India should adopt, adapt AI before frontier development: World Bank

World Bank's Vincent Belinga advises India to prioritize adopting existing AI and adapting open-source models for local needs. He recommends a three-stage approach, focusing on broad adoption across sectors before investing heavily in new frontier models.

India should focus on expanding the use of existing artificial intelligence applications across the economy and selectively invest in frontier AI capabilities instead of making broad investments in developing advanced models from scratch, Vincent Belinga, Senior Economist at the World Bank, told ANI in an exclusive conversation.

A Three-Stage Approach to AI Development

Belinga said India should follow a three-stage approach to AI development: adoption, adaptation and frontier development. "One is adoption, meaning using existing application. The second is adaptation, meaning taking open source AI and adapting them to local conditions. And then the third stage is frontier model, meaning developing the next capability from scratch," Belinga said.

Prioritise Adoption and Adaptation

He said India's immediate priority should be to deepen AI adoption among firms and workers, as adoption is increasing but remains limited across the economy. "What we recommend for India is doing the two. Meaning, deepening adoption so that many firms, millions of firms, millions of workers use AI more productively because so far we see that adoption is increasing but not broad based," he said.

Belinga said India already has promising AI applications in areas such as education and healthcare, but many of these initiatives are either limited to specific states or remain at the project stage. The focus, he said, should be on taking such applications to a larger population. "Some promising small AI in education, in health that are either state specific or that are still at project level. So, what we need is to scale up these initiatives so that they are broadly shared across all the states or enriching the broader population," he said.

The World Bank report highlights local AI applications in India, including Bhashini and AI4Bharat for multilingual access to services, DIKSHA for personalised learning, KATHIR and the PM-KISAN chatbot for agriculture, and AI-assisted tuberculosis screening from chest X-rays.

Driving Broad-Based Productivity Gains

Belinga said broad-based deployment across sectors would be more important for India than relying on productivity gains from only a few industries. "AI is going to raise productivity," he said, adding that India needs to ensure that AI is adopted across sectors and by workers.

The World Bank estimates that 14 per cent of Indian workers are in jobs where AI can complement their work and raise productivity. It also estimates that AI could add 0.1-0.8 percentage points to annual labour productivity growth over the next decade, depending on the pace of AI deployment and complementary investments.

Selective Investment in Frontier Development

Belinga said India should also be selective in developing frontier AI capabilities because such development requires significant resources. "Frontier development is very resourceful. I mean, you see what happening in the US, you have billions of dollar to be invested in frontier development. And given that India has competing development needs, what we suggest is to invest selectively in frontier capabilities where India has a comparative advantage," he said.

He said the broader objective should be to deploy smaller AI applications across sectors such as education, healthcare, agriculture, finance and IT so that productivity gains are spread across the economy rather than concentrated in individual sectors. "So you have to make sure that broadly, AI is being adopted and you deploy these small applications across all the economy so that we can have larger productivity gain instead of having productivity gain at sectoral level," Belinga said. (ANI)

(Except for the headline, this story has not been edited by Asianetnews Editorial staff and is published from a syndicated feed.)