In a groundbreaking initiative, the open source activists association Swecha and the International Institute of Information Technology – Hyderabad (IIIT-H) have joined forces to launch viswam.ai, a Centre of Excellence dedicated to developing AI solutions tailored to the unique needs of the Global South.

Kiran Chandra, Centre Head of Viswam.ai, and Ramesh Loganathan, Professor of Practice, Co-Innovations, at IIIT-H, this ambitious project seeks to **democratise access to AI** by **leveraging community resources and expertise**.

“The driving force behind viswam.ai is a belief that current AI models, often developed by large corporations in the Global North, are ill-suited to address the challenges faced by the Global South,” Y Kiran Chandra, Centre Head of viswam.ai and the Founder of Swecha, told businessline.

Stating that cultural context holds the key to the development of AI applications in general and large language models (LLMs) in particular, he said language, as a “compressed format of expression,” is deeply intertwined with cultural nuances.

“Existing models lack this crucial understanding and risk perpetuating a homogenisation that fails to cater to the diversity of the Global South,” he pointed out. 

Ramesh Loganathan, Professor of Practice, Co-Innovations, at IIIT-H, said that resource-poor countries often lack the resources to invest in expensive hardware and data collection processes, making traditional AI development approaches untenable.

Viswam.ai’s approach is rooted in community-driven innovation. One of their key strategies is crowd-sourcing data.

“This approach has proven remarkably successful in their previous project, Bhashini, where we amassed 2,000 hours of high-quality speech data in a single weekend with the help of 50,000 student volunteers,” he said.

Lightweight AI model

Building on this experience, Viswam.ai aims to “develop lightweight AI models that require significantly less computational power”. This approach reduces reliance on expensive hardware and aligns with their commitment to environmental sustainability, as Loganathan points out the alarming increase in power consumption associated with large AI models.

Crowd-sourced computing power

One of Viswam.ai’s most innovative aspects is its exploration of community-sourced computing power. Drawing inspiration from projects like SETI@home (a UC Berkeley-led initiative that sought to build a network of computing for the Detection of Extraterrestrial Signals), Loganathan said the centre is planning to tap into the idle processing power of mobile phones and desktops.

While acknowledging the privacy and security concerns associated with such an approach, he believes that through careful design and transparent communication, they can foster the necessary trust within the community.

By empowering communities to participate in the development and deployment of AI, they aim to ensure that these technologies serve the needs of the many, not just the privileged few.  

Viswam.ai, which is in the advanced stage of building a Telugu LLM, plans to replicate the models in other Southern languages and some underrepresented languages like Gondi, Koya, and Lambadi so that people speaking those languages will also benefit.