AI for Good: Transforming humanitarian response 

Artificial intelligence is transforming every sector and industry, and the U.S. nonprofit sector is no exception. By leveraging AI, nonprofits can unlock powerful insights, make informed decisions, and optimize operations in ways previously unimaginable.  

Yet many nonprofits hesitate to adopt AI, often due to perceived challenges around cost, expertise, and ethical considerations. Concerns about AI’s complexity and the technical skills needed to ensure data privacy, security, and fairness are valid and warrant consideration.  

At DataKind, we partner with organizations to develop AI tools to increase impact. From our experience of more than 350 completed projects over a decade, we know nonprofits can embrace AI effectively, even on a budget. One of the most promising applications is in humanitarian response, where rapid access to reliable data can mean the difference between life and death.  

Humanitarian response as a use case: The Humanitarian Data Insights Project 

Humanitarian ​​​​organizations struggle to find and use accurate, timely data effectively. Where data is available, it’s often scattered across reports, spreadsheets, and platforms, which creates silos of information, making it time-consuming to derive actionable insights.  

Recognizing the need to transform data chaos into information organizations can use to direct humanitarian response efforts, we developed the Humanitarian Data Insights Project (HDIP). It’s an AI-powered tool designed to enable humanitarian organizations to rapidly use incoming data to inform decisions in high-stakes scenarios. For example, during civil conflicts, humanitarian organizations need information from multiple data sources to evaluate humanitarian access, urgent needs, and critical gaps. HDIP can help an organization save days of work to locate data streams on, for instance, population movement and which organizations are already in the vicinity with what provisions. This helps speed up and informs how an organization can best respond to reach more people, without duplicating efforts.  

Developed with support from Microsoft and guidance from an advisory group led by Save the Children (that also included Catholic Relief Services, Mercy Corps, UNOCHA, Norwegian Refugee Council, and others), HDIP addresses the challenges of data access in crisis response. 

Bridging data gaps for humanitarian organizations

Following an initial period of user research to identify the most pressing data issues hindering efficient humanitarian response,​​​​ we tackled two key priorities:  

1. Metadata prediction. HDIP uses generative AI to streamline data accessibility by predicting metadata. Metadata provides essential context and documentation that makes the data usable, trustworthy, and maintainable. There are thousands of humanitarian datasets that don’t have metadata. Using genAI helps humanitarians quickly locate and correctly combine datasets to analyze and use in critical situations. 

2. Conversational analysis. HDIP introduces the ability to ask data-related questions in plain language using a chat interface. Users start with a question and continue with further questions to get more detailed information. Without extensive technical knowledge, anyone can intuitively and quickly retrieve information from multiple data sources and perform meaningful analyses to guide their response efforts. 

Designing AI tools with humanitarian needs in mind 

HDIP isn’t just about delivering information—it’s about delivering it responsibly, sustainably, and in support of existing workstreams. The priority needs of humanitarian organizations are reflected in the tool’s unique features:  

• Trusted data sources. HDIP leverages data from the Humanitarian Data Exchange (HDX), which is maintained by UNOCHA and widely used by the humanitarian response community. While HDX serves as the initial data source, HDIP will accommodate data from additional sources in the future, including an organization’s internal data, offering even broader access and nuanced insights. 

• Data visualization. HDIP can generate charts, graphs, and maps to help users visually interpret data, making it easier to understand and communicate insights in a way that supports planning and coordination.   

• Transparency and human-in-the-loop verification. HDIP uses a human-in-the-loop model for quality control. “Data recipes,” or AI-generated reusable code to perform analyses, is validated by a human expert to ensure accuracy, ethics, and alignment with the user’s question. These codes are then made available for use by people without extensive quantitative data analysis skills. The data recipes are saved in a catalog for reuse, ensuring that high-quality responses are accessible for common queries.  

• Sustainability and cost efficiency. By reusing stored query recipes, HDIP minimizes the need to generate new AI queries for each question, reducing both the environmental impact and the token costs associated with genAI. This model makes HDIP a financially and ecologically responsible tool for humanitarian organizations working with limited resources. 

A vision for data-driven humanitarian response—and beyond 

HDIP will be available in a public beta version ​on​ December​ 10,​ 2024. We invite humanitarian practitioners to test the tool, provide feedback, and participate in its ongoing refinement. Those interested in shaping the tool’s development can join our advisory group. 

HDIP is just one example of how AI, when thoughtfully implemented, can elevate an organization’s ability to serve. In adopting AI, nonprofits are investing in their mission, building a more data-informed, impactful future. We’re excited about those future possibilities, because the approach adopted with HDIP is broadly applicable. We know a similar approach in other data-rich sectors like education, health, housing, and economic inclusion could be equally transformative. 

If you’re interested in enhancing data accessibility for greater impact, we encourage you to reach out. Together, we can bring the benefits of AI-driven insights to more organizations and communities in need.  

Photo credit: Jason Whitman via iStock