India’s Artificial Intelligence Story Does Not Need Hype, It Needs Realism

India's AI conversation needs to be grounded in reality—neither brochure-like optimism that promises everything will magically improve overnight and catapult India to superpower status nor incurable pessimism that sees only doom and job losses. What is required is a clear-eyed discussion on what can realistically be achieved, especially in AI-driven healthcare where lives are at stake.

Visit forestarrow.help for more information.

The conversations at the India AI Impact Summit in Delhi this week were revealing. Alongside the buzz around innovation, one heard words like “ethics”, “localisation”, and “trust”. These will be central to India’s AI story in the days ahead. There are lessons from elsewhere in Asia.

Think back to IBM Watson for Oncology, introduced as a groundbreaking AI solution for cancer diagnosis and treatment. In 2012, IBM partnered with Memorial Sloan Kettering Cancer Centre in New York to train AI to make treatment recommendations. A few years later, top hospitals in Thailand, India, and South Korea signed adoption agreements, drawn by the promise of bringing world-class cancer care to their patients.

Watson’s mission was to export a Manhattan “Gold Standard” by digitising the institutional memory of Memorial Sloan Kettering. The idea was noble—to democratise the expertise of its oncologists and provide a decision support system for institutions like Manipal Hospitals in India, Bumrungrad International in Thailand, and Gachon University in South Korea.

Yet, it did not quite work as intended. By 2022, IBM had divested Watson Health into Merative, which was sold to a private equity firm.

The case remains a foundational warning about the fallacy of a universal brain that ignores specificities.

What went wrong? Experts agree on one key reason—a total lack of local context. Watson was trained on “synthetic cases”—hypothetical scenarios written by US-based doctors—rather than the reality of wards in different medical and economic universes.

This lack of local realism collided with clinical reality in India and Southeast Asia. Hospitals in India, Thailand, and South Korea found IBM Watson for Oncology (WFO) achieved high agreement with doctors for common cancers like breast and lung. However, it showed significant limitations with region-specific cancers such as gastric cancer, local drug availability, treatment variations, often a “Western bias”, and reduced accuracy with elderly patients. The WFO frequently recommended newer, expensive immunotherapy and targeted drugs that were unavailable or financially inaccessible locally.

“The generally affluent population treated at Memorial Sloan Kettering does not reflect the diversity of people around the world. The cases used to train Watson therefore do not consider the economic and social issues faced by patients in poorer countries…” noted an April 2025 article in the International Research Journal of Innovations in Engineering and Technology (IRJIET).

The fallout from Watson’s failure birthed a new era of “Sovereign AI”, words heard a lot at the recent AI Summit.

India uses AI in its public healthcare system, including government hospitals, with tools like Qure.ai for quick detection of TB and lung diseases through chest X-rays. Startups such as NIRAMAI (AI-based breast cancer screening) operate in over two dozen cities. The momentum will grow steadily.

And yet one must circle back to the cautionary side of adopting AI in healthcare, best articulated by those who see patients.

“We use AI at Tata Memorial Hospital only in research mode so far. We are working on solutions that look at pathology and radiology image analyses for various diagnostic, prognostic, and predictive outcomes,” says Dr C S Pramesh, the Director of the Tata Memorial Hospital, Mumbai.

“Many datasets on which AI solutions are based are highly biased and from high-income country (typically Caucasian) datasets. This can lead to very variable outcomes when applied in a different setting. Hallucinations (AI errors) are a real problem, though there is progress in this in recent months. Appropriate clinical validation (as opposed to technological validation) is a very common problem with many AI solutions. In summary, while I am convinced that AI solutions will transform healthcare for the better in the months and years to come, we need to be cautious about these aspects. In addition, we need to ensure that we emphasise ethical and responsible AI to make sure these solutions are person- (and patient-) centric and reduce inequities rather than exacerbate them,” he adds.

Nalagarh Police Station Blast: 2 More Key Perpetrators Held

Dr Anant Bhan, a leading public health researcher, argues that AI oversight is crucial but requires a focus on quality and the inclusion of the user in the development process. “In a recent paper published with colleagues in Wellcome Open Research based on conversations with diverse participants, we learnt that AI oversight is crucial but will require us to focus on ensuring the quality of the AI interventions while also working on making users of AI a key part of the development and usage process and crucially working to enhance AI literacy among communities, regulators, and ethics committee members,” he told me.

Bhan stresses the importance of ensuring technology developers understand the need to follow and respect ethics and regulations, that health professionals understand how to leverage AI with responsibility, and that regulators and ethics committee members can understand the technology and provide oversight. “We also need to ensure that we have public engagement around AI to understand both the benefits but also potential risks.”

In 2026, no serious hospital, regulator, or clinician would accept “Trust me, I’m an AI” as a clinical protocol. India must focus on building AI that fits its own healthcare needs, with ethics, trust, and local context at the centre.

Patralekha Chatterjee is a writer and columnist who spends her time in South and Southeast Asia, and looks at modern-day connects between the two adjacent regions. X: @Patralekha2011

Read full story at source