Most AI investment targets commercial markets, but some of the most meaningful applications address problems markets neglect: disasters, poverty, conservation and access to services. “AI for social good” has produced real successes — and many projects that never moved beyond pilots. Here are examples that work and the lessons they teach.

Flood forecasting

Floods affect more people than any other natural disaster, and many vulnerable regions lack river gauges for traditional forecasting. Google researchers published a study in Nature in 2024 showing that a machine learning model could forecast extreme river floods in ungauged watersheds with reliability comparable to or better than a leading global model, extending useful warnings to regions with little data. Forecasts are shared through public platforms and alerts, covering many countries.

Early warnings save lives only when they reach people who can act — which requires partnerships with governments, humanitarian organizations and local communities.

Wildlife conservation

Camera traps produce millions of images, most of them empty or showing common species. Platforms like Wildlife Insights, supported by conservation organizations and Google, use AI to identify species automatically, letting researchers spend time on analysis instead of sorting photos. AI also analyzes acoustic recordings to monitor birds, whales and illegal logging (by detecting chainsaws), and satellite imagery to track deforestation.

Humanitarian response

After disasters, AI analysis of satellite images helps map damaged buildings quickly, guiding rescue and relief. Humanitarian organizations also use AI to translate messages from affected people, forecast displacement and analyze needs.

Poverty mapping and cash transfers

Researchers have used satellite imagery and mobile phone data to estimate poverty at fine geographic scales where survey data is scarce. Some programs have used these estimates to target cash transfers during crises, reaching people faster than traditional surveys could.

Health access

  • AI screening for eye disease, tuberculosis on chest X-rays and other conditions in places with few specialists;
  • chat and voice services giving health information in local languages;
  • supply chain forecasting for medicines.

Accessibility and inclusion

Tools that describe images for blind people, caption speech for deaf people and translate sign language extend access to information — see our coverage of accessibility and inclusion.

Lessons from what works

  1. Start with the problem and the people affected, not the technology.
  2. Partner locally: governments, NGOs and communities make adoption possible.
  3. Plan for the last mile: a prediction is useless if no one acts on it.
  4. Design for low resources: limited connectivity, cheap phones, local languages.
  5. Evaluate impact rigorously, not just model accuracy.
  6. Protect vulnerable people’s data, which can expose them to harm.
  7. Plan for sustainability: many projects fail when pilot funding ends.

The ITU’s AI for Good platform, organized with dozens of UN agencies, convenes projects and organizations around the UN Sustainable Development Goals.

How to evaluate a social-impact AI project

Before funding, joining or deploying a project, ask:

  • Who asked for it? Was it requested by the people or organizations it serves?
  • What decision will it change, and who will act on its outputs?
  • How was it tested, in the real setting where it will be used?
  • What happens when it’s wrong, and who bears the consequences?
  • Who owns and controls the data?
  • Who will maintain it after the initial grant or volunteer effort ends?

Projects that answer these questions convincingly are far more likely to deliver lasting benefits.

What to watch next

  • Expansion of AI early warning systems for floods, heat and other hazards.
  • Funding models that sustain social-impact AI beyond pilots.
  • Open datasets and models for conservation and humanitarian work.
  • Data protection standards for humanitarian AI.

Key takeaways

  • AI flood forecasting extends warnings to regions without river gauges.
  • Wildlife Insights and acoustic monitoring speed up conservation research.
  • Satellite AI supports disaster damage mapping and poverty targeting.
  • Success depends on local partnerships, last-mile delivery and rigorous evaluation.
  • Protecting vulnerable people’s data and planning for sustainability are essential.

Frequently asked questions

Is “AI for good” just marketing?

Sometimes. But there are well-documented projects with measurable benefits, especially in forecasting, conservation and humanitarian mapping.

How can nonprofits access AI?

Through donated or discounted tools, partnerships with universities and companies, open-source models and dedicated grant programs.

What are the main risks?

Harmful errors, data exposure of vulnerable people, dependence on external providers and projects that end when funding stops.