India is a testing ground for ‘good AI’ infrastructure and inclusive applications

Few countries have the market heft, and impact opportunity, that India does. India’s fast-growing consumer class, and also its large rural and urban poor populations, are proving ground for companies designing AI applications and infrastructure for good.

“Unless AI benefits the bottom half of the Indian population, we’re not going to see a huge amount of impact,” said Vinod Khosla, the Indian-American billionaire and venture capitalist in a talk at the recent India AI Impact Summit in Delhi, which drew more than 70,000 people last month. “If we don’t do that, it’s a massive opportunity lost.”

Khosla stressed three urgent applications: AI-based personal tutors for children in rural India where teachers are lacking; AI-based primary medical care and chronic disease management; and AI agronomists that support India’s farmers.

A key challenge is how to build such tools to be effective at scale, for use by a huge population that speaks multiple languages.

In the health sector, for example, “there is limited evidence about what interventions can feasibly be integrated into health systems and primary care, to strengthen service delivery and improve health in low- and middle-income countries,” said Charlotte Watts of UK charitable foundation Wellcome.

Wellcome is partnering with the Gates Foundation and Novo Nordisk Foundation on a $60 million collaboration to evaluate AI-enabled decision support tools in Africa and South and Southeast Asia. The Evidence for AI in Health, or EVAH, initiative will fund assessments for AI tools meant to help health workers do clinical tasks like triage, diagnosis and referral.

Government agencies in India are also partnering with startups, companies and nonprofits to support high-impact applications of AI. 

The Indian Council of Medical Research, a government-funded agency that works on biomedical research, prepared an image repository from more than 10,000 tuberculosis patients, which a nonprofit in the state of Gujarat is using to develop a tool for TB diagnoses. 

ICMR scientists are also using AI to map genes and prepare for new vaccine development, so that when there’s another pandemic, “we will have the vaccine in months,” said Taruna Madan, an ICMR scientist.

Data centers

The summit was a means to center India in the global AI race. In dozens of sessions over five days, experts at the India AI Impact Summit shared research, pilot projects and new applications from India, Africa and other markets. A number of major investments for data centers and AI infrastructure by Indian conglomerates like Reliance and the Tata Group were also announced. 

OpenAI announced that it would be the first customer of Tata Consultancy Services’ HyperVault data center business, which will enable OpenAI to run advanced models in India. Larsen & Toubro Group is teaming up with Nvidia to build AI infrastructure, advanced computing platforms and other support for companies running AI workloads in India. 

Google, Microsoft and Amazon have also pledged billions of dollars to build data centers and other AI infrastructure in the country. 

Blackstone, Ontario Teachers’ Pension Plan and other investors committed $600 million to Mumbai-based startup Neysa to expand India’s AI computing capacity. 

Missing from the conversation: the environmental impact of running these centers. In one interview, Open AI’s Sam Altman denied that excessive water use to run ChatGPT and other applications are “totally fake,” though he admitted that the massive use of energy is real.  

AI for farmers

One sector that drew attention for the high-impact potential of AI was agriculture. 

The Indian Institute of Technology in Ropar, in Punjab, is building a large language model for answering farmer questions and providing accurate weather data to farmers. 

The UN World Food Programme showcased its work to optimize the transport of grains from farmers in India to warehouses to the more than 600,000 fair price shops, from where it is distributed to the poor. At the event, UNWFP said it had launched a mobile robot, that circulates around the warehouse when spoilage is happening, or to detect pests.

There was also a presentation by French nonprofit Current AI of a prototype of a handheld device that can listen to, process and respond to farming queries in 22 Indian languages. In one example, a farmer displays a bottle of crop treatment that he had sprayed on his field and asks the device, in Hindi, when he needs to spray again.

“Administer this medicine again after three to five days,” the device responds in Hindi. 

The prototype was developed using funding pledged by the French government, the MacArthur and Ford foundations, Salesforce and others at last year’s AI Action Summit in France. Current AI is working with the Indian government’s Bhashini initiative to leverage technology for speech recognition and language translation. The intention, said Ayah Bdeir of Current AI, is “to get as close as possible to the individuals and the communities.”

AI applications for language was a broader theme at the summit. India has 22 officially recognized languages and hundreds regional languages and dialects. Most AI applications, however, are in English. 

Startups like Bangalore-based Gnani.ai are now providing AI-led virtual assistants for customer engagement. Its speech-to-text model works in 11 Indian languages. 

Limitations to use

For all the promising AI applications on display, the shortcomings and risks were equally evident.

Arun Pratihast, a senior researcher at agriculture-focused Wageningen University in the Netherlands noted that many promising AI and digital tools for farming fail because they’re designed to address a global issue rather than local farmers’ needs. There’s also a lack of data in emerging economies.

“That hinders AI models,” Pratihast said.

A recent study on the effectiveness of AI solutions in agriculture in low and middle-income countries, funded by the UK Foreign, Commonwealth and Development Office, showed that most models are early-stage and don’t stick, in part because of a lack of trust from farmers.

One reason is that solutions are being designed to solve specific problems like pest control and crop productivity, rather than more holistic issues like income security or livelihood improvement, observed Zeba Siddiqui of Athena Infonomics, which helped conduct the review.

Farmers are also skeptical because of lack of feedback and accountability mechanisms, noted Kavya Dashora of the Indian Institute of Technology in Delhi. “If something goes wrong, who will be responsible?”