6 Examples How Humanitarian AI Increases Impact Around the World

As the ChatGPT news makes the round, articles pop up discussing what role similar AI tools can play in the humanitarian sector .

However, while ChatGPT might be the new shiny tool that made the news, deploying AI in the humanitarian sector is not a new thing. In fact, there are plenty of examples of AI and machine learning in particular already in use today. Naturally, we at engageSPARK are interested in seeing what’s out there.

In this article, I’ve tried to gather a few very different examples from around the internet to show the breadth that AI usage has reached already—in search of making humanitarian projects faster, cheaper, and more accessible.

Use AI to improve maps (image processing)

When setting up support infrastructure for migrants in Uganda, some of the first questions that will need answering are: how many are where? Where are nearby facilities? How can we get stuff there?

The more real-time the answer and the more accurate, the better the humanitarian response. That’s where the Humanitarian OpenStreetMap Team (HOT) helps, using the OpenStreetMap technology and ecosystem to provide accurate maps to, well, everyone.

Usually, the maps are updated with volunteers. And that goes only so fast. That’s where Microsoft’s “AI for Humanitarian Action” program came in—they used Bing map data to feed satellite images into their machine-learning tool and asked it: what do you see? With a bit (read: a lot!) of training and fine—tuning, it could learn to identify among other things 18 million building “footprints”.

Read more about it from the HOT team , or from Microsoft . If you’d like to learn more about how exactly AI was used to improve the maps, and some of the challenges, watch the following presentation by the HOT team from a mapping conference. Yes, it’s as nerdy as it sounds. ??

Explaining and forecasting migration flows

If employment in Ethiopia is on the rise, how will that affect migration to the UK? Answering complex questions like this one helps understand migration flows, thereby supporting the people who are migrating.

The Danish Refugee Council in cooperation with others built the Foresight tool to predict how migration will change in the coming years for some two dozen countries. See for example the graph under the headline “displacement trends” on the page for Ethiopia .

To do this, they gather information on how an economy is doing, about governance, environment, etc—some 120 indicators in total—and feed all that data into a machine-learning model. Read this Guardian article for more info. If you care about the technical ins and outs (again: nerdy!), then read this post from IBM Research .

Matching refugees with a job (using machine learning models)

Once they arrive in their host country, refugees often struggle to find work and earn. One of the issues is matching refugees and their skills with communities and available jobs.

The International Rescue Committee (IRC) partnered with Stanford University Immigration Policy Lab to help match refugees with a job. Machine learning was used to do the matching, trained on existing data. Read more about this project here or on The Verge . Unfortunately, this project fell prey to COVID disruptions.

Local communities help build better AI tools

One well-reported problem with many machine-learning tools (or rather their underlying models and data) is bias—for example, racial bias. But there is also a general availability bias towards people from highly industrialized nations. There are clear economic reasons for this—and it leaves us with AI tools that are less than perfect for humanitarian settings. Recognizing these shortcomings, Nesta and the IFRC collaborated to “localize AI ”, building AI tools with the communities involved.

In this case, the AI tools built were used to predict the aid needed by households and a COVID-19 fact-checking tool. Interestingly, this case isn’t only about fancy AI helping the humanitarian sector and the local communities. It’s also about these communities helping to build the tools—to serve them better.

The COVID-19 fact checker uses SMS and—forgive the plug—might have been built using our Inbound SMS Rules or SMS Survey .

WhatsApp chatbot

Frequently shoved under the “AI” umbrella, WhatsApp chatbots have become increasingly popular in the humanitarian sector.

The Sésamo Chatbot is one such example: The Sesame Workshop with a consortium of organizations such as the Norwegian Refugee Council built a WhatsApp chatbot for Venezuelan migrants . Through it, kids and their caregivers can access educational content in the same way they chat with their relatives and friends.

Related, do read our blog post on how IOM used WhatsApp to survey illiterate people .

Accessible phone surveys

SMS Surveys are often a great and affordable tool in low-resource settings. You can use it to gather data for monitoring, evaluation, accountability, and learning programs (MEAL), or make educational information easily accessible to constituents. There is one catch though: you’ve to be able to read. Also, while 2-way SMS has many advantages, for some situations they’re also not the right tool .

Enter automated phone call surveys. (Yes, something we proudly do at engageSPARK.) Using a human voice, and the urgency of a phone call alleviates many issues with SMS-based interventions. But there remains the problem with answering: people can only reply with keypresses, or by speaking … and then someone has to listen to dozens (thousands?) of recordings. Needless to say, this is expensive and slow.

This is where AI comes in: speech recognition allows people to talk and the survey system automatically translates this into text (or intentions).

The World Food Program (WFP) has been using a custom-built speech recognition tool to power surveys in remote areas.

I’m happy to share that our Voice IVR Survey supports speech recognition as well. Chat with us if you’d like to learn more. ??

Humanitarian AI is already here

In short, AI has long been helping in the humanitarian sector in a variety of use cases. I hope these examples help you find new ways to do old things better.