India’s a priority market for Google, excited by tech innovation here: Senior VP Prabhakar Raghavan

Terming the demands for a “pause” in AI development as “uncalled for, as of now”, senior vice president of Google Prabhakar Raghavan said rules to govern AI should be “based on science and a deep understanding of the subject”. He was voicing his opposition to calls for a blanket ban on the technology, which is taking the world by storm. The way ahead to develop artificial intelligence (AI) systems is to “engage with scientists and technologists who understand” the technology behind it, the top Google executive said.

In an exclusive interview with ET, Raghavan – who is a key member of Alphabet CEO Sundar Pichai’s leadership team – said AI regulations must balance innovation and its potential to uplift economies like India. The country remains one of the two priority markets for Google and is adding a lot of innovative products to the company’s repertoire, he added.

Noting that “India is growing really fast, right on a fairly significant pace,” he said the next big startup from India could be the one which caters to the unique needs of Indian users. For instance, startups that solve the problems of logistics for companies delivering to remote corners could gain the most. Google, which has developed a lot of India-first products such as Google Pay, offline maps, flood forecast warning, and map directions for two-wheelers, among others, is now finding ample users for these in other geographies as well. Now, with generative AI coming into the picture, there will be newer opportunities for Google’s products, especially in geographies like India, Bangladesh and others, according to the IIT graduate who is now responsible for Google Search, Assistant, Geo, Ads, Commerce, and Payments products at the Mountain View-based search giant.

“We have fascinating new opportunities in building better Indic language models because the data sparsity is different. We are trying to develop better resources to train across languages,” said the 62-year-old Raghavan.

Google’s focus was not just pure languages such as Hindi, Bengali or Tamil, but dialects such as Hinglish, which were a mix of two pure spoken and written languages, according to the 62-year-old computer science engineer who has a PhD from the University of California, Berkeley. Training mixed language models, as well as pure language models, was also important for Google as search queries from India had shifted from the traditional text and image to a combination of voice and image or just voice.

“In India, about a third of our search queries are spoken. That is in contrast to like three or 4% in the US. So, it’s a 10x increase. That is because people here are culturally really comfortable pulling out the phone and talking to it. In many other countries, people are not comfortable,” Raghavan said, adding that in future, search queries would arise from multiple modalities such as text, voice and image.

“You point the camera and then ask a text query such as if there is something wrong with this plant, how do I fix it? And we tell you to go buy this pesticide, sprinkle on top etc. These are emerging behaviours that drag the core products in different directions,” he noted.

Though Google has found and implemented ample use cases for its generative AI and derived models, Raghavan believes that a pause in the development of the technology is uncalled for as of now. He was responding to a query on the call – by technology entrepreneur Elon Musk – to hit a pause on AI, the Google executive said “there are many prominent scientists who truly understand what these things do and don’t do. My suggestion would be to engage them and try to get from them exactly what they are thinking of. Ultimately, it is a hard science underneath, and it is a branch of computer science.”

Raghavan, who is currently on a visit to India, is of the view that smaller companies in the generative AI ecosystem will make as much of a difference as the larger firms such as Google, Microsoft or OpenAI and that it was not “hopeless” for the smaller players to compete with the bigger ones in the market.

The smaller companies, he said, would make use of the multiple language learning models and tweak them to fit a vast variety of applications. The threshold for such companies was not whether they had trained a language model by now as that is something many people have done.

“It is what novel and interesting user benefit can you derive from it? People are getting creative in many startups everywhere. It is round one of a ten-round match,” he said.