Tech wants govt to prioritise access to data under IndiaAI Mission

Startup Mahakumbh Event 2024: Prime Minister Narendra Modi said India should not let the AI opportunity go

The government's IndiaAI mission takes an all-round approach in supporting the emerging technology's development in the country, with plans to fund high-end chips, train students in AI, back startups in the sector and encourage use-cases in areas like agriculture and health.

One area that the tech industry hopes the government would put greater focus on, though, is to prioritise the creation of a framework to facilitate the availability of open, anonymised data in the country.

Experts say that new chipsets and techniques are evolving fast, and if processes for data collection are not sped up, India would be left behind.

Data is the fuel on which AI models run. The more diverse and plentiful the data is, the better the model is at identifying patterns, making decisions and offering predictions. However, most current AI models in India are trained on data from other countries, for example, OpenAI's ChatGPT.

This is where training AI models on Indian datasets is important, as it can lend local context and also help in improving accuracy of generative AI applications. The work on gathering Indian datasets is ongoing, and it would take time for a country as diverse as India to have expansive datasets.

"If the government can push the availability of open datasets in various domains on a priority, it’s a good thing,” said Pramod Verma, former chief architect of India's digital identity programme Aadhaar. Also the chief technology officer of the Nandan Nilekani-backed EkStep Foundation, Verma said the government has to take the task to completion.

“If they (government) leave it to chance, it may not happen, or if it is too delayed also, it may be of less value,” Verma told Moneycontrol. “So it has to be done fast and it’s a good thing if they (government) can pull it off.”

Verma has many feathers to his cap including of being the co-creator of the open-source Beckn protocol, which is the base for India’s Open Network for Digital Commerce (ONDC) and Namma Yatri, a ride-hailing service app.

In fact, the concept of an India Datasets Platform (IDP) was floated in October last year by the government-created committee of experts mandated to come out with a comprehensive roadmap for a policy push to AI. The conceptualised platform will make all discoverable and usable datasets available for research and to startups with role-based access.

Also read: Inside India’s grand artificial intelligence plan to add $967 billion to economy

“Technology will change in six months; it will get outdated. So we have to think of using our exchequer money wisely and how it will go the longest distance,” said Tanuj Bhojwani, head of PeoplePlusAI, which is a part of the EkStep Foundation, the digital-focused not-for-profit organisation.

Experts say the usefulness of anonymised and concerted datasets cannot be understated. Towards this end, some work is in progress in India. For example, an AI language project by the government called Bhashini has funded and mandated over 70 research institutes to collect open-source data.

The data thus collected by institutes such as IIT-Bombay and IIT-Madras, on behalf of Bhashini will then be trained to build use-cases.

Solving the compute problem

After datasets, the other hindrance to building home-grown foundational models is compute capacity. One of IndiaAI mission’s focus is to make available over 10,000 graphics processing units (GPUs).

GPUs, primarily designed to handle rendering of images and videos quickly and efficiently, are the necessary hardware on top of which AI models can be trained. However, experts are apprehensive whether 10,000 GPUs are enough to train the ocean of data that will flow in for building such models in India.

“My estimate is that 10,000 GPUs is just the tip of the iceberg, we are going to need much more than that,” said Satya Gupta, member of the Confederation of Indian Industry’s (CII) national committee on electronics manufacturing. Besides, associating just a number with the GPU will not give the right answer, as the capacity of each GPU should also be taken into consideration, Gupta explained.

For reference, Facebook’s parent company Meta owns over 340,000 Nvidia H100 GPUs, which is among the fastest GPUs. Meta plans to increase this number to 600,000 in the near future.

Regardless, Gupta said it is a good start but India will need much more than the present number in the future.

“In a sense, we are a GPU-poor nation… We need more access to compute and for these large language models and, in general, AI, we need some pretty heavy-duty compute infrastructure,” said Mahesh Makhija, technology consulting leader, EY India.

Ajai Chowdhry, cofounder of HCL Infosystems, told Moneycontrol that unlike non-AI tech products, the “magic” of AI is heavily dependent on developing models that in turn require “costly” computing.

“We need to do more of the same in the near future to help India become an AI product nation,” said Chowdhry, who is also the founder and chairman of Epic Foundation. Chowdhry is also a member of the advisory board to the Ministry of Electronics and Information Technology (MeitY).

Currently, American multinational NVIDIA is the largest manufacturer of GPUs, followed by other players such as AMD and Intel. Union minister for electronics and information technology Ashwini Vaishnaw recently told Moneycontrol that the government is in discussion with players like NVIDIA, which have committed to support the AI mission in a very “big way”.

Furthermore, experts said it is necessary to provide access to computing resources, techniques and building capabilities within the tech ecosystem. And while the government can play a significant role in facilitating access to compute and datasets, the onus also lies on the tech industry to develop its capacity.

"Finally, the tech ecosystem has to create their own capacity, they have to recruit (and) train, the government can't solve all that for everybody," Verma said.

The government has to provide open and anonymised datasets, experts reason, as private players may not be in a position to do this effectively and efficiently. When asked whether provisions already exist in the IndiaAI mission for anonymised data collection, Gupta said they do, but that the government needs to act on them.

“Suppose I just make a comment in the policy that we will make available health datasets, it won’t happen with just that statement, the mechanism has to be created to collect this data,” Gupta said.

Verma concurred, saying, “It will help the industry access them, train their models.” It will boost various AI applications in sectors such as agriculture, healthcare, etc.

Also read: Through IndiaAI, we will develop indigenous AI foundational models: Rajeev Chandrasekhar

Additionally, experts agreed that the focus on foundational models designed for Indian languages and datasets will be a significant step towards creating AI solutions tailored for the country's specific needs.

"This can lead to the development of innovative AI solutions that cater to the diverse needs of Indian businesses and consumers," said Venugopal Ganganna, chief executive officer of AI-powered digital marketing company Langoor Digital.