AI is an enabler and India wants to fully leverage the technology to tackle societal issues. Wadhwani AI, a non-profit AI institute set up by the Wadhwani brothers – Romesh and Sunil Wadhwani – is fully dedicated to tackling societal challenges in India with AI.

“We are probably the only non-profit in the world devoted exclusively to AI for social impact. We want to ride the digitisation and data creation wave to build robust AI solutions that will help people who typically don’t use technology; and that’s our mission,” Alpan Raval, chief scientist – AI/ML, Wadhwani AI, told AIM. “We work on developing AI solutions for underserved communities in the Global South. Our primary focus at the moment is underserved communities in India but our charter is broader than that.” 

Established in 2018, and inaugurated by Prime Minister Narendra Modi, Wadhwani AI has developed a host of AI-powered solutions in multiple domains such as healthcare, agriculture, and education. Currently, we find ourselves at the cusp of generative AI and Wadhwani AI is also actively exploring the potential of this technology to drive social impact. 

Generative AI for social impact

Raval reveals that since the launch of ChatGPT last year, they have seen a huge spike in interest from various government ministries and agencies for generative AI. “The health ministry and the agriculture ministry, they all want to know what this exciting new technology is and how we can leverage it for social impact.”

Wadhwani AI is continuously exploring the use of large language models (LLMs) and more generally foundation models across the board. Even though GPT-4 is the most advanced LLM out there and is available through APIs, the institute isn’t using the models extensively. Instead, they are experimenting with many open source models like Falcon, a fully open-source LLM developed by the Technology Innovation Institute (TII) in UAE. “There are limitations to using GPT both from a cost perspective, as well as the data perspective. We have to be very careful about where the data goes. In most of our applications, we have to ensure that the data stays in India,” Raval said.

In most cases, Wadhwani AI prefers fine-tuning on custom data to create its own fine-tuned language models. Generative AI is in the process of being integrated into its clinical decision support system. “We use large language models in order to extract knowledge from medical texts and use that to make decisions.” 

The non-profit previously developed a disease surveillance system utilising language models. Now, they are working on a more advanced generative AI system to augment its capabilities. This system scans local news sources, extracts relevant disease events, and reports them to a central cell in Delhi. Wadhwani AI is also exploring the use of generative AI to power Kisan Call Centres. “Our approach involves augmenting human expertise by utilising models that provide automated responses based on a knowledge base. We create these knowledge bases using current government reports and documents, and our speech interface enables a conversational AI system for seamless interactions,” he said.

Not only this, the institute is exploring the use of generative AI in many other areas. “We’re using generative AI to create eye images which the ophthalmologists cannot distinguish from real images. Generative AI is being used for projects like diabetic retinopathy detection and pest management to augment training datasets. “We are also in talks with partners for developing financial inclusion tools that use language models, as well as automatic translation models,” Raval said.

Developing AI solutions for governments

The institute not only works closely with numerous governments, they also help establish AI centres of excellence within government organisations. “This has helped us tremendously in terms of getting access to datasets, getting the requisite permissions, and opening up deployment pathways because it builds trust. We have our own employees stationed within these government departments,” Raval explained.

The institute has developed an AI-powered tool for pest management for cotton farmers in India. Called CottonAce, and available through a mobile app, it helps farmers protect their crops by determining the right time to spray pesticides through immediate, localised advice. The app can work offline in remote areas as well and is available in nine languages — English, Hindi, Marathi, Gujarati, Telugu, Kannada, Tamil, Odia, and Punjabi, and is being used by more than 33,000 farmers across six states and over 140 talukas.

“We are in the process of generalising that to other pests and crops and so on. And we are also in a partnership with the Ministry of Agriculture & Farmers’ Welfare to be an integral part and develop solutions around agri stack.

Similarly, in healthcare, Wadhwani AI has been involved in multiple Ministry-led projects like maternal and child health, tuberculosis, and disease surveillance, among other things. For example, their AI-powered disease monitoring solution has already been deployed and is being used actively by the Government of India’s Media Scanning and Verification Cell. Since its launch in April 2022, it has evaluated 9 million health articles, 788 health events published and 115 outbreaks confirmed.

Furthermore, their Clinical Decision Support System has also been integrated with eSanjeevani 2.0. Since deployment from March 1 to April 15 2023, the AI generated smart differential diagnosis has enabled over 5 million Al-powered consultations. “Our ongoing anthropometry project involves developing an app that uses short videos of babies taken in rural home settings. The app accurately estimates the baby’s weight and other anthropometric metrics, achieving an impressive error margin of only 150 grams,” said Raval.

There were hardly any pre-available datasets to build these tools. The institute collaborated with partners, including NGOs and other government partners, to collect ground-level data from hospitals and community settings. “Additionally, for certain projects like our TB solutions, we utilised government-collected longitudinal data on TB patients, tracking their journey from diagnosis to treatment completion. Given the number of TB patients in this country, the data set is huge and very informative.”

Recently, Wadhwani AI also ventured into education and is developing AI solutions in collaboration with UNICEF and the Gujarat government. “We are looking into AI for climate change. We are also interested in the use of AI in genomics and molecular biology, among other things,” Raval concluded.