Even as the debate rages in the country about whether India should build its own Large Language Model (LLM) or use the already existing models to develop applications, the government would like to follow a three-pronged approach. “Our approach will be to promote building applications because that’s where the benefits are realised to build on top of existing open AI models and open source models of AI, adapt them, train them on better Indian data, and build on India’s own knowledge to create a foundational model,” S Krishnan, Secretary, MeitY, has said.
He said the emergence of DeepSeek gave the confidence that building a foundational model can be done with frugal innovation.
DeepSeek and data security
“If you use DeepSeek on an app or through their portal, the data goes to China. One company has already deployed it on their servers in India, and they say that when it is deployed on Indian servers, the data stays put in India, but the issue is being examined,” he said.
Krishnan sees India playing a significant role in AI development in the Global South. He emphasises that India’s focus on the inclusive use of AI resonated at the recently concluded AI Action summit in Paris.
Krishnan said AI offers India an opportunity to innovate and leapfrog in its development, similar to what it achieved with the IT revolution and mobile technology.
“France views India as a key player in AI, distinct from the United States and China. There is a ‘third way’ of doing AI and India has a role there,” he said.
On regulation, he said the government believed in enabling regulation rather than restrictive regulation and would look at what parts of regulation hinder acceptance or the rollout of AI.
He said the global mood on AI has shifted from concerns about safety and security to the possibilities of innovation and its potential impact on the economy.
Replying to a question on the likely impact of the US and the UK refusing to sign the Paris declaration, he said that the US did not sign the declaration due to concerns about the diffusion of AI technologies and because the news US administration is still figuring out its pathway, while the UK government has changed since the Bletchley Park summit in 2023.
“It is not an all-inclusive summit and I wouldn’t treat it like some veto in a UN Security Council. I don’t think it’s the end of the world. There would be more opportunities and by the time we get to the Delhi or the India summit sometime later this year, maybe we’ll get everyone to sign it,” he said.
AI compute
He said there are two approaches to AI compute – a viability gap funding-based approach to building a new AI facility and a voucher-based approach, where the market for use of AI compute facilities is underwritten, but the private sector builds it.
The voucher-based approach was chosen because many private companies had already started installing GPUs, and a government subsidy for some GPUs would make the market unequal. Bids for about 18,000 GPUs are already in progress in the first instance and are updated every quarter.
Job losses
On the issue of job losses due to AI, he said India’s situation is little different from its peers in the West. “Unlike some Western countries, India doesn’t have as many white-collar jobs that might be immediately threatened by AI,” he said.
He said there was a debate in India about whether to build a foundational model at all, but after consultations, it was decided to proceed with an open-ended RFP.
