New Delhi, February 18 Artificial intelligence must be safe, responsible, and inclusive, especially when deployed in sensitive areas such as healthcare, education, and applications focused on children, said AI4Bharat Founder Mitesh Khapra on Wednesday.
It should work the same way for everyone, regardless of their demographic background and language, Khapra said during a conversation at the ongoing AI Impact Summit 2026 here.
He also raised concerns from an Indian perspective about AI, such as the sovereignty of data used in services like education and healthcare.
"If we don't have sovereignty, if we don't have control over what our AI is, especially AI driving critical applications in the government, healthcare, and education, then we have a challenge," said Khapra, who is also an Associate Professor at IIT Madras.
"If we don't have sovereignty, we don't have control over what goes into training those models," he cautioned, noting the risks of relying heavily on foreign-built solutions.
AI4Bharat is a research lab at IIT Madras dedicated to advancing AI technology for Indian languages through open-source contributions.
Khapra emphasized that reliability and accuracy are critical when AI systems are used by professionals or vulnerable groups.
"AI has to be safe. It has to be responsible," he said, underscoring the importance of trust in technology.
He added that inclusivity is equally vital, with AI expected to work uniformly across demographics, languages, and districts.
Highlighting the Indian context, Khapra pointed out that dialects such as Haryanvi, spoken across an entire state, deserve mainstream support alongside widely recognized languages like Gujarati or Marathi.
Acknowledging that several Indian AI models have been deployed in recent weeks, Khapra stressed the need for a long-term national agenda.
Khapra mentioned the Bhashini initiative, which aims to expand beyond the current 22 languages. He said in the second phase, Bhashini expects to capture many dialects such as Maithili, Haryanvi, Chhattisgarhi, Khariboli, and Kanoji, among others, to ensure inclusivity in India's linguistic diversity.
He emphasized that regional collaborations will be crucial for collecting localized data, while a centralized mechanism will be needed to standardize processes across languages and dialects.
"We hope to move beyond the 22 languages and focus on dialects as well as more languages," he said.
Expressing similar sentiments, Elevenlabs Co-Founder Mati Staniszewski said India faces a challenge of supporting 22 languages from a voice perspective.
"You really need to build the architecture from the start to support it," he said, adding that it currently supports 11 languages across 11 labs, and hopefully all 22 will be supported later this year.
ElevenLabs is a Free AI Voice Generator & Voice Agents Platform.
"We hope that the future will be different, where you will be able to enjoy all that content, all those stories, all those movies in the original language, with the right voices, the right intonation, the right emotion, and that's where Elevenlabs was born," he said.
