AI in Humanitarian Research: How Can We Harness Its Potential and Mitigate the Risks?
Emerging tools offer significant opportunities for research. But which tasks can we delegate to AI — and where does human judgement and accountability remain essential?
On 26 October, the Humanitarian Research Hub of the Alliance of Ukrainian Civil Society Organizations hosted an online session exploring the use of artificial intelligence in humanitarian research.
The session featured Thomas Byrnes — a researcher and consultant with extensive experience in the humanitarian sector, including work with the Norwegian Refugee Council and UNICEF.
The discussion focused on the practical potential of generative AI, the reliability of its outputs, the protection of sensitive data, and the responsible adoption of new technologies by humanitarian organizations.
AI Is Already in Use. Are Organizations Ready?
One challenge highlighted by the speaker was the gap between how staff actually use AI and whether their organizations have policies governing its use.
This contributes to what is known as shadow AI — the use of AI tools by staff outside formally approved procedures. Rather than relying on bans or avoiding the topic, the speaker encouraged organizations to discuss AI’s benefits and risks openly, invest in staff training, and establish clear rules for its use.
The Key Question: What Exactly Are We Delegating to AI?
AI can help organize large volumes of information, support the initial analysis of interviews, surveys and community feedback, and accelerate the preparation of analytical materials and reports.
Yet even a seemingly simple summary of a long document requires a series of choices: what to include, how to interpret it, and what to consider important. This is where context can be lost, findings distorted, or the experiences of particular population groups overlooked.
In humanitarian work, such errors can affect decisions about assistance. It is therefore essential to distinguish tasks that AI tools can support from decisions for which people must remain accountable.
AI can help us conduct research faster. But verifying evidence, interpreting findings and taking responsibility for decisions remain human tasks.
Six Recommendations for Humanitarian Organizations
Drawing on the presentation and discussion, the following practical principles can help organizations use AI responsibly.
Use AI to Support Researchers, Not Replace Them
Use AI for preliminary data processing, classification and information organization. Interpretation, risk assessment and decision-making should remain under the supervision of qualified professionals.
Verify Sources and Outputs
Check that cited sources exist and genuinely support the claims made. Compare AI-generated summaries of interviews and research with the original material to ensure important context is preserved.
Protect Sensitive Data
Before uploading materials, understand how the service stores and uses data. Exercise particular caution with information concerning internally displaced persons, survivors of violence and recipients of humanitarian assistance.
Do Not Delegate High-Stakes Decisions
Automatically selecting assistance recipients or deciding whether to provide or deny support without appropriate human oversight can create serious risks to people’s rights and safety.
Establish Internal AI Policies
Clearly define within your organization:
- approved tasks and tools;
- data that must not be shared with external services;
- who is responsible for reviewing outputs;
- how the use of AI should be documented.
Invest in Team Training
Teams need more than the ability to operate AI tools: they must know how to formulate precise tasks, provide relevant context, verify findings and understand the technology’s limitations.
Tools and approaches
How Should We Choose AI Tools?
Thomas Byrnes emphasized that the choice of tool should depend on the research task, the organization’s resources and its data-protection requirements.
During the discussion, he highlighted NotebookLM as an option for working with uploaded sources within a defined information context. However, no tool can guarantee reliable results without human verification.
Before you begin
Five Questions to Ask Before Using AI
1. What is the task?
2. What data will be used, and is any of it sensitive?
3. Which tool are we choosing, and why?
4. What boundaries and safeguards are needed?
5. Who will verify the output, and how?
From Using AI Tools to Responsible Research Practice
Adopting AI requires more than technical solutions. It also calls for organizations to rethink how they conduct research, manage data and take responsibility for their findings.
AI opens up new opportunities for the humanitarian sector, but its use must be grounded in the principles of reliability, transparency, data protection and do no harm.
The Alliance of Ukrainian CSOs Humanitarian Research Hub will continue to provide a space for sharing experience, strengthening research capacities and discussing the ethical and methodological challenges of humanitarian research.