As India pushes forward with its vision for sovereign AI, it must prioritize the release of government-held data to counter the Western-centric perspective of current global AI models, says Rajnish Gupta of EY India.

Ahead of the Union Budget, Gupta called on the government to make extensive public datasets accessible to local developers, enabling the creation of AI that reflects India’s distinct identity. He warned that most popular Large Language Models (LLMs) today are biased because they are predominantly trained on data from the US and Europe.

"One of the big things that we could do a little more in India is make a lot of data available publicly. (Most of) the answers that you get are very Western. They're very much from a US or a European standpoint," Gupta said.

He emphasized that India’s cultural diversity, multiple languages, and religious intricacies are often misunderstood or overlooked by global AI models. ".We have our own culture. We have our own languages, our own nuances… A lot of this data can be generated by the government and made available to the people who are developing LLMs," he added.

Comparing the development of AI to India’s achievements with Aadhaar and UPI, Gupta suggested that treating data and computational resources as public goods could democratize AI in India. He noted that while public investment is crucial for laying the foundation, it should not control the direction of innovation.

Gupta also advocated for a hands-off regulatory approach. He praised the government for not replicating the EU’s AI regulations, saying minimal interference will allow Indian entrepreneurs to experiment and develop culturally relevant AI models.

"Don't tinker with this," Gupta advised. "If you don't license, don't create frictions, don't introduce any new taxes... and just let them operate, then I think somebody sitting over here will create something."