The rise of first-generation vernacular AI

Despite hurdles in data management and funding, the rise of vernacular AI in India promises to revolutionise business communication and responsiveness

- VERNIKA AWAL

With 2023 firmly being the year that solely belonged to generative artificial intelligence (AI), it is no surprise that a host of Indian startups, some now fuelled by early-stage venture capital money, are going after creating 'Indian' AI models. Beyond the nationalist noise, each of these startups is looking to capture a piece of a potentially significant pie-that of businesses looking for India-specific AI tools and models.

The central government, too, is in on this game. Last year, Union Minister of State for Information Technology, Rajeev Chandrasekhar, underlined the Centre's work on the India AI Programme, which aims to establish Indic language datasets and indigenous AI compute infrastructure to meet the country-specific AI needs. The Centre, through the Ministry of Electronics and Information Technology (MeitY), also runs Bhashini, a central repository of organised Indic language data.

All of this points to the rise of the first generation of vernacular AI models, one that could either just serve the hype cycle and die out, or power the future of India's local language Internet.

WHY DO WE NEED VERNACULAR AI?

First and foremost, vernacular Al, or India-specific generative Al, refers to the creation of AI models that will cater to Indian languages, and take into account the country's range of biases in terms of culture and sociopolitical factors. This is imperative because the most prominent AI models globally, which include OpenAl's Generative Pre-trained Transformer (GPT), Meta's Llama and Google's Gemini, were trained primarily on a global dataset, which enterprises and the government believe may not adequately represent biases specific to Indian culture, which is important for local usage.