Pinpointing where poverty is most severe and tracking its changes over time is crucial for helping communities effectively. However, traditional benchmarks like household surveys and national censuses often fall short—they’re expensive, slow, and infrequent. In countries like Sri Lanka, this means we’re often relying on outdated information, hindering our ability to respond to sudden economic shocks or disasters.

On top of that, poverty cannot be determined by income data alone, rather its multidimensional, where factors such as infrastructure, access to services, and economic activity also play a role in determining a community’s well-being. To capture these complexities, our DAP team (Data, Algorithms, and Policy) explored something different: how to rethink the way we measure poverty in Sri Lanka using AI with non-traditional data sources?

In this study, we used anonymized Call Detail Records (CDR) and remote sensing data—to estimate poverty levels at the most granular administrative levels (Grama Niladhari Divisions) in Sri Lanka. This work makes two main contributions. First, we adapted existing machine learning approaches to the Sri Lankan context, to provide fresh insights into the spatial distribution of poverty. Second, we address significant gaps in literature by proposing frameworks for validating model outputs on their ability to inform poverty alleviation initiatives. We evaluate the models’ ability to identify GN divisions with the highest poverty.

And the results were promising, our models successfully identified 22 out of the poorest 25 Divisional Secretariat Divisions (DSDs) as established from the poverty metrics of Household Income and Expenditure Surveys (HIES), and Census data.

We’re excited to share that our research was selected for presentation during the 39th edition of the flagship conference—The Association for the Advancement of Artificial Intelligence (AAAI) —one of the world’s leading conferences dedicated to advancing the scientific understanding of Artificial Intelligence and Machine Learning. In addition to the main conference, AAAI organizes a series of focused workshops that bring together researchers, practitioners and policymakers around key challenges and opportunities in the AI landscape. Our work was featured in “AI for Public Missions” workshop—a space dedicated to showcase innovative applications of AI for social good. We are honored that LIRNEasia’s research was one of nine projects selected for presentation.

This research is a part of LIRNEasia’s broader initiative, “Harnessing Data for Democratic Development in South and Southeast Asia,” which aims to use data-driven approaches to shape evidence-based policymaking and improve governance. One of the major objectives of this project is to explore the intersection of AI and non-traditional data sources in tackling socio-economic issues, like poverty and inequality.

Representing the team, researcher Chanuka Algama presented the findings in a poster session at AAAI-25, engaging with experts from academia, industry, and government institutions. The discussions that followed went beyond just the technical details. Attendees were interested in our model training process, and how the models performed in real-world scenarios with actual poverty data, and steps we took to address the privacy concerns in CDRs. There was a lot of curiosity around how this approach could be adapted to the regions where mobile data might not be as widely available. Notably, several professors and researchers from leading global institutions expressed interest in collaborating with LIRNEasia to explore AI-driven solutions for public missions, particularly in Asia and developing regions.

LIRNEasia’s work was presented alongside research from some of the world’s most renowned institutions, including:

1. The Folly of AI for Age Verification – Harvard University
2. A Framework for Evaluating Vision-Language Model Safety: Building Trust in AI for Public Sector Applications – Baylor University (School of Engineering and Computer Science, Texas, USA)
3. AI-Enabled Support Services for Migrants: A Community-Centered Design and Development Approach – DePaul University (Chicago, Illinois, USA)
4. Benchmarks Assessing LLMs Legal Reasoning over the Biological Weapons Convention – Information Sciences Institute, USA & RAND Corporation
5. Disseminating Authentic Public Messages using Chatbots – University of South Carolina, USA
6. Differentially Private Synthetic Time-Series Data Generation – De Montfort University, UK & University of Basel, Switzerland
7. Knowledge Graph Analysis of Legal Understanding and Violations in LLMs – USC Information Sciences Institute
8. A Mission-Driven Conversational Consultant for Pre-Venture Entrepreneurs in Food Deserts – San Diego Supercomputer Center, University of California, San Diego & NYU Stern School of Business & UC San Francisco School of Medicine

Moving forward we plan to enhance model performance through incorporating visual data sources, such as satellite imagery as an additional layer of information alongside remote sensing and CDR. Subsequently following this recognition, LIRNEasia looks forward to expanding collaborations with global AI researchers, policymakers, and institutions to drive AI-powered solutions for societal impact. The enthusiasm from AAAI-25 attendees has paved the way for potential partnerships that could extend the reach and influence of LIRNEasia’s work beyond South Asia.

For those interested, LIRNEasia’s full research paper is available below. 

To learn more about AAAI-25, visit the AAAI-25 AI for Public Missions official website here.

  Download PDF   Email