Chennai: Researchers from the Wadhwani School of Data Science and Artificial Intelligence (WSAI) at IIT Madras have proposed a responsible ‘Frugal AI Framework’ to guide India’s development and deployment of AI systems.The framework, outlined in a white paper by the Centre for Responsible AI (CeRAI), calls for AI systems that deliver public value with minimum computational resources, energy and financial investment. It recommends that India focus on smaller, specialised, domain-specific and multilingual systems suited to its infrastructure and diverse population.“AI does not have to know everything to be useful. It has to know the right things, deeply, for the task at hand,” IIT Madras director V Kamakoti said, stressing the need for AI that is useful, affordable and sustainable.Balaraman Ravindran, head of CeRAI and WSAI, said, “India does not need to replicate the capital-intensive AI development models pursued by countries such as the US and China, nor remain a downstream adopter of foreign AI technologies. Instead, it can develop an approach rooted in its own strengths in frugal innovation, engineering capabilities and digital public infrastructure.”The framework recommends assessing whether AI is necessary and considering alternatives such as rules-based systems, retrieval-based approaches and conventional machine learning. It also calls for shared public AI infrastructure, interoperable systems, vernacular-first development and testing under Indian conditions, including varying bandwidth, device capabilities and energy availability.“Our framework calls for first establishing whether AI is required for a particular task and then identifying the smallest sufficient AI model or technique capable of delivering the desired outcome,” said Gokul S Krishnan, research director at CeRAI, IIT Madras.The white paper proposes a seven-point checklist covering deployment constraints, including resource tracking, total cost of ownership, documentation, auditing and human oversight. It recommends publishing incidents and failures from AI deployments to improve future implementation.
