Google’s head is in the cloud but its AI is 'grounded'

GOOGLE CLOUD NEXT, SAN FRANCISCO – Google spent a lot of time talking up the capabilities of artificial intelligence (AI) at Cloud Next this week. But a perhaps unappreciated undercurrent running through its announcements was reassurance that its AI tools won’t run amok. Or, more specifically, executives repeatedly insisted that the responses generated by models built on Vertex AI or Duet AI are “grounded.”

Put another way, Google is ensuring its AI cites its sources as a way to counteract the potential for hallucinations.

The topic of grounding came up multiple times during Tuesday's opening keynote session, with Google Cloud CEO Thomas Kurian, Vertex AI product lead Nenshad Bardoliwalla and Google Cloud Security’s Head of UX Steph Hay all mentioning the term while onstage. In fact, the company actually announced the launch of an enterprise-grade grounding service for Vertex AI foundation models and conversational tools which will allow companies to tie AI-generated responses to actual company data. It’s testing a similar grounding service for Google Search.

Keep up with the all the news from Google Cloud Next 2023 with our dedicated news hub here.

Bardoliwalla explained: “We can actually show you exactly where each of the individual pieces of content come from so that all of your data is grounded.”

Hay also noted grounding is also being applied to strengthen Google Cloud’s security offerings.

“Now, with Duet AI, I see a summary on this threat actor instantly generated for me. Further, the Duet AI model here is grounded, which reduces LLM hallucinations, so I have accurate summaries,” Hay said during an onstage cybersecurity demonstration using Mandiant Intelligence.

Why AI hallucinations matter

Google’s move to introduce grounding tools comes as the industry looks to head off potentially serious issues with AI. Security software provider Cato Networks recently flagged AI hallucinations as a key threat to enterprises.

“AI hallucinations will be the number one problem with the future adoption of this technology with privacy and prompt injection attacks following behind,” Etay Maor, Cato’s senior director of security strategy, told Silverlinings.

Read up on why AI hallucinations can be so scary, here.

Want to learn how to maximize your cloud investment? Check out our Cloud Executive Summit in beautiful Sonoma, Calif., from Dec. 6-7.