HITS

Google Cloud Exec: Company Continues to Make Major AI Investments

NEW YORK – Google continues to make major investments in deep learning and other associated areas that fall under the artificial intelligence (AI) umbrella, according to Ron Bodkin, Google Cloud technical director of Applied Artificial Intelligence.

“Google has been invested in deep learning” for quite a while, he said May 2 during a keynote at the Artificial Intelligence (AI) Conference in New York. “We really started working in this space with Google Brain at the start of the decade and, in the last three years, there’s been a significant ramp in adoption of TensorFlow, our open source framework…inside of Google for building products,” he said.

Google Brain is a deep learning AI research team that the company started several years ago, while TensorFlow is the open source machine learning framework for high-performance numerical computation that the team originally developed for internal company use and released in 2015. Since then, many other companies have started using TensorFlow, including Advanced Micro Devices (AMD), Coca-Cola, Dropbox, ebay, Intel, Nvidia, Qualcomm, Twitter and Uber, according to Google.

Late last year, there were nearly 7,500 TensorFlow models to “check into our source code control system, and it continues to ramp rapidly,” Bodkin said, adding: “We’re using it in all kinds of ways,” including for gaming, as well as “real problems like how do we improve search quality?” The technology “really helps improve the quality of search for complicated queries,” he noted. Google is also “using it to drive improvements in advertising,” including for “smart bidding” and ad campaigns, he said.

Google is also using the technology “to enable entirely new products ,” including Google Home, which he pointed out “relies on a variety of deep learning capabilities that we’ve developed,” including language understanding, audio generation, and converting from audio to text,” he said.

The company is “using these techniques across all seven of our billion-plus user products and beyond, so we’re truly an AI company,” he told the conference.

Meanwhile, “one of the things that I think is really exciting is the continued participation of the open community,” he said, noting Google publishes information on the research it’s doing and the applications it’s creating, “often in collaboration” with others.

Google is “leveraging research,” while “continuing to develop better and better tools, and to make it more and more possible for our teams to use these capabilities with TensorFlow, and we’re open-sourcing and expanding the scope of TensorFlow,” he told the conference.

As Google continues to “invest in deep learning at scale,” it’s built “optimized” hardware as well, he said, pointing out “we’re onto our second generation” of cloud Tensor processing units (TPUs). Google started working on the TPUs “when we realized that if every Google user did three minutes of voice search, we’d need to build a dozen new data centers and we had to come up with another way to drive efficiency,” he said. The TPUs are now available in “public beta,” he noted, telling the conference that companies including Lyft are using them.

Google is also “doing a lot of research on raising the bar to make it easier to apply” deep learning, he said.

In a separate conference session that followed, Bodkin went on to explore in more detail how Google is using AI to enhance the digital experience.

As an example, he noted that AI now helps to reduce the “strain” of the email inbox by predicting responses based on the user’s personal email history. Twelve percent of all Gmail responses are “driven by AI,” according to him.

Spotify, meanwhile, is using the technology to increase customer “stickiness” through personalization in which users can now get up to seven music playlists customized for them each day and are also able to get near real-time music recommendations, according to Bodkin.

The conference, presented by Intel and O’Reilly Media, will return for its fourth year in New York April 15-18, 2019. The companies will also be presenting another AI Conference in San Francisco Sept. 4-7 this year, followed by one in London, Oct. 8-11 this year.