• by Laurie Sullivan  @lauriesullivan, January 13, 2025

Google will release one of its more challenging projects this year -- a way for companies to adhere to global regulations and policies related to personally identifiable data. 

The project centers on its BigQuery system that processes and analyzes multimodal data inputs and outputs, supported by artificial intelligence (AI) that can reason and find the best way to carry out the task.

National and multinational companies will have access to this automated system. It will rely heavily on autonomous agents that can decipher policies and restrictions in any region worldwide.

These AI agents will make decisions in milliseconds based on privacy regulations and policies, identify device locations and determine the exact information that consumers are willing to share.

“We are building toward this vision,” Yasmeen Ahmad, product executive for data, analytics and AI at Google Cloud, told MediaPost. “Today, we see companies try to manually adhere to policies and regulations. It requires a human to sit there and interpret all the policies and code them back into SQL, Python, and Spark.”

With these new capabilities -- including some launched last year -- Google is building governance rules and policies that the platform will manage automatically.

The foundational building blocks were launched in 2024, with plans to roll out autonomous models this year. BigQuery will become the heart of the project. 

“We are able to detect PII and access data for regulations and policies,” Ahmad said. “Regulations will evolve, and it will lead Google to autonomous governance.”

Many of the Google Cloud enterprise offerings originated in the technology infrastructure that runs consumer products like Search, YouTube, Gmail and more. They were packaged into enterprise applications in Google Cloud, which is where BigQuery, a database and analytics platform, was born.

Ahmad mentioned DeepMind’s AI reasoning and thinking models as a key element of Reasoning that puts Google that much closer to making all this a reality with multimodal data. These agentic capabilities built into a privacy system will sit in BigQuery. It will act autonomously, make decisions, and take actions to achieve specific goals without constant human intervention, such as protecting consumer data.

A retailer would connect its data into Google’s multimodal foundation and use BigQuery’s governance capabilities to activate autonomous agentic agents. It’s not clear if the system would identify the device’s IP address to determine the regulatory policies in that part of the world.

“It’s a vision coming to reality in 2025,” she said.

BigQuery will have autonomous AI capabilities such as automatic data quality checking, and monitoring datasets for personally identifiable information (PII) information. It has become an intelligent data platform to support traditional uses.

It will draw from the same types of technologies Google has used for years to pull from real-time ad serving, discounts and offers from merchants, customer support, and retailers. These companies have relied on immediate processing times -- fast enough to make the process act in real time.

Ahmad said Google uses a variety of its CPUs like the Tensor Processing Unit (TPU), a custom-designed application-specific integrated circuit (ASIC), to run real-time datasets in BigQuery, capable of microsecond response times.

Google has a variety of processing units that sit on top of BigQuery to support all types of businesses from advertising and healthcare to package delivery. Inputs include all types, too, including engines such as SQL and Python, programming languages used for data processing, analysis, and machine learning.

UPS, for example, built a live app that runs real-time data that enables the company to predict parcel theft with a platform called Delivery Defense. The platform, which sits on top of BigQuery, uses algorithms to assign a delivery score, so when LLMs run it benefits from Google’s multiple types of TPU technology.

Ahmad called it “the economy of now,” the ability to give customers answers about a product or service — personalized and in the moment. Not just process text, but video, audio and other multimodal inputsl and then deliver them back to the consumer.

Multimodal data is often called “dark data,” because Google, until now, did not have the ability to analyze the data.

“Business moves so fast, so you need to stay with the customer,” she said. “Multimodal AI is important to make that change.”