Anticipatory action and artificial intelligence: plotting a pathway forward

Artificial intelligence (AI) has great potential for strengthening anticipatory action, for example by improving hazard forecasts or supporting decision-making processes. To be effective, however, it must be tailored to real-world humanitarian contexts. A workshop held in Heidelberg, Germany, in June 2025 explored pathways to ensure that AI is used responsibly and effectively for anticipatory action. 

Over two days, anticipatory action practitioners, researchers and technology experts discussed the role of AI within three areas of anticipatory action work:  

1. forecasting and early warnings 
2. risk mapping and vulnerability assessments 
3. analysis, communication and collaboration.

Forecasting and early warnings 

AI tools (e.g., generative and machine-learning models) can improve the accuracy and granularity of hazard forecasts, as well as increase the lead times between a forecast and a hazard occurring. This gives humanitarian actors more time to implement anticipatory actions – especially critical for rapid-onset hazards such as flash floods – and more accurate information about how and where the hazard will hit.  

However, for anticipatory action to be effective, forecasts must go beyond predicting the hazard; they need to be linked to the expected impacts on people. While AI can help to combine weather data with information on (local) vulnerabilities, past disaster impacts and other risk indicators, in many high-risk areas, especially those affected by conflict, this kind of data is often incomplete or completely unavailable. It is therefore essential to increase the availability of locally validated data through ground-truth observations to translate improved forecasts into targeted anticipatory actions. 

Risk mapping and vulnerability assessments 

By processing large volumes of historical and real-time data (e.g., satellite imagery), AI tools can support humanitarian actors in mapping both risks and vulnerable populations, resulting in more dynamic and detailed risk assessments. This can help with identifying or confirming the priority areas and communities for interventions, assessing the structural vulnerability of housing and road infrastructure, or mapping access to critical services and healthcare facilities among other benefits. This leads to better-targeted and more efficient anticipatory actions. 

But technology alone is not enough. During the workshop, participants emphasized the importance of ensuring that AI tools reflect local realities and do not reinforce or introduce algorithmic bias. When used responsibly, AI can help elevate community-driven insights and support bottom-up approaches – but this requires ethical safeguards, human oversight and meaningful engagement throughout the process.   

Analysis, communication and collaboration  

AI is not only helping to improve forecasts and map risks, it also shapes how this information is shared and acted upon. Examples include the use of language-processing tools to create multilingual early-warning and sensitization messages, including in minority languages, or the use of large language models to summarize evolving risk situations in real time. AI tools also offer opportunities to adapt solutions from other sectors, such as logistics or finance, to support the planning and delivery of anticipatory actions. 

At the same time, the workshop participants reiterated the gaps and challenges of using AI for analysis, communication and collaboration, such as addressing the persistent data gaps that affect many communities. They also looked at unexplored opportunities in this area, which include consolidating evidence and learning about the impacts of anticipatory action, and using AI to monitor and document the risks from compounding hazards.   

Building a pathway forward 

To capitalize on AI’s potential to improve anticipatory action, targeted and sustained exchange between humanitarians, researchers and technology developers is essential. The workshop ended with the participants developing a roadmap to guide this collaboration. This links concrete challenges – for example, working in data-scarce or conflict-affected settings, integrating local knowledge and advancing impact-based forecasting – with those best placed to address them. The aim is to ensure that AI-supported tools for anticipatory action are not only innovative, but also ethical, inclusive and tailored to the humanitarian realities in which they are needed most.