India Stack 2.0: Using AI to Democratize Access to Tech

India’s methodical build of digital public goods, starting with Aadhaar, followed by Jan Dhan (bank accounts for all), and accompanied by affordable mobile connections (courtesy Jio) has led to successes such as the Unified Payments Interface (UPI). The next step in this journey seems likely to be to ensure that India Stack becomes truly transformative using Artificial Intelligence (AI) and reaches India’s 1.4 bn faster. This holds the promise of democratizing access to cutting edge technology for the masses whilst addressing issues which are unique to India e.g., giving low-income earners access to subsidies & loans; providing literacy and skilling from school through to vocational education; and providing SMEs access to voice powered bots which can service customers, take feedback, and improve the underlying product.

“After the reforms and redirection of technological change in the second half of the nineteenth century, a degree of hope seemed warranted. For the first time in thousands of years, there was a confluence of rapid technological progress and institutional preconditions for the benefits to be shared beyond a narrow elite.” – Daron Acemoglu & Simon Johnson in ‘Power and Progress: Our Thousand Year Struggle Over Technology and Prosperity’, 2023

Introduction

The rise of America in the 20^th century demonstrated that enabling infrastructure (railroads, electricity, telegraph, etc) which lowers the cost of transport & communication tends to precede economic growth (see chart above). In contemporary India, as is well understood now, the methodical build of each layer of the India Stack to leverage technology as a public good (the so called ‘Digital Public Goods’) has been tremendously successful – see our previous notes most notably the one dated 7^th February 2024, The Polarization of Corporate Profits in India is Reducing, we had explained that:

“The JAM trinity (Jan Dhan, Aadhaar, and Mobile) has essentially pulled the country into the digital realm, by giving each citizen an identity (using Aadhaar), access to financial means (using Jan Dhan bank accounts), and overall access to the internet. Joining all of this together seems to have created a robust ecosystem that facilitates democratization of information and costless access to the financial system.

Built on top of this is the Unified Payments Interface (or UPI) that facilitates transactions seamlessly and instantaneously across bank accounts [with no transaction costs at all for the payor or the receiver]. After growing exponentially through Covid, as of December 2023, UPI transaction value crossed INR 180 tn (or ~USD 2.17 tn per annum), more than 50% of Indian GDP! This rapid spread of a low cost, easy-to-use digital payments system has conferred many benefits on SMEs and MSMEs including: (1) a reduction in the time taken to settle transactions and, as a result, a reduction in working capital cycles; and (2) elimination of the transaction costs associated with plastic cards, cheques (which entailed visits to the bank) and cash (which also could entail visits to the bank and/or physically travelling to take or make a payment).”

Interestingly, whilst technology in the form of the India Stack has had a transformative impact on the Indian economy, two large constituencies have not yet been able to experience its full benefits – as of 2022, roughly 26%  of the Indian population was illiterate and as of 2024, roughly 47.6% of people in India do not have access to internet.

In fact, the literacy rate of 74% is for people literate in any language (i.e., literate in their respective mother tongues and not necessarily in English). This becomes a real obstacle to digitalization in India as most web applications and data on the internet are developed in the English language, which may serve the purpose for people living in English speaking countries, but doesn’t go very far in a country as diverse as India, where the language and dialects change every kilometer and where only 10.6% of Indians are proficient in the use of the English language.

As the mode of access to the internet becomes increasingly through voice powered bots using natural language processing, the need for adapting the India Stack grows. Secondly, as the use of AI to enhance functionality and improve the user experience becomes widespread, various types of opportunities open up to take the India Stack to the next level.

India stack 2.0: the application of AI to the India Stack

The development of AI in both America and China has been characterized by large tech companies spending billions of dollars in training their bots to solve problems for which users are willing to pay a fee. This “privatization” of AI for corporate profit is a necessary engine for the further development of this nascent technology.

However, what this also means is that AI as developed in the West seems unlikely to address the needs of the majority of Indians who are:

(a) not going to be in a position to be able to afford to pay for AI; and

(b) likely to have use cases which the West won’t have e.g., a Tamil speaking farmer seeking to identify which of the hundreds of central & state governments schemes, subsidies and loans she’s eligible for; or an Assamese speaking 18-year old Scheduled Tribe female high school student in a village near Digboi seeking to figure out which government or private sector scholarship can help her pay her way through university.

Delving deeper into the use cases where AI could help India, it becomes obvious that the Western training data fed into Western AI tools like ChatGPT will render these tools relevant to only that section of the Indian economy whose problems are similar to that of the West e.g., a financial analyst in the Indian stock market trying to synthesize learnings from the last 20 years of published financial data, annual reports and conference call transcripts of a listed company in India; or an executive seeking to book a camping holiday for four in the Himalayas.

To harness the power of AI for the uses cases which are more unique to India, EkStep Foundation, co-founded by Nandan Nilekani, Rohini Nilekani and Shankar Maruwada in Bengaluru is using the tech prowess of its volunteers, who are also stalwarts in their respective fields, to build an AI stack which can be labelled as India Stack 2.0.

To that end, EkStep funded a lab in IIT Madras called AI4Bharat, which among other things, has been responsible for building an open stack in AI, comprising of high quality data, models and tools around Indic languages. AI4Bharat is the digital public good powering India’s National Language Translation Mission or BHASHINI. Bhashini is a DPI (Digital Public Infrastructure) providing language translation AI technology as a service to many Government initiatives. Bhashini currently boasts proficiency in 11 Indian languages in speech and 22 languages in text and is the first and arguably the most crucial step in making cutting-edge technology accessible to all (read more about Bhashini here).

It is important to note here that the Western tech giants, Google and Microsoft, too have AI initiatives built around India’s vernacular languages. However, given that these are listed for-profit enterprises, to what extent they will harness AI for use cases central to India’s economic development is not obvious. Furthermore, given their priorities, the case for them to create something that would address the need for social development of the country isn’t clear either.. The limited role of the Western tech giants in creating India Stack 1.0 suggests that they are likely to remain marginal players as India’s best brains work on creating India Stack 2.0.

An indigenous AI module for India has a few remarkable advantages

Made-in-India AI solutions that are targeted at specific issues – resolution of which are central to India’s development – have the capacity to drive meaningful change by being:

• Use case specific where distinct issues which are specific to India can be addressed and resolved. For instance, if the government wishes to understand how effective a particular scheme like MNREGA has been for a particular state like Madhya Pradesh (or a particular district in Madhya Pradesh), government officials should be able to create a bot that would not only record responses from lakhs of people but also analyze and synthesize the data to help draw actionable insights.
• More inclusive so that everyone can derive benefit from it regardless of which strata of society they belong to and which language they speak. Bhashini’s development is crucial here for to overcome the English language hurdle that the internet currently poses for the majority of Indians. The ‘affordability’ issue around AI is the second leg of the inclusivity challenge that the team at Ek Step is working on.
• Easier to build powerful new applications so that the building blocks of the solution will be available to all (on an ‘open source’ basis), to be leveraged free of cost and as per need. Sunbird (org) is one such digital public good that creates the building blocks of code which anyone can use to create solutions as per the issue or use-case they are targeting. For instance, the QR code on the Cowin vaccine certificate came from a block of code on Sunbird.

These three advantages make the indigenous application of AI to the India Stack an economic necessity.

A few initiatives are already underway

Various teams at EkStep are working with IIT Madras, Bhashini, and many other Government departments, startups and big tech players have already begun working on digital public goods (DPGs) which marry AI with the India Stack. Summarized below are three DPGs which we reckon could have a transformative social and economic impact:

• ‘Jan Ki Baat’ idea (see more here): ‘People plus AI’, the team behind the Jan Ki Baat idea, define it as a game changer which can truly touch the lives of people and make an impact by listening to them, understanding their concerns, synthesizing information, and generating actionable insights for organizations that cater to a very large population (such as the Government of India).

Traditional bots which are currently in use answer questions asked by users (think of the chatbots available on the private sector bank’s websites or apps). Jan Ki Baat reverses this paradigm by asking targeted questions to people, eliciting responses from them conversationally i.e., a bot talks to you as if there is a human being at the other end of the line wishing to understand what issues you faced or what you found really helpful, let’s say, when applying for a home loan under PM Awas Yojana. This shift in paradigm helps the bot get more information – not just what is being spoken, but also the tonality of the speech, emotions (like anger or happiness due to the service provided) and other non-verbal cues. Right now, at a demo stage, the People plus AI team believes this bot and the general Jan Ki Baat idea can revolutionize how the government can assess success of its schemes and enhance them as per on-ground feedback at a population scale. For the corporate sector, such a concept can be used to launch products, collect customer feedback at speed, and adapt the launch plan (or the product) as per the feedback.

• e-Jaadui Pitara (see more here): The Government of India has through the years launched innumerable schemes to improve the education system. Given the practical difficulties of implementation, these plans often stay on paper which get filed away in large and clunky rooms of some government office, never to be looked at again. One such recent policy was the introduction of a Jaadui Pitara or a magic box by NCERT as part of implementing the NEP (National Education Policy)– an actual box filled with puzzles, games, toys etc. targeted towards the goal of learn and play. This box turned out to weigh 11 kg and cost Rs. 10,000+, making it financially and logistically impossible to be procure and be shipped to all the government run schools in the country.

Inspired by this, the EkStep Foundation partnered with NCERT and the Education Ministry to come up with a digital solution in the form of e-Jaadui Pitara, an app and a bot (can be used on WhatsApp) to help make the process of learning fun and easy for all. The bot currently has 3 modules – Teacher Tara, Parent Tara, and Katha Sakhee – each targeting a specific need and demography.

For example, if a teacher sitting in a village in the Latur district of Maharashtra wishes to know how she can incorporate into her lesson play for her students some of whom are visually challenged, she can ask Teacher Tara bot for help (by typing or by speaking) in 11 different languages.

The bot will promptly refer to the existing official and authoritative material on the subject, come up with an actionable way of incorporating play for visually challenged students and will also give reference to the source material. Similarly, parents and teachers can use Katha Sakhee to generate stories for their children by just giving a few character references and morals they wish to instill in their children.

The bot is built using the Open AI stack of AI4Bharat, Bhashini, and Sunbird AI Assistant, and leverages the existing national scale digital infrastructure of DIKSHA (diksha.gov.in) .

The availability of this open AI stack means that the bot can also be repurposed by companies can give their consumers actionable solutions to installation or tech support related problems (for example, instruction to self-install a water purifier).

• Open Network for Education and Skilling Transformation (ONEST) (see more here): The issue of unavailability of information for students seeking to undertake skilling courses or apply for apprenticeships & scholarships is one that most of us at Marcellus have lived through in our teenage years. So is the problem on the supply side with employers unable to reach out to talent all across the country to either fund via their CSR funds or to recruit. ONEST solves this problem to some extent by connecting siloed platforms via an open network. Utilizing the same tech architecture as ONDC (Open Network for Digital Commerce), ONEST is a network that connects the demand and the supply side with each other.

On ONEST, a biotech student in a remote village in Uttar Pradesh can undertake courses relevant for her field to further enhance her skillset. Once she completes the course, she gets a digitally verifiable credential which can be stored on her ID (her identity is verified via Aadhaar and PAN). An employer in Mumbai seeking to employ a biotechnologist can leverage ONEST and find the best student who has aced a practical course on the network. If the employer is also seeking to utilize its CSR funds on such students where it sees potential, it can find them via ONEST and easily keep track, no matter where in the country the beneficiaries reside. This has the potential to democratize access to opportunities and address the permanently pertinent issue of demand-supply mismatch in the labour market.

These three initiatives are not an exhaustive list of areas where AI can be leveraged to enhance the India Stack thereby producing transformative Digital Public Goods. India’s brightest brains are working on more such initiatives to democratize the citizenry’s access to affordable modern technology which can improve social and economic outcomes.

India specific AI is not without its limitations

Even as it becomes increasingly evident that the application of AI to India’s problems holds great promise to transform economic & social outcomes, the challenges of using AI in India are also coming to the fore.

• There is a marginal cost to using AI: Unlike in the case of internet where the marginal cost or the cost for providing a service to an additional person over the internet is close to zero, the marginal cost for providing AI based solutions is not ZERO. To be specific, the marginal cost of AI is the GPU burn or in simple terms the cost of providing computing capacity every single time AI is used. This makes wide-scale implementation of such modules a little trickier than what it has been under India stack 1.0 with a very tough question lingering – who will pay for this technology? To be specific, when an Odiya farmer who is earning a living just above the poverty line, uses AI to identify which subsidy or loan he’s eligible for, who will pay for the GPU burn is something that needs to be worked out.
• Ground up development of use cases: Because India is at a very different stage of economic development relative to the West or relative to its developing peers, the use cases which are required to tailor AI based solutions will need to be created from scratch. Much of the training data might also have to be created from a zero (or near zero) base. The critical use cases that merit the application of AI along with key stakeholders, beneficiaries and service providers will need to be identified, training data will have to collected and then once the bots are implemented to address that India-specific use case, ongoing surveillance will have to be put in place to prevent both ‘hallucinations’ (where AI models start observing nonexistent patterns or patterns that are imperceptible to humans in data often leading to nonsensical results) and misuse of the installed apparatus.
• Need to collect and train vast amounts of data: Because of unique characteristics of a country like India (where language, dialects, sensibilities, and even behaviors in similar situations keep changing with geography), much of the training data will have to collected in the real world (rather than relying on data already available data over web). This is also because, the data currently does not capture everything, and especially for those problems which require the most urgent remediation – for instance, there would hardly be any data surrounding which fertilizer is to be used for a crop, let’s say, bajra or millet, in a small hamlet of a district in Bihar, in the regional dialect that is understood by the people and farmers living there.

Investment Implications

The India Stack 1.0 has already democratized access to the markets and thus democratized corporate profits in India (see our note dated 7^th April 2024, The Underdog Ascends: The Rise of a New Indian Elite). India Stack 2.0 can further this process, effectively reducing the gap between small and large companies when it comes to the use of tech to build competitive advantages. That in turn could make the economy more competitive and improve India’s chances – especially the chances of smaller Indian firms – to build businesses which can compete successfully in the global market. Beyond IT Services and Pharmaceuticals, India has struggled to build an export economy capable of competing on the world stage. India Stack 2.0, if built and executed at population-wide scale, promises to help India build competitive advantages at the economy-wide level.

Nandita Rajhansa and Saurabh Mukherjea work for Marcellus Investment Managers (www.marcellus.in). Amongst the companies mentioned in this note, Microsoft is part of Marcellus’ portfolio. Nandita and Saurabh may be invested in these companies and their immediate relatives may also have stakes in the described securities. The described stocks/securities are for educational/illustration purpose only and not recommendatory. We thank EkStep Foundation for their help and support in explaining their research and work in revolutionizing AI in India.