Generative artificial intelligence (AI) research projects are on the rise in top engineering colleges in India, with a particular focus on creating tools similar to OpenAI’s ChatGPT but in Indian languages. Even as technology giants like Microsoft and Google promote generative AI platforms like ChatGPT, Bing, and Bard, Indian researchers are grappling with challenges such as the availability of ample data in Indic languages, the high cost of projects, and the scale of computing power needed.
Researchers at institutions such as the National Institute of Technology (NIT) Rourkela and the Indian Institute of Technology (IIT) Madras are leveraging language models, specifically the transformer architecture, for various tasks such as data classification, question answering, machine translation, and building chatbots. While global platforms largely operate in English, Indian researchers are focusing on languages like Hindi, Bangla, and Kannada, creating models that can process questions in these languages and generate output in English. These researchers have achieved impressive scores on the industry-standard BiLingual Evaluation Understudy (BLEU) test, with NIT Rourkela scoring between 25 to 30 on Hindi to English and 19 on Bangla to English, surpassing OpenAI’s GPT-4 model’s scores of 22.9 on English to French outputs.
Generative AI projects are not limited to language translation alone, but also extend to other areas. For example, researchers at the Indian Institute of Technology (IIT) Guwahati are working on creating affordable visual animation models that study eyes and facial movements from open-source visual databases to replicate the process. Similarly, at IIT Delhi, researchers have developed a language model called ‘MatSciBert’ specifically for material science research, aiming to discover new materials with the help of AI by processing scientific articles and extracting knowledge about materials and their properties.
However, one of the biggest challenges faced by researchers is the scale of computing power required for training large language models. The cost of a single training run of models like OpenAI’s GPT-3 can be exorbitant, making it unaffordable for most academic institutions and Indian companies, except for top tech firms. The lack of access to India’s supercomputer infrastructure, owned by the Centre for Development of Advanced Computing (C-DAC), has also been a hindrance to researchers’ progress, with limited clarity on enabling access to these resources.
Another challenge is the availability of data in Indian languages. While researchers use public datasets such as the Samantaral database released by IIT Madras and scrape newspapers to create their own datasets, there is a need for more comprehensive and diverse data sources for Indian languages to train robust generative AI models.
Despite these challenges, generative AI research in India is gaining momentum, with researchers making significant progress in developing models for Indian languages and other domains. The government and academic institutions need to address issues related to computing power, data availability, and access to supercomputer infrastructure to foster further growth in this field. With continued efforts, generative AI has the potential to revolutionize various industries in India and contribute to the country’s growth in the field of artificial intelligence.