AI in disaster risk reduction: Who is being left out?
Artificial intelligence (AI) is becoming part of disaster risk reduction (DRR). It is already being used to forecast hazards, map exposure and guide decisions about where assistance may be needed.
That potential is real. But so is the risk.
Over the past few years, I have been esearching how AI may affect marginalised groups in disaster contexts. One point has become increasingly clear: AI does not arrive in DRR as a neutral tool. It enters systems that already struggle to see, reach and support some of the people most at risk.
The problem starts with data
Most AI systems learn from already-existing data. In DRR, this might include census records, hazard maps, satellite imagery or health data. That sounds useful, and often it is, but not everyone is equally visible.
People living in informal settlements may be missing from official maps. Undocumented migrants and stateless people may be absent from official identity and registration systems. Nomadic communities may not fit into systems built around fixed addresses. People with disabilities, gender and sexual minorities, informal workers and people facing social stigma may be poorly recorded, misclassified or not counted at all.
When AI is trained on data that contains these gaps, it does not correct them. It learns from them. A model may appear accurate while still producing an incomplete picture of who is at risk, where they live and what support they need.
That matters when AI is used to inform response, recovery and preparedness.
It's not just data
Even where an AI system produces useful information, people still need to receive it, understand it and act on it. That usually requires access to a phone, internet connection, electricity, literacy, digital skills and, in some cases, formal identity documents.
These are not small barriers. A flood warning sent by app or short message service (SMS) will not reach someone without a working phone. A chatbot may be of little use if it does not work in the languages people use, or if people do not trust the institution behind it.
During a crisis, these barriers can get worse. Couple them with power failures, lost phones and network outages and people are further displaced from the services they rely on.
This creates what I have been calling “double exclusion”: people are first missing from the data used to build AI systems, and then unable to access the systems those tools support. For inclusive DRR, that is a red flag.
The risk is not just technical
There is a temptation to treat these issues as technical problems, something that can be solved with better data, better models and better user platforms.
But the rapid integration of AI into our DRR systems raises questions of power and accountability. Who decides what data is used? Who defines vulnerability? Who checks whether an algorithm is producing unfair outcomes? Who can challenge a decision if an automated system deprioritises their community or increases their exposure to risk?
After all, as we know from experience, in many disaster contexts, the people most likely to be harmed by exclusion are also those least able to question how decisions are made.
What needs to change
The answer is not to reject AI, the same as the question is not whether AI belongs in DRR. The question is how it is used, who shapes it, and whose interests it serves.
A more inclusive approach could start with several commitments:
- Organisations should assess who is missing from the data before relying on AI outputs.
- Digital tools should never be the only route to warnings, registration, feedback or assistance.
- Affected communities, including those often left out of formal consultations, should be involved in the design, testing and review of AI-enabled systems.
- There must be clear routes for people to question decisions, report errors and seek redress.
There is an important role for donors and governments as well. Funding innovation should include funding for independent evaluation, bias testing, digital inclusion and community engagement. Procurement rules should require transparency, especially where tools inform high-stakes decisions.
Inclusion must come first
DRR has long recognised that risk is shaped by inequality, discrimination and power and AI does not sit outside those realities. It is built within them.
If the future of DRR is digital, inclusion cannot be an afterthought. It has to shape the data, design, deployment and governance from the start. Otherwise, we risk building tools that work well for those already visible, connected and recognised, while leaving the most marginalised further behind. That is the opposite of what DRR is meant to do.
Kevin Blanchard MSc FRGS (he/him) is a policy specialist, trainer and researcher with more than fifteen years’ experience working across disaster risk reduction, climate adaptation and inclusive governance. His work focuses on strengthening how policy and practice respond to marginalised and hyper-marginalised groups, with particular attention to intersectionality, rights, and structural exclusion.
Kevin works internationally with grassroots organisations, universities, national governments, UN agencies and civil society organisations. He supports the development of inclusive policy frameworks, practical guidance and training that translate evidence into action, helping institutions embed equity, participation and accountability into decision-making before, during and after crises.
Alongside his professional work, Kevin sits on the Editorial Board of the Journal of Disaster Studies and convenes and hosts the #DRRLive seminar and webinar series, creating space for dialogue on inclusive disaster risk reduction, emerging risks and governance challenges.