J-PAL’s Project AI Evidence: Testing AI to Reduce Poverty
Overview: What is Project AI Evidence (PAIE)?
MIT’s Abdul Latif Jameel Poverty Action Lab (J-PAL) has launched Project AI Evidence (PAIE), a global research and policy initiative to rigorously test how artificial intelligence can reduce poverty and to guide responsible scaling of effective tools. The program connects governments, nonprofits, and technology companies with economists to run randomized evaluations and related studies across education, health, climate, and social protection [1][2]. J-PAL Project AI Evidence matters because it brings scientific rigor to fast-moving AI deployments where stakes are high and public resources are limited [1][2].
Why PAIE matters for governments, NGOs, and funders
Public agencies and social-sector organizations face pressure to adopt AI but lack credible, context-specific evidence on impact and risk. PAIE addresses this gap by prioritizing randomized evaluations, building partnerships through J-PAL’s regional offices, and creating a clear path from pilots to policy at scale. This approach informs procurement, supports return-on-investment decisions, and helps manage risks tied to inclusion, safety, and effectiveness—core needs for responsible AI scaling [1][2][4][5].
Priority research areas and questions
PAIE’s early priorities target entrenched challenges and high-stakes use cases:
- Can AI-assisted teaching improve learning for all students?
- Do AI-enabled early-warning systems better protect people from floods and other disasters?
- Can machine learning reduce Amazon deforestation?
- Can AI chatbots and mobile apps improve health and mental health outcomes?
These questions reflect a cross-sector agenda spanning education, climate resilience, environmental protection, and health—areas where robust causal evidence can accelerate or restrain adoption based on real-world outcomes [1][2].
Examples of funded projects and early pilots
PAIE has already funded at least eight research projects across geographies and sectors, including:
- An AI-based mental health app for immigrants in the United States.
- AI-assisted mobile platforms for maternal health in Kenya.
- Smartphone-based infant weight estimation tools in India.
- Machine learning approaches to cheaper predictive tests for elderly health screening in Tamil Nadu.
These efforts illustrate how the initiative leverages J-PAL’s regional infrastructure to enable implementation support and policy partnerships while stress-testing AI in diverse contexts [1][2].
J-PAL Project AI Evidence: methodology and staged pilots
Randomized evaluations are central to PAIE’s strategy for establishing causal impacts of AI interventions on poverty-related outcomes. The initiative runs recurring funding competitions that include pilot grants for refining models, conducting in silico evaluations, and testing engagement strategies. Promising pilots can then move into larger randomized evaluations with potential policy impact at scale—a disciplined path designed to minimize harm and inform responsible AI scaling [1][2][3].
Eligibility, partnerships, and how to engage
Participation is limited to J-PAL affiliates, selected collaborators, and invited researchers, with specific eligibility rules (including those for students) and requirements for coordination with J-PAL regional offices. Governments and social-sector organizations can collaborate with PAIE to pilot AI within public programs, aligning operational needs with rigorous evaluation plans and potential scale-up pathways [2][3].
For organizations exploring AI policy partnerships or PAIE funding opportunities, J-PAL provides clear guidance via its initiative page and request-for-proposals materials [2][3].
Ethics, safety, and deciding what to scale
PAIE’s broader goal is to identify which AI innovations work, for whom, and under what conditions—and to ensure only effective, inclusive, and safe tools are expanded. Interventions found ineffective or risky are intended to be scaled down or avoided, not promoted. This evidence-to-policy discipline builds on J-PAL’s long-running efforts to help governments use data and rigorous evaluations in program design and implementation [1][2][4][5].
Implications for businesses and tech teams
For AI builders in education, health, climate, and social protection, PAIE sets clear evidence expectations: iterate via pilots, test models in silico where feasible, and demonstrate causal impact through randomized evaluations before pursuing scale. Companies looking to collaborate with governments and NGOs should anticipate structured learning agendas, outcome measurement, and implementation support through regional policy partnerships [1][2][3].
Organizations seeking broader context on AI governance and deployment can also explore J-PAL’s institutional work and mission via J-PAL (external) while keeping up with our coverage of the space in our AI news hub [2][4].
Next steps and resources
To learn more or explore PAIE funding opportunities, see the Project AI Evidence initiative page and the RFP overview. For program context and recent developments, read the MIT News coverage [1][2][3].
Sources
[1] New J-PAL research and policy initiative to test and scale AI innovations to fight poverty
https://news.mit.edu/2026/new-j-pal-research-policy-initiative-to-test-scale-ai-innovations-fight-poverty-0212
[2] Project AI Evidence (PAIE) | The Abdul Latif Jameel Poverty Action Lab
https://www.povertyactionlab.org/initiative/project-ai-evidence-paie
[3] Project AI Evidence: RFP Overview
https://www.povertyactionlab.org/sites/default/files/251205_PAIE_RFP_Overview_0.pdf
[4] J-PAL expands evidence-to-policy government partnerships to fight poverty
https://news.mit.edu/2022/j-pal-expands-evidence-policy-government-partnerships-0809
[5] J-PAL North America and Results for America announce 18 collaborations with state and local governments
https://news.mit.edu/2023/j-pal-north-america-results-america-announce-collaborations-18-state-local-governments-0929
