Sovereign AI: GovAI emerges as the core to secure economic, national future

Repeat after me: AI is the coming wave, and we are living in the age of AI.

 

I request that you read my first article of the series in Business Standard on January 10, 'Sovereign artificial intelligence: A strategic imperative for coming wave', before you read this piece. I had argued that India—or indeed any country in today’s AI age—must have a Sovereign AI strategy. More than any single technology, AI is both a challenge and a threat to a nation’s economic and national security.

 

Here, I lay out the way to build our Sovereign AI strategy.

 

GovAI + Private AI = Sovereign AI

 

 

AI allows you to do more with less. AI is a layer on the Internet. Since the launch of Digital India, India has deeply digitalised public services and governance—with Digital Public Infrastructure (DPI) and the India DPI stack.

 

DPIs represent vast amounts of data, which can be the building blocks that segue DPIs to what I refer to as GovAI. GovAI (AI deployed in governance use cases) is the killer app for AI. It will impact citizens, who will benefit from the smarter, predictive, and more efficient ways AI can deliver services and solve problems. GovAI is, therefore, the most logical way to build our Sovereign AI capability, which can also be applied to other domains.

 

DPI + AI = GovAI

 

DPI has demonstrated the power of technology as a transformative force. India DPI + AI (GovAI) will be the kinetic enabler of next-gen DPIs, while DPIs can help build GovAI.

 

Data is the fuel to build and train AI models. DPIs have huge repositories that can be transformed into consented personal data and/or non-personal training data for AI models. An India Datasets Programme, consisting of non-personal or anonymised data, is already envisaged in the IndiaAI mission blueprint prepared through a partnership between the government and AI practitioners in academia and industry.

 

Digital India has resulted in a large basket of DPIs—digital payments, banking, insurance, health, education, skills, agriculture/farmers, women, and more. There are over 35 major DPI apps (UPI, Digilocker, Diksha, Umang, Cowin, NHM, etc) and many other data, image, audio, and video repositories, along with several smaller repositories at the central and state government levels (no formal inventory available). Organised into datasets, these can become formidable inputs for training and building India’s AI capabilities. It must be understood, however, that this is neither trivial nor does it have a shortcut road to building these sovereign capabilities.

 

An architecture for GovAI

 

Each of these DPI-derived datasets can be used to train models focused on the use case or public service the DPI delivers. For example, the past and future datasets of Diksha (education platform) can train AI models that predict and design smarter education outcomes analysis and delivery.

 

The GovAI architecture will resemble the DPI framework, where models for specific use cases—small language models (SLMs)—are built and trained with datasets specific to that DPI or domain. These SLMs and/or their datasets can then train and infer from each other, or “co-operatively” build more complex trillion-parameter large language models (LLMs). For instance, learning outcomes of students can be correlated with health patterns, income levels, and districts to enable predictive and planned interventions.

 

India’s DPI architecture is intuitive and elegant because it was never designed as a stack. It began as individual services that organically evolved into a stack. There is work to be done to make it function as an integrated stack rather than as a loose collection of applications. Similarly, GovAI is envisaged as a collection of SLMs solving independent use cases and concurrently evolving into a more intelligent, intersectional LLM or super LLM.

 

I’ll make a bold prediction: the future AI landscape will evolve into one where sovereign AI models collaborate with other trusted AI LLMs to create a “better” or “more trusted” AI. For example, India’s GovAI or Sovereign AI may train and infer with other LLMs, representing diverse datasets, to create even more powerful learning. The path to the holy grail of artificial general intelligence (AGI) will require cooperation and mutual learning agreements between models representing different datasets and people. A robust debate on our approach to Sovereign AI is welcome and is one intention behind writing this series of articles.

 

Govt + Innovators: The partnership for the coming wave

 

India’s Sovereign AI ambition will need the support of Prime Minister Narendra Modi, just as he has championed Digital India and the Semiconductor Mission. The DPI framework grew organically and rapidly after Aadhaar, UPI, and Digilocker were rebuilt. Bootstrapping GovAI will be more difficult, but once the first use case SLMs are demonstrated, the same momentum will occur in other domains as with DPI.

 

A deep, talented, and trusted partnership between the government and AI practitioners in the private sector and academia has to be at its core. This partnership should aim to build and train models, platforms, apps, agents, research, and talent across the full spectrum. DPI spawned a huge innovation ecosystem, far outstripping its original objectives. Similarly, GovAI can create Sovereign AI capability and position India to shape the future of safe, trusted AI, accessible to all in the Global South, while future-proofing our economic growth and national security.

 

---

{Rajeev Chandrasekhar is a former Union Minister. This is the second in a three-part series exclusively for Business Standard.}