Last year, AIM probed when India’s ‘UPI Moment’ in AI would arrive, pointing towards the potential for a locally contextualised AI model as a digital public good. Little did the world know that the work had already started and that it was arriving sooner than expected.
“We will see a UPI moment in India for generative AI through the usage of AI, more than the production of AI,” said Pramod Varma, expressing uncertainty over whether another OpenAI will emerge from India but but wished AI companies in India (the likes of Krutrim AI, Sarvam AI, and SML) well on their effort in building advanced foundational models.
Varma is the former chief architect of Aadhaar & India Stack and the founding architect of the Beckn Protocol. He is currently serving as the chief technology officer at EkStep Foundation and building digital public infrastructure (DPI) for generative AI in India through People+AI’s initiative of Open Cloud Compute.
A few days ago, OpenAI vice president Srinivas Narayanan expressed his excitement about potential collaborations and praised the diversity in AI applications being built in Bengaluru. “It is inspiring to see the ambition in their thinking and the diversity in the applications,” he said.
In an exclusive interview with AIM, Varma showed a similar excitement. He believed that India’s ‘UPI moment’ in AI would arrive quicker if we maximised the utility of existing models to address challenges instead of building foundational models like GPT-4 from scratch.
“I have a feeling that we will see generative AI penetrating through art, media, the movie industry, transport, commerce—everywhere,” said Varma, adding that he sees no reason for it not to happen because we currently live in a very low-performing equilibrium.
A recent report showed that only 22% of Indians are leveraging generative AI for work purposes in healthcare and research, while 76% plan to use it in the next two to five years. “I can only say there is no human being who is not working on generative AI,” said CP Gurnani, the former Tech Mahindra chief, at AIM’s MachineCon GCC Summit recently.
“The Indian path in AI is different. We are not in the arms race to build the next LLM. Let people with capital, let people who want to pedal ships do all that stuff… We are here to make a difference and our aim is to give this technology in the hands of people,” said Nandan Nilekani, the co-founder and non-executive chairman of Infosys.
On the contrary, Ola Krutrim chief Bhavish Aggarwal holds a different opinion. He recently said that India has the largest number of developers, silicon designers, data, and IT services in the world. “India can do to AI what China did to manufacturing,” he added, saying that it won’t automatically happen unless we make it happen.
DPI to the Power of AI
In a recent interview, Nilekani discussed how India has benefited immensely from DPI, Aadhaar, and UPI and how AI could lead to similar transformative changes.
He explained that applying AI at a population scale could have an even greater impact. “We call this DPI to the power of AI,” Nilekani added, emphasising that AI can elevate DPI to the next level.
Varma explained that AI is already integrated into GST and Aadhaar. “AI is used extensively within GST to detect circular loops, fraudulent patterns, and process invoices,” he said, highlighting its widespread application.
“You can now introduce voice interfaces. If you can do voice-based payment, UPI will go through the roof. We will see AI play into DPI to further expand and bring efficiency and expansion of DPI,” he continued.
Varma believes that India’s future is voice-based AI. “Indian entrepreneurs should really look at voice as a completely new human-computer interaction method. It could be very powerful, and I think it’s gonna happen because voice is natural to humans.”
In a previous interaction with AIM, Sarvam AI said that it is currently working on a voice-based Indic LLM and plans to release it this year.
Meanwhile, the IISc’s ARTPARK (Artificial Intelligence & Robotics Technology Park) is preparing to open-source 16,000 hours of spontaneous speech data from 80 districts as part of Project Vaani, under the Ministry of Electronics and Information Technology’s flagship AI initiative, BHASHINI.
India’s Tryst with Generative AI
In India, approximately 10 AI startups are currently engaged in developing LLMs from the ground up. Recently, Gurnani unveiled Project Indus, a ChatGPT-like product developed for less than $5 million.
Despite initiatives like Krutrim AI and Hanooman, these chatbots have yet to achieve the sophistication of GPT-4o or Claude 3.5.
According to Similarweb, Krutrim saw a decline in user numbers from 498.7K visits in April 2024 to 309.2K visits in June 2024. In contrast, SML’s Hanooman had only 80K visits in June 2024, but the company aims for Hanooman to reach 200 million users within its first year of launch.
“I feel it’s unfair to tell them that it’s not working. It takes time. Every business goes through the cycle of production, distribution, and unlocking value for their customers,” said Varma.
Citing healthcare, Varma said that hundreds of enterprises won’t rely solely on one advanced model such as OpenAI’s GPT-4 or Anthropic Claude 3.5 Sonnet, and said: “Both specialised and generic models will coexist and that one is not a replacement for the other.”
“It’s the wrong dream to believe we only need one API. We will need both mega models aiming for AGI and specialised micro models that can integrate into today’s workflows.
“These models will bring enormous efficiency, accuracy, and cost savings to various sectors—from industrial to healthcare,” he said, giving the example of a specialised model that understands Indian laws and can draft contracts in Indian languages.
AWS’ chief medical officer Rowland Illing also told AIM how AWS is looking to democratise access to foundational models, saying, AWS wants to achieve that with partnerships like Anthropic, offering early releases of models such as Claude 3.5 through Bedrock, providing users with a wide selection of models and tools for secure and efficient generative AI applications.
During the pandemic, Co-WIN was built natively on AWS, supporting 2.2 billion vaccination doses across the country. DigiLocker, also powered by AWS, saw the transition from paper-based certificates and documents to digital storage.
