Picture two people. One is a cotton farmer in Vidarbha who wakes up to a message in Marathi warning that pink bollworm is spreading in her district. She checks her field, acts in time, and saves her crop. The other is a woman applying for a welfare benefit. An automated system rejects her because her fingerprint failed to match, or her name was spelled differently in two databases. Nobody explains why, and there is no simple way to appeal. Both women have met artificial intelligence. One was lifted by it; the other was locked out. The underlying technology is broadly the same. What differs is how it was designed, whose data trained it, and who answers for it when it fails.
AI could be one of the most powerful tools we have for reaching the Sustainable Development Goals (SDGs). But for the Global South, it will only deliver if it is built to be responsible and inclusive from the start, not fixed up afterwards.
Why the SDGs are the right yardstick
The SDGs give governments and businesses a shared scorecard. They shift the question from “Is this AI impressive?” to “Does it improve lives, and whose?”
That scorecard is flashing red. The UN’s Sustainable Development Goals Report 2026, released in July, finds that only 36 per cent of the 139 measurable targets are on track or making moderate progress. Nearly half are moving too slowly, and 15 per cent have slipped below their 2015 levels. Not one gender equality target is on track, and official development assistance fell by a record 23 per cent in 2025.
There is good news too. Internet access has risen from 40 to 74 per cent of the world’s population since 2015, and the UN now names digital technologies, including AI, as one route back on track. But the core promise of the 2030 Agenda is to leave no one behind. That promise is exactly what inclusive AI means. If AI helps the connected, urban and English-speaking race ahead while everyone else stands still, it will widen the very gaps the SDGs exist to close.
Where AI is already working
India offers three examples that are real, working and scaling.
Health (SDG 3). India carries roughly a quarter of the world’s tuberculosis cases. During this year’s TB Mukt Bharat campaign, an AI model built by Wadhwani AI helped Maharashtra flag 11,091 villages and urban wards as high-risk for transmission, so health teams could focus screening where it mattered. The state detected more than 6,000 new cases in the campaign’s first 35 days, according to reports from the state assembly.
Agriculture (SDG 2). The government’s Kisan e-Mitra chatbot lets farmers ask about schemes such as PM-KISAN, crop insurance and the Kisan Credit Card by voice, in 11 Indian languages. It has answered more than 93 lakh questions. AI-based monsoon forecasts have also reached about 3.88 crore farmers by SMS, and between 31 and 52 per cent of those surveyed said they changed their sowing decisions as a result (TelecomTalk).
Language and inclusion (SDGs 4 and 10). India has 22 scheduled languages and hundreds more in daily use. Bhashini, the national language platform, offers translation, speech recognition and text-to-speech across all 22, through open interfaces any department, startup or NGO can use for free.
What links these successes is that none was built by a single company. Each sits on public digital infrastructure: TB programme data, PM-KISAN records, a shared language layer. India’s experience with Aadhaar and UPI already showed that open, interoperable rails make inclusion scale. AI can run on the same rails, and much of the Global South is watching.
How AI can widen the gaps
Every one of those successes has a shadow.
Data deserts. AI learns from data, and data is unevenly spread. Roughly half of all websites with an identifiable language are in English, while Hindi, spoken by hundreds of millions, makes up under 0.1 per cent (W3Techs). A model trained mostly on English, urban, Western data works less well in rural Jharkhand or northern Nigeria. It mishears accents, misses context, and fails quietly for the people least able to complain.
A gendered digital divide. The GSMA finds women in low- and middle-income countries are 14 per cent less likely than men to use mobile internet. In South Asia the gap is 32 per cent, the widest anywhere, and 885 million women in these countries are still offline. If AI arrives mainly through smartphones, many women will reach it years late.
Automated exclusion. When an algorithm decides who gets a loan, a ration, a pension or an interview, an error is not a statistic. It is a family going without, and the cost falls hardest on those without documents, literacy or connections.
Deepfakes and trust (SDG 16). Synthetic media now spreads through messaging groups faster than corrections can follow. India’s IT Rules amendments, in force since February, require platforms to label AI-generated content and remove certain harmful material within hours. Rules help, but they also need detection tools, local-language fact-checking and media literacy.
Energy and water (SDGs 6 and 13). The International Energy Agency projects data centre electricity use will more than double by 2030, to around 945 terawatt-hours, slightly more than Japan consumes today. In water-stressed regions, where and how we build AI infrastructure is a development question, not just an engineering one.
Five principles for getting it right
Inclusive by design. Build with communities, not just for them. Test in local languages, on cheap phones, over patchy networks, with users who may not read. Kisan e-Mitra works because farmers can simply speak to it.
Representative data. AI that understands India needs Indian data, collected with consent and handled under the Digital Personal Data Protection Act.
Human oversight and redress. In high-stakes decisions on welfare, credit, health or justice, a human stays accountable, and everyone affected can ask “why?” and appeal.
Transparency and accountability. Test for bias across gender, caste, region and language before launch, publish limitations, and keep auditing. MeitY’s India AI Governance Guidelines already name people first, fairness and accountability among their seven sutras. They now need to become practice.
Sustainability. Bigger is not always better. Smaller, efficient models can be cheaper, greener and more private. Track AI’s energy and water use like any other cost.
For policymakers, this means treating open datasets and multilingual models as public goods and keeping regulation risk-based. For business leaders, it means adding the SDGs to the AI scorecard, measuring who a product reaches and who it misses. And for the Global South as a whole, it means sharing tools and lessons across countries like India, Kenya, Brazil and Indonesia, so that we help write global AI rules rather than simply adopt them.
The choice is ours
The farmer and the welfare applicant met the same technology. What separated their outcomes were human choices about data, design and accountability.
With fewer than four years left to deliver the SDGs, AI can help us move faster, but only if we judge it by a simple test: not how many people use it, but who it reaches, and who it leaves behind.
