With artificial intelligence (AI) technologies gaining currency and being applied in various business operations, tech firms are now working on developing large language models (LLM) in local Indian languages.

Speaking at the first day of the Global Partnership on Artificial Intelligence (GPAI) Summit, experts and industry veterans said that developing models in local languages will deepen the usage of AI in sectors like agriculture, health, education, and fintech in a much wider form.

“Whether it is agriculture, health, education, we are making sure that we collect the data from the region, for which the application is being developed. So we are ensuring that regional LLMs are not just for Google, but for everyone, using that data,” said Harsh Dhand, head, APAC Google Research and AI Partnerships at Google.

“Most of us have not been able to crack local languages and particularly what we call low resource languages. There are so many languages unique to our country, from which datasets can be built,” said Akrit Vaish, co-founder and CEO of Jio-owned Haptik.

Vijay Shekhar Sharma, CEO of Paytm said, that since there are several Indian languages, developing AI bots on them can see the education sector benefit immensely.

The Indian government is already working on making available non-personal datasets for companies to train their AI models, under the India AI programme. In terms of regional language repository, India’s Bhashini platform can help in developing LLMs, experts said.

Currently, the biggest source of information for generative AI platforms is Wikipedia, which is largely in English language.

Representatives of global tech giants such as NVIDIA, Google, OpenAI, along with United Nations Educational, Scientific, and Cultural Organisation (UNESCO), said that local LLM models would reduce the gaps between leading AI economies like the US, China, and the global south and would enable cross border flow of data to train models.

“The truth is that the transformative power of AI is still highly concentrated in a few hands, few companies, and few countries. So, AI is highly concentrated in terms of development, which may lead to discriminatory patterns,” said Gabriela Ramos, assistant director-general for social and human sciences at UNESCO.

According to Ramos, in the last decade, there has been a 28 fold increase in investments in AI, and this has occurred only in the US, China, and some parts of the world.

For countries in the Global South, including India, to bridge the gap between leading economies in terms of AI development, they will have to invest in their own technologies so that they can up their representation, Ramos said.

Talking about democratisation of AI, and making it more unrestricted, affordable, and user-friendly, John Ashley, director at NVIDIA AI Technology Centres said, “in order to address inequality of AI, we need to take action instead of focusing on having a perfect plan first, so that there are no AI haves and have nots”.

Further, to bring equality for AI use and leverage it for the masses, companies like OpenAI want the government to start using the tools first.

“My biggest recommendation for GPAI would be that governments should start adopting these tools and integrating them. This will strike the balance in terms of use of tools and also help people to leverage the technology,” said Anna Makanju, vice president of global affairs at OpenAI.