Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
Conversational AI using Gemini helps organizations find new insights using natural language
Being able to interrogate complex data for answers to real-world questions without having to code how to do so has been the dream ever since Captain Kirk first said “Computer?” from his command chair.
Generative AI has finally made this interaction accessible to anyone, but not always reliably due to unchecked source data and hallucinations. Data agents bring business value to natural-language inquiries and functions by using LLMs with a company’s own data.
Google Cloud uses conversational data agents powered by its Gemini generative AI tools to help users analyze data, answer questions and perform tasks, enhancing data analysis tools like Looker and BigQuery enterprise data warehouse.
Peter Bailis, Google Cloud vice president of engineering for data analytics, explained that traditional business intelligence tools offer easy access to data via a dashboard but digging deeper can be a time-consuming and complex process.
“Using a dashboard like Looker, you can click and filter but can’t say, why is this metric not changing? What is going on here? Drill down and give me more detail. It's still a manual process,” Bailis said.
“What LLMs and specifically data agents let us do is to have that conversational experience, as Gemini Consumer or ChatGPT do with data on the web, but with a company’s own business metrics. With this integrated stack, we're enabling these agents to apply multi-step reasoning to data.”
Bailis demonstrated the data agent’s capabilities with a call center use case using BigQuery, Looker and Gemini with an agent built to take the data and semantics and perform complex, multi-step question answering and advanced data analysis. This could include requests such as creating a pivot table broken down by call reason – the LLM interprets what is meant by call reason.
“I can go beyond typical questions as well. Looker is largely focused on conventional analytics but doesn’t give insights into why or what's going to happen,” said Bailis.
“Because we're hooked into Gemini, which is very good at coding and understanding data science tasks, I can ask much more complicated questions than I would ask a traditional data analysis tool, like, ‘How is call volume expected to change?’ I don't have to be very precise, I just chat with the LLM. It expands the potential and the ability for people to ask these questions they don't get to ask today.”
Because BigQuery is an integrated data stack, organizations use it to store semi-structured and unstructured data such as text images and video. Gemini is multimodal and can process all of these data types and this means the data agent can process this multimodal input too.
For the call center use case example, a field agent could upload a photo of a router and ask: “What’s wrong with this?” The data agent could interrogate PDFs of manuals for that model to deliver an answer.
Bailis added that while a company’s data and Gemini are hosted on Google Cloud, that data remains within a customer’s environment and Google does not train its models using that data.
“Having data privacy and governance taken care of out of the box has been hugely important in building trust, not just in the quality of the outputs, but also that your data is not going to be sucked up into someone's training,” he said.
Google Cloud has had a public preview of its conversation analytics available since October and promises announcements and customer testimonials at its Next event in April.
You May Also Like