Sovereign in AI: Why India must have its say in emerging technology

Sovereign development of the technology can be used for solving socioeconomic issues

Arvind Krishna, chief executive officer (CEO) of IBM, made headlines when he said recently that all nations should consider having their own sovereign capability in artificial intelligence (AI).
 

As the world debates the possibility of AI revolutionising socioeconomic areas ranging from healthcare to transportation and national security, the technology has stirred curiosity and caution in equal measure.

Krishna, while visiting India, said that countries should develop their own AI large language models (LLMs), pre-trained tools that understand and generate text in a human-like fashion. GPT-3 and GPT-4 from OpenAI, LLaMA from Meta, and Google's PaLM2 are examples of LLMs.

“In many nascent technologies, you often need the government to step in first before others will follow. The government should set up a national AI computing centre, some pools of data can be shared – maybe agriculture, health,” said Krishna after meeting senior Indian government officials in August.

What is sovereign AI?

“A sovereign technology is the one for which a country owns end to end, including intellectual property. For instance, India should have the capacity to run any AI model that we want on computing resources that reside within the borders of India, and are within the regulatory control of India,” said Balaraman Ravindran, professor of Computer Science, Indian Institute of Technology Madras. He heads the Robert Bosch Centre for Data Science & Artificial Intelligence and the Centre for Responsible AI at the institute.

The European Union (EU) is building sovereign AI capabilities by framing a data strategy that facilitates the development of interoperable cloud computing services. It plans to invest €2 billion in the European High Impact Project to develop data-processing infrastructure, sharing tools, architecture and governance mechanisms. 

The purpose of sovereign AI capability is that a country should not be reliant on somebody else for supplying data, intellectual property (IP), and cloud services. One other argument in favour of states building their own AI systems is that most large datasets are owned by giant technology firms, which gather information by scraping the internet.

Ravindran said India has already put into motion multiple initiatives to develop sovereign AI technology. “We are looking at building datasets not just under the national AI programme, but (by) various ministries.” 

The Ministry of Electronics and Information Technology (MeitY) has launched the National Programme on AI (NPAI) to develop computing systems, drive domestic IP innovation, and use datasets for governance. It has consulted academics and industry stakeholders.

The Indian Council of Medical Research (ICMR) is collaborating with the Indian Institute of Science Bengaluru to build datasets for healthcare applications. 

The NITI Aayog has suggested that the government set up an AI-based cloud computing infrastructure known as AIRAWAT (AI Research, Analytics and Knowledge Assimilation). This platform will assist in research and developing new technologies for solving business and governance issues.

The infrastructure will use high-performance supercomputing to create new applications for healthcare, agriculture, weather forecasting and other sectors.

AI for governance

The central government is using AI for applications of India Stack and language model Digital India Bhashini. It plans to use AI-driven algorithms to analyse data generated by users of applications like Aadhaar, Unified Payments Interface, and DigiLocker. 

The government will spend some $200 million (around Rs 1,635 crore) to develop an AI ecosystem and make e-governance platforms more intelligent, said Rajeev Chandrasekhar, minister of state for electronics and information technology, earlier this year at the Business Standard TechTalk conference. The government also plans to set up three centres of excellence (COE) for AI. 

Experts say sovereignty in AI should be viewed in the context of building domestic capacity and reducing reliance on a private big technology firms. Such capacity is needed due to concerns about how LLMs like GPT-4 were trained and what information they were fed. 

“If you don’t own and somewhat control your data in a country, can you have sovereign AI? Because it is data that gets converted in AI and data sources are not owned. The other part is sovereignty over computing infrastructure. Nowadays, computing demands have become so intensive that you can’t buy it, so it will turn into a grid, similar to electricity. Today we are dependent either on Amazon or Microsoft for this,” said Parminder Jeet Singh, executive director at IT for Change, a non-profit.

Singh said the proposed COE should focus on building a public computing infrastructure that is available to small businesses.

“The government of India’s recent data infrastructure policies are largely inclined towards an assumption that technology companies will share their datasets voluntarily. But as Kris Gopalakrishnan's committee suggested, you have to select some important datasets which need to be mandatorily shared (for domestic use). The government has to play a big role in mandating this data sharing,” he said . 

In 2020, a government committee headed by Infosys co-founder Gopalakrishnan suggested that non-personal data generated in India be allowed to be harnessed by domestic entities.
 

The government should create a framework where data and computing infrastructure are available to everyone by sharing, said  Singh.