Gemini at Work event. The company also unveiled new versions of its AI model, Gemini.  

Collaboration between Google Cloud and the company’s research arm, Google DeepMind, can help accelerate product to market, executives said. That’s important as investors press parent company Alphabet Inc. to show how AI investments can turn into new business.

It’s closer work between two business units that both have something to prove. Google Cloud, which has long trailed Amazon.com and Microsoft in the cloud computing market, is trying to use the excitement around generative AI to expand its business. Google’s AI researchers invented much of the technology that underpins the latest wave of AI, but after OpenAI leapfrogged Google with the launch of ChatGPT, the search giant has been under pressure to smooth the path from research labs to products for consumers and businesses.

As progress in the field accelerates, Google’s AI researchers have had to “pick up the pace,” said Eli Collins, vice president of product at DeepMind.

“The researchers are incredibly motivated by seeing their work in the hands of real products and users,” Collins said. “That’s also been a culture shift, not just within Google, but the rest of the research community, where so many of the leading research labs are actually product companies at this point.” 

In recent years, Google has lost a series of researchers who were impatient to launch products. But one of the most outspoken members of that camp, Noam Shazeer, recently returned.

Across Silicon Valley, companies have struggled to unite research and product teams, which often have different incentives, said Sharon Zhou, CEO of Lamini Inc., a startup that helps companies use AI systems known as large language models. 

“Production AI and AI research teams are notoriously hard to get collaborating well,” Zhou wrote in a message. “Whoever figures out how to get these teams collaborating well will be a market leader. They are in strong tension at many companies, from Google to their competitors.”

In an interview earlier this year, Google Cloud Chief Executive Officer Thomas Kurian said strengthening ties with DeepMind has been a key focus, noting that teams in London, Seattle and the Bay Area are working closely, sometimes even sitting side by side. In one session in June, Google Cloud employees presented customers’ experiences with generative AI to DeepMind researchers, and some features discussed by the teams were included in Tuesday’s model launch, a Google Cloud spokeswoman said. 

To sharpen the focus on product, DeepMind researchers are working not only with Google Cloud but also with corporate customers, Collins said. In conversations with customer Snap Inc., the owner of Snapchat, the company shared how users were often using the app for homework help. DeepMind tailored its models accordingly, Collins said. 

“Cloud customers are our customers,” Collins said.

© 2024 Bloomberg L.P.