DIME Artificial Intelligence

Development Impact AI Lab

The Development Impact AI Lab is an applied AI lab in the World Bank’s Development Impact group. Our team of 30+ scientists, engineers, and researchers builds AI systems used by governments and international institutions worldwide. What makes us distinctive: research and product are tied. We publish the science, ship the systems built on it, and rigorously evaluate their real-world impact, all in the same place.

How we work

We operate across the full chain, from applied research, to product development, to field experiments that test real-world impact. We focus on three areas where AI can move the needle for development: policy decision-making, aligning AI systems with people in low-resource contexts, and planetary risks like food crises.

Three principles guide our work. Trustworthy: open, reproducible, and benchmarked against rigorous baselines. Timely: insights when policymakers need them, not years later. Culturally aware: built for the contexts where AI is deployed, not the ones it was trained on.

What we do:

We build public-good AI systems for development: evidence tools, crisis early-warning dashboards, domain-specific models, and curated datasets.

ZeroHungerAI (public dashboard launching 2026). Real-time, district-level food crisis early warning system powered by news streams, covering 82 IDA countries with weekly updates.

HateDay (2026). First global hate-speech dataset representative of a single day on Twitter, covering eight languages and four English-speaking countries.

ImpactAI (public launch October 2025). Conversational AI agent that synthesizes thousands of rigorous causal studies, giving policymakers instant access to high-quality development evidence with full source attribution and standardized comparisons across interventions. (Visit, Learn more, Demo)

NaijaXLM-T (2024). Context-specific large language model for hate speech detection in Nigerian online contexts. Achieves 90% accuracy versus 60% for generic frontier models. (GitHub, Hugging Face)

NaijaHate (2024). First dataset annotated for hate-speech detection on a representative sample of Nigerian tweets. (GitHub, Hugging Face)

EconBERTa and EconIE (2023). Domain-specific language model pretrained on 1.5 million economic research articles, plus an annotated dataset for named-entity recognition in economics. (GitHub, Hugging Face)

Our research develops and evaluates AI methods that turn complex data into decision-relevant evidence for development policy and practice.

Can social media reliably estimate unemployment? PNAS Nexus, December 2025. AI-based methodology that forecasts US unemployment insurance claims up to two weeks ahead of official data, at national, state, and city levels, by detecting unemployment self-disclosures on social media. Paper

HateDay: Insights from a Global Hate Speech Dataset Representative of a Day on Twitter. ACL 2025, Outstanding Paper Award. The first global hate-speech dataset representative of a single day on Twitter, covering eight languages and four English-speaking countries. Paper

NaijaHate: Evaluating Hate Speech Detection on Nigerian Twitter Using Representative Data. ACL 2024. Shows that models trained on US data substantially overestimate real-world performance in Nigerian contexts. Paper

EconBERTa: Towards Robust Extraction of Named Entities in Economics. Findings of EMNLP 2023. Domain-specific language model and benchmark for extracting named entities from economic research. Paper

Predicting food crises using news streams. Science Advances, 2023. AI methodology that anticipates food crisis outbreaks up to 12 months in advance by analyzing 11 million news articles. Paper, GitHub

Large-Scale Demographic Inference of Social Media Users in a Low-Resource Scenario. ICWSM 2023. Paper

Multilingual Detection of Personal Employment Status on Twitter. ACL 2022. Active-learning methodology for identifying disclosures of employment status on social media in three languages. Paper

Ongoing research

Synthetic Personas (2026). Methodology for augmenting societal surveys with LLM-simulated responses in low-resource contexts, with rigorous evaluation of cultural fidelity across regions.

Impact Predictions (2026). Method that extrapolates from causal evidence to predict intervention effects in context.

All our code and datasets are open-source on GitHub and Hugging Face.

Insights

Our insights translate the Lab’s research and products into accessible briefs, interviews, and analysis for policymakers, funders, and practitioners.

The Next Frontier in Policy Decisions: Between Information and Insight, Part 2: Building an Evidence Infrastructure in the Age of AI. Technical Brief, April 2026. On building the infrastructure that makes causal evidence comparable, contextual, and usable at scale. Read

Extracting Experimental Evidence at Scale: Building the World’s Largest Open RCT Database. Technical Brief, March 2026. How we built a multi-stage, LLM-powered pipeline to extract, standardize, and organize causal evidence from thousands of RCTs in development economics. Read

Spotlight Interview: Karan Bhatia (Google), “Collaborative Innovation Shaping Global Development.” January 2026. Google’s Global Head of Government Affairs on partnership models for AI in development. Watch

Spotlight Interview: Bosun Tijani, Minister of Communications, Nigeria, “Building Capacity to Harness AI for Shared Prosperity.” January 2026. Nigeria’s digital economy minister on national AI capacity-building. Watch

Spotlight Interview: Jamie Zimmerman (Gates Foundation), “Small AI with Big Impact.” January 2026. Gates Foundation’s Deputy Director, Digital Connectivity, on small AI for global development. Watch

The Next Frontier in Policy Decisions: AI-Powered Causal Insights, Part 1: Why We Built ImpactAI. Technical Brief, 2025. On translating rigorous causal evidence into smarter, more reliable policy decisions. Read

Context-specific AI for Low-resource Languages, Part 1. Technical Brief, September 2025. How our context-specific AI model for Nigerian social media dramatically improved performance over generic frontier models. Read

Context-specific AI for Low-resource Languages, Part 2. Technical Brief, September 2025. How ad-hoc benchmarks vastly overestimate AI performance in low-resource contexts, and why representative evaluation datasets matter. Read

AFW LEADS Workshop. August 2025. The AI Lab’s recap from the workshop in Togo, where we showcased ImpactAI and deepened collaborations with regional stakeholders. Read

Togo Data Lab. July 2025. On Togo’s pioneering use of AI and data tools to improve public services, showcased at the 2025 LEADS Workshop. Read

Nature, “Can AI help beat poverty?” February 2025. Feature on AI for the world’s poorest. Read

Spotlight Interview: Hear from our Program Lead, Samuel Fraiberger. December 2024. On how the lab leverages our partnership with Google to build impactful AI solutions. Watch

Apolitical Government AI 100 (2025). December 2024. Apolitical’s annual list of public servants leading on AI in government. Read

Understanding Socioeconomic Disparities in Mobility Behavior During the COVID-19 Pandemic. DIME AI Blog, November 2024. Insights on mobility patterns and inequality in developing countries. Read

Exploring Generative AI for Development Impact. DIME AI Blog, October 2024. Learnings from a high-level panel hosted by DIME, supported by Google.org. Read

Green shoots, AI roots: Responding to crises and building sustainable futures with gen AI. Google Cloud, October 2024. On nonprofits, including DIME AI, building AI for global challenges. Read

AI for Social Good E-Book. Google, October 2024. Google’s showcase of how nonprofits, including DIME AI, are tackling societal issues with AI. Read

1,302 real-world gen AI use cases from the world’s leading organizations. Google Cloud, September 2024. ImpactAI featured among leading gen AI deployments. Read

From Research to Action: Introducing ImpactAI at the Google.org Impact Summit. DIME AI Blog, September 2024. ImpactAI’s debut at Google.org’s flagship summit. Read

21 nonprofits join our first generative AI accelerator. Google.org, March 2024. ImpactAI joins Google.org’s inaugural GenAI Accelerator. Read

Axios, “AI Can Now Forecast the Next Food Crisis.” March 2023. Coverage of the Science Advances paper on predicting food crises from news streams. Read

Predicting food crisis outbreaks using natural language processing of news streams. World Bank Data Blog, 2023. How non-traditional data sources can improve food crisis forecasts in data-scarce environments. Read

Reading between the lines: Leveraging news data for AI-powered food insecurity forecasts. VoxDev, 2025. How AI-powered news analysis improves food crisis forecasting. Read

Events

Hype or Help: Scaling AI for Development Under Real-World Constraints. World Bank Annual Meetings, April 2026. Exploring what it takes to build AI systems that are accurate, trusted, and actionable for development decision-making. Watch

Unlocking the Power of Science & AI for Development Impact. World Bank Annual Meetings, October 2025. Public launch event for ImpactAI, convening leaders from policy, research, technology, and philanthropy. Details

AFW LEADS Workshop. Lomé, Togo, May 2025. Regional workshop convening pre-selected World Bank and IFC-financed projects across Africa West, where the AI Lab showcased ImpactAI. Details

UNICEF Global Impact Evaluation Forum. December 2024. Invited talk on AI and program evaluation methodologies. Details

Google.org AI for Social Good Webinar. December 2024. Featured speaker at Google’s flagship AI-for-good webinar. Watch

GenAI 4 Development Impact. World Bank, October 2024. Hosted by the World Bank with support from Google.org, bringing together leaders to explore generative AI for development. Details

Measuring Development 2024: AI, the Next Generation. May 2024. The tenth annual conference, co-hosted with DECDG, CEGA, and the University of Chicago. Includes “ImpactAI: Impact Evaluation Using Generative AI” (Watch) and “Detecting Hate Content” (Watch). Details

Food and Humanitarian Crises: Science and Policies for Prevention and Mitigation. Pontifical Academy of Sciences, Vatican, May 2023. Invited workshop on AI for early warning of food crises. Watch

Recognition

ACL 2025 Outstanding Paper Award, for HateDay, at the leading international conference in computational linguistics.

Google.org GenAI Accelerator: selected for the inaugural 2024 cohort of Google.org’s $20M Generative AI Accelerator.

World Development Report 2026 on Artificial Intelligence: Core contributor to the World Bank’s flagship publication on AI.

Contact

For partnerships, demos, or general inquiries: impactai@worldbank.org

Back to Programs

Get to know DIME AI

We are a multidisciplinary team of AI engineers, development economists, and software developers using impact evaluation and AI for social good.

Meet Our Team

Explore ImpactAI

The World Bank's Development Impact Group is using generative AI to help policymakers and practitioners access development economics research and make better policy decisions.

Learn More

Impact Intelligence: DIME AI Blog

Explore the DIME AI Blog for updates on our events, in-depth looks at innovative tools like ImpactAI, and the latest in research and projects driving data-driven solutions for global development.

Read DIME AI Blog