Data for what? AI for understanding data use and improving responses to forced displacement
Read about the commitment on “Innovation to increase the quality, timeliness, and accessibility of socioeconomic data on forced displacement” here.
AI for Data
Artificial Intelligence (AI) is transforming how we engage with development data. The World Bank’s AI for Data program is at the forefront of this change, innovating on evaluating data quality, enhancing metadata, boosting data dissemination, and measuring data use. The program also goes beyond the technical progress of new tools by strengthening the positive feedback loop through which better data enables better AI, and better AI enables more effective data use. The good health of this virtuous circle is key to maximizing the contribution of both data and AI to progress towards the sustainable development goals.
Forced Displacement Data Use
The World Bank UNHCR Joint Data Center’s Commitment to Data highlights the potential for innovation to improve the quality, timeliness, and accessibility of socioeconomic data to inform policy and operational responses to forced displacement. One of the ways we are delivering on this commitment is through partnering with the World Bank AI for Data program, adapting tools to meet the needs of UNHCR and its partners. The value of data is in its use, and understanding how data is used by policymakers and operational teams is especially important in fragile and displacement-affected contexts, where high-quality socioeconomic data can be scarce.
Pilot Study
As a first step, we conducted a pilot study in which we applied a model that was developed to detect dataset mentions in academic papers (example output in Fig. 1) to a corpus of policy and operational documents from forced displacement affected contexts. The results showed promise but also limitations of the direct application of a model pipeline trained on and tuned to structured academic research papers, which have relatively strong conventions and norms on data citation. The ways that data are mentioned and referred to in policy and operational documents is much more varied and complex, which means that the model pipeline will need to be adapted to enable data mention detection with sufficient accuracy to be operationalized by teams.
Fig 1. Word cloud of data mentions detected in a set of research papers.
Model Adaption
The pilot study has taken us a step closer to understanding data use for informing sustainable responses to forced displacement. The team is now working on adapting the model pipeline to deal with the additional complexity of operational and policy documents relative to structured
Model Adaption
The pilot study has taken us a step closer to understanding data use for informing sustainable responses to forced displacement. The team is now working on adapting the model pipeline to deal with the additional complexity of operational and policy documents relative to structured academic research papers. Tools developed as part of the wider AI for Data program can help, with semantic search and metadata enrichment playing key roles in the innovative model pipeline under construction.
The Last Mile
By refining these tools, we can better trace the journey of data from collection to impact, identifying which datasets are most influential, where gaps exist, and where we could effectively invest in improving discoverability and accessibility. In contexts where data is scarce and has a shelf-life, we must do what we can to make the best possible use of existing data. This work will help ensure that data investments translate into more effective, inclusive, and sustainable responses to forced displacement.
What’s next?
Stay tuned for updates in 2026 as we continue to refine the model and expand its application. We’ll be sharing findings and tools, and will invite policymakers, practitioners, and researchers from the UNWDF and Commit to Data community of practice to engage with us in co-developing solutions.