On the one hand, the government aims to use the platform to “democratise” AI by not allowing any single entity to dictate terms while providing AI-as-a-service to its citizens, institutions and companies to build products that will help society in areas such as agriculture, healthcare, and education. On the other hand, the India AI stack is aimed at helping the country become ‘aatmanirbhar’ or self-sufficient in the AI space to compete with countries including the US and China.
The government plans to adopt the Digital Public Infrastructure (DPI), also known as the ‘India Stack’, approach to build this AI platform. And, with good reason. DPIs are digital networks that help provide citizens with social services.
Over the past decade, the Centre has implemented and scaled DPI in payments, financial services, healthcare, transportation, and digital identities including Aadhaar, Unified Payments Interface (UPI), Open Network for Digital Commerce (ONDC), Account Aggregator (AA), Fastag, and Ayushman Bharat Digital Mission (ABDM).
Union minister for IT, electronics, railways and broadcasting Ashwini Vaishnaw made this blueprint evident at the inaugural session of the Global IndiaAI Summit 2024 in New Delhi on 3 July.
“Whether it is called big tech, (or) whether it’s some cases (where the) government (is) controlling everything, the approach that our Prime Minister has always adopted is that technology should be accessible to everybody. So, the Digital Public Infrastructure is a classic case where no single payment provider, no single service provider has monopoly over the service. The government invests in the platform, and everybody basically becomes a part of that,” Vaishnaw asserted, adding that the government will “adopt the same approach in AI.”
To be sure, DPIs have been built by collaborating with regulators, academic institutions, private sector companies, volunteers, and startups. Yet, the India Stack represents disparate technology products and frameworks that are not collectively owned by one agency.
Aadhaar products such as e-auth and e-KYC are owned by the Unique Identification Authority of India (UIDAI), while eSign and Digilocker are maintained by the ministry of communications and information technology. UPI is owned by the National Payments Corporation of India (NPCI) while the AA framework is regulated by the Reserve Bank of India (RBI).
The government hopes the DPI approach will help it make AI accessible to everyone by discouraging monopolies and ensuring interoperability, use of open source technologies, transparency, inclusion, collaboration, and sufficient guardrails in this space. DPIs could help India become an $8 trillion economy by 2030, according to a February report by Nasscom and consulting firm Arthur D Little International.
DPI numbers show promise. About 1.3 billion, or 95%, Indians have an Aadhaar number. India recorded about 131 billion UPI transactions with a total value of ₹200 trillion in FY24, and UPI transactions now account for 80% of all digital payments in India, according to the RBI.
Touted as the ‘UPI of commerce’, ONDC had 21.5 million transactions across categories in Q4FY24. About 300 million cards were issued under the Ayushman Bharat Pradhan Mantri-Jan Arogya Yojana (AB PM-JAY) scheme as of 12 January 2024.
Will DPI approach work for the AI stack?
While the thinking is right, the approach may take quite some tweaking to achieve the desired results. ONDC, for instance, is built using an open-source, decentralised protocol called Beckn, which is used for location-aware, local commerce.
AI, however, is not a single technology. Broadly, it’s the desire to make machines as smart as human, if not more intelligent (read: artificial general intelligence, or AGI). AI comprises many technologies including machine learning (ML), deep learning (a subset of ML), image recognition, computer vision, natural language processing (NLP), and now generative AI (GenAI).
AI is also used in conjunction with humungous amounts of data generated by other technologies such as sensors, which are part of the Internet of Things (IoT), blockchain, augmented reality, virtual reality, digital twins, 5G and 6G. And all these technologies are continuously evolving.
Hence, building a country-specific AI platform involves many components including massive localised (multilingual) datasets for training, local data centres, a multi-layer cloud services model, a secure and advanced computing and storage infrastructure, specialised hardware, a semiconductor ecosystem comprising fab plants and chip design capabilities, sophisticated software and algorithms, and a workforce of AI researchers, data scientists, and engineers.
The process also envisages AI model creation, testing, and deployment, while taking account of ethical and legal issues such as privacy and intellectual property protection, copyright violation, and bias mitigation. And it will need an effective application programming interface layer for users and companies to tap into for their specific needs. Many of these architectural needs were detailed in a 2 September 2020 paper by the Department of Telecommunications.
But with the advent of generative AI, which unlike traditional AI, can not only do analysis and predictions based on historical data but also create new text, images, audio, video, and even code, the task of building an AI platform only becomes more complex and warrants additional safeguards.
Further, all this can cost billions of dollars, depending on the project’s scale, existing infrastructure, development timeline, and specific use-cases. For perspective, ChatGPT creator OpenAI reportedly spent $540 million on GPT-3, while Google’s DeepMind incurred losses of about £1.6 billion over six years.
Countries embarking on this journey must consider not only initial development costs but also expenses for maintenance, updates, and further research. Additional indirect costs include developing supporting industries, education programmes, and regulatory frameworks.
Regardless, the government can play a key enabler role in developing such platforms to stabilise and build a sustainable ecosystem such as decentralising data centres.
“With increasing data localisation and data sovereignty policies evolving at the state level and even from the perspective of vernacular languages content being routed locally to avoid latencies (delays in transmission) and provide much better user experience, there is an increased need for data centres to be located in tier-2 and tier-3 towns of India. And, the government can play a huge role in enabling that,” said Jayanth N Kolla, founder and partner of deeptech consultancy firm Convergence Catalyst.
He added that when it comes to building large language foundational and frontier models, the government can make the critical and often costly compute infrastructure available for young companies at a subsidised rate, thus “creating a large compute infra and platformising it.”
The government, according to Kolla, also needs to establish and empower a data fiduciary – a neutral entity that ensures the ownership of the data remains with the citizens while the government remains the custodian of the data, thus making sure these datasets are secure and safeguard privacy while making them accessible for training and developing AI models.
“At the AI apps and solutions development layer, the government can create a platform of both potential customers and investors on the one side, and young, innovative startups on the other, ensuring trusted matchmaking and expedited scaling,” said Kolla.
What AI global leaders are doing
The US and China are way ahead in the AI race with both accounting for almost 50% of the world’s 36,000 AI companies, according to a report jointly released by KPMG International and China’s ZGC Industry Institute. The UK, India and Canada come next, with the 2,367, 2,080, and 1,515 companies, respectively (a little over 3,600 in India, according to a June report by Nasscom).
The US and China are also leaders in AI investment, with the former having invested almost $250 billion in 4,643 companies since 2013, according to Stanford’s AI Index 2023. China ranks second with $95 billion invested in 1,337 AI startups during the same period, while India has invested $8 billion ($10 billion till date, according to a June report by Nasscom) in 296 startups, which implies that funding is hard to come by.
Over the past decade, China’s Big Fund has raised hundreds of billions of dollars and acquired stakes in dozens of microelectronics companies. It is pumping in about $47.5 billion into a third investment fund – the China Integrated Circuit Industry Investment Fund – in a bid to reduce foreign reliance in its domestic chip industry.
The US, which has sanctions against China’s semiconductor sector, is investing $50 billion to boost its semiconductor manufacturing capabilities. India’s new investments, on the other hand, will manage to make chips of only 28-40 nanometres, while sophisticated plants globally have moved on to 2-3 nm.
Further, China-based inventors are filing the highest number of GenAI patents, far outpacing inventors in the US, Republic of Korea, Japan and India, which comprise the rest of the top five locations, according to a report released by the World Intellectual Property Organization this month. China accounted for 38,000, or 70% of the 54,000 GenAI patents filed in the decade through 2023, six times more than the US.
The top five inventor locations are China (38,210 inventions), the US (6,276 ), Republic of Korea (4,155), Japan (3,409) and India (1,350). The heartening fact is that India posted the highest average annual growth rate among the top five leaders, at 56%.
On the data front, too, while India may generate 20% of global data, how much of it is good quality data? Many of the 22 official Indian languages do not have digital data, which makes it challenging to build and train AI models with local datasets. Bhashini, a unit of the National Language Translation Mission, has so far spent $6-7 million to collect data from different sources.
How India is keeping pace in AI race
To be sure, India has already taken the first steps with a budgetary outlay of ₹10,371.92 crore (about $1.25 billion at today’s prices) for its AI mission that was approved by the Union cabinet in March. The India AI mission aims to develop a manufacturing base for graphics processing units (GPUs) in a public-private partnership, and multi-modal domain-specific large language models. The government hopes to launch the mission in the next 2-3 months.
Vaishnaw also outlined the government’s plan to invest in building a “public platform” comprising computing power, high quality data sets, a common set of protocols, and a common set of technical and legal frameworks. The idea is to have startups, entrepreneurs, academicians, and people working on different applications across sectors like agriculture, medicine, healthcare, and education, to leverage this common platform and accelerate their efforts, an approach that “is consistent with the last 10 years of Digital India.”
In a bid to present itself in a new avatar, for instance, state-owned TV channel DD Kisan recently presented two AI anchors named ‘AI Krish’ and ‘AI Bhoomi’ to help farmers and villagers with information on topics such as agricultural trends and animal husbandry.
KissanAI (formerly known as KissanGPT) is a multilingual AI chatbot that provides farmers with personalised, voice-based assistance. The startup has also released Dhenu 1.0 (named after the cow goddess Kamadhenu), an agricultural language model with seven billion parameters.
10BedICU, which creates critical care infrastructure in rural and smaller government hospitals, is developing three OpenAI-powered tools: CARE Scribe transcribes doctor-patient interactions into EMRs (electronic medical records), reducing data entry time by over 50% and improving data quality. CARE Device Connect integrates data from incompatible monitors for continuous monitoring. CARE Discharge Summary automates patient records, saving paperwork time.
There would be many more such examples if startups can tap into the AI stack.
Vaishnaw said the government plans to invest in an AI compute infrastructure of 10,000 or more GPUs, an AI innovation centre, and focus sharply on AI skills development as it has done for semiconductors, 5G and 6G development by tying up with universities.
The government also plans to accelerate deep tech and AI financing “because we know that venture capital will come at a point where the returns start being visible. The phase before that is most vulnerable,” according to Vaishnaw.
India’s AI and emerging technologies group has been promoting the use of cutting-edge technologies such as 5G, AI, blockchain, augmented reality and virtual reality, machine learning and deep learning, robots, natural language processing, and quantum computing.
The country has taken a big step towards achieving self-reliance in electronics by laying the foundation stones for three semiconductor facilities in India, worth almost ₹1.25 trillion. These include a fab by Tata Electronics with Taiwan’s PSMC in Gujarat; an outsourced semiconductor assembly and test (OSAT) facility in Assam, also by Tata Electronics; and an OSAT facility by CG Power in partnership with Renesas.
Indian entrepreneurs have released local language models including Bhavish Aggarwal’s Krutrim, Tech Mahindra’s Project Indus, Sarvam AI’s OpenHathi series, AI4Bharat, SML’s Hanooman series, Sutra series from Two AI, and CoRover’s BharatGPT.
India has been recognised in the Stanford AI Index Report 2023 for the release of its National AI strategy (2018), the rising number of AI publications, AI applications, machine learning systems, contribution to the development of large language and multimodal models, GitHub AI projects contributed by software developers in the country, hiring of AI talent, relative AI skill penetration rate (across gender too), private AI investments, and increasing government implementation of AI curricula.
The three countries or regions with the highest AI skill penetration rates were India (3.2), the US (2.2), and Germany (1.7). as of 2022. The numbers would have only improved since then.
Where AI skilling is concerned, the government cannot work alone with startups, according to Kolla.
“It needs to create a robust platform with equal commitment and participation from global big tech companies (that are also operating and vying for their share of the Indian market), large Indian business conglomerates (that bring in the India market knowhow and experience) and startups (that bring in the innovation and agility) for sustained up- and cross-skilling at population scale. And, such a platform, in any market, can only be created by the government,” he asserted.
On the regulations front, India is already a founding member and the Lead Chair (2023-24) of the Global Partnership on Artificial Intelligence (GPAI), an international forum that “aims to bridge the gap between theory and practice on AI by supporting cutting-edge research and applied activities on AI-related priorities.”
India does not have a separate AI regulatory bill as yet. But the country’s Digital Personal Data Protection (DPDP) Act, 2023, which was notified last year, has eased the stance on cross-border data transfer restrictions. Further, the proposed Digital India Act 2023, which aims to replace the Indian IT Act 2000 and also regulate AI, may undergo more iterations before being enacted.
To sum up, India has many tools and skillsets to make the AI platform a successful reality. All stakeholders, though, will do well to realise that Rome was not built overnight.
