Google Cloud Next '24: AI regulation needs to be technologically feasible, says CEO Thomas Kurian

Google Cloud CEO Thomas Kurian (extreme left) along with other executives addressing a media roundtable at the Cloud Next '24 in Las Vegas on March 10

As various governments across the world deliberate on introducing legislation to regulate technologies such as artificial intelligence, Google Cloud CEO Thomas Kurian on March 10 said that there was a need to ensure that any policy introduced is implementable technologically.

Speaking at a media roundtable at the Google Cloud Next '24 in Las Vegas, Kurian said, " In discussions with many governments around the world, United States, Europe, Latin America, on policy related models, we've always said welcome to policy, and regulation and models."

"But we need to make sure that whatever is defined in policy can actually be implemented technologically. Because eventually, you know, all this is piece of software, and you would want to put a policy on there, that can be implemented on software," Kurian said.

The Google Cloud CEO's comments come at a time when several governments, including India have attempted to regulate AI, in a bid to keep a check on harms related AI and Gen AI, for example, deepfakes and other AI-generated misinformation.

The Indian government for instance, a month ago, introduced an which said that artificial intelligence platforms will have to seek its permission before launching an AI product in the country. However, after widespread backlash from the industry, the IT ministry scrapped the said advisory.

Apart from that Kurian during the press conference, highlighted the company's rapid growth, platform's choice and flexibility in deploying models, while emphasising the importance of openness in working with enterprises.

Kurian also spoke about Google Cloud's data centers, and how the company is currently also focusing on working on newer designs of data centers that are purely optimised for artificial intelligence.

"We are also working on newer designs of data centers that are purely optimized for AI. And they are designed to lower substantially the cost of training models. There are many, many things we've done and pioneered, for example, our systems are water cooled. And it allows us to get 35 to 40% of throughput for the systems," Kurian said while responding to a query from Moneycontorl.

"And it's because of that we're also optimizing data center design itself around your systems. So we'll do a lot you'll see a lot from us as we go forward with that," he added.