Govt invested in indigenous AI models to tackle bias: MeitY Secy

The government has invested in developing indigenous and sovereign artificial intelligence models to address concerns related to bias and to ensure that India has AI systems primarily trained on Indian data, a senior official said on Tuesday.

S Krishnan, secretary in the Ministry of Electronics and Information Technology (MeitY), said that sovereignty is a very important issue, and India needs to have AI models where "nobody else has the kill switch, and it was always with us".

According to him, a country of the size of India needs to necessarily have control over the way AI models function.


There was a debate initially whether India should actually invest in creating indigenous or sovereign models or should it use open source models which have been created elsewhere in the world, because creating an indigenous model was felt to be expensive and time-consuming, the secretary said.

But after some discussion and debate, India went ahead and created its own AI models for certain important reasons.

"First and foremost is to address the issue of bias...Making sure that we have models which are based primarily on Indian data and data which is created in a multiplicity of Indian languages. That is very critical," Krishnan said.

He was speaking at a session in the AI Impact Summit.

"So in order to have truly represented India's cultural and linguistic diversity and represented adequately Indian data, it was important and critical to have a model of our own," he said.

He said the second reason is India's linguistic diversity, as people would prefer to access AI in their own languages.

It is important to have voice first and voice-text models to come out of India. India's sovereign model is BharatGen.

Launched in October 2024, BharatGen is a government project to develop a sovereign AI model which will provide services like Automatic Speech Recognition (ASR) models and Text-to-Speech (TTS) models for Indian languages.

He added that India's goal is not necessarily to build large language models (LLMs).

The aim, he said, is to create things which can be used by people in their own language and in areas such as science, agriculture, and manufacturing.

"For these reasons, it is critical to fund these things," the secretary said.