AI for Social Good: What It Really Takes in India

“AI for social good” is one of those phrases that sounds self-evidently virtuous and turns out to be enormously difficult to live up to. It conjures an image of clever technology solving stubborn problems — disease detected earlier, crops advised more wisely, services reaching people they never reached before. The promise is real. But the distance between a promising demonstration and a lasting improvement in people’s lives is vast, and in India it is paved with lessons that the enthusiasm often skips over.

Having watched many well-meaning technology efforts over the years — some that endured, many that quietly faded — we have come to believe that the model is rarely the hard part. The hard part is everything around it.

The demo is not the deployment

A great deal of AI-for-good work stops at the demonstration. A system performs impressively in controlled conditions, a pilot generates encouraging numbers, a report is written, and attention moves on. The gap between that moment and real, sustained use in the field is where most impact is lost.

In the field, the assumptions of the demo meet reality. Connectivity is intermittent. Devices are shared and old. The data that flows in looks nothing like the clean data the system was trained on. The people meant to use the tool are busy, sceptical, and have seen initiatives come and go. A model that scored well in testing can be useless in a setting it was never designed for. Building for social good in India means treating the deployment, not the demonstration, as the real product — and being honest that the deployment is ten times harder.

Context is not a detail

Technology built without deep local context tends to solve the wrong problem, or solve the right problem in a way no one can adopt. An advisory tool that ignores what a farmer can actually afford to do, a health application that assumes a level of literacy that is not there, a service that requires a document many people do not possess — each fails not because the AI is weak but because the context was treated as a detail rather than the foundation.

Context in India is layered: language, livelihood, local institutions, seasonal rhythms, and the everyday constraints of time and money. Understanding it cannot be done from a distance. It requires spending time with the people the tool is meant to serve, learning how they actually live and work, and letting that understanding shape the design from the start. This is slow, unglamorous work, and it is the difference between a tool that gets used and one that gathers dust.

Build with, not for

Perhaps the most important lesson is also the simplest to state and the hardest to practise: build with communities, not merely for them. The instinct of technologists is to arrive with a solution. The wiser path is to arrive with questions, and to treat the people affected as partners who understand their own lives far better than any outsider can.

This is not sentimentality. It is a practical route to systems that work. Communities know where the real bottlenecks are, which fixes will be trusted, and which will be quietly ignored. They notice failures early. They can sustain a tool long after the initial enthusiasm has moved on — but only if they had a hand in shaping it and see it as theirs. A tool imposed from outside is fragile; a tool built alongside its users has roots.

The unglamorous multipliers

Some of the most decisive factors in whether AI does social good have nothing to do with AI at all. Trust — whether the community believes the effort is genuine and will last. Intermediaries — the local people who can carry a tool the last mile and explain it. Maintenance — the unexciting commitment to keep something working after the launch. Humility — the willingness to admit when a technological answer is the wrong answer and a simpler intervention would serve better.

These multipliers rarely feature in announcements, yet they determine outcomes more than any model architecture. An initiative rich in clever technology but poor in trust and maintenance will fade. An initiative modest in technology but rich in local partnership and staying power can quietly change lives for years.

A long, patient road

None of this is an argument against using AI for social good. It is an argument for doing it with clear eyes. The technology can genuinely help — to extend the reach of scarce expertise, to lighten burdens, to bring services to places they struggle to reach. But it helps only when it is embedded in context, carried by trusted people, maintained with discipline, and shaped with the communities it serves.

At Svadrishti Foundation we hold this as a guiding conviction: that technology for the public good is measured not by how advanced it is but by how faithfully it lifts those it was meant to lift. That is a slower and humbler ambition than the phrase “AI for social good” usually implies. It is also, we believe, the only version of it that lasts.