Argomenti trattati
J‑pal launches Project AI Evidence to test AI interventions for poverty reduction
The Abdul Latif Jameel Poverty Action Lab (J‑PAL) at MIT has launched Project AI Evidence (PAIE), a research program to evaluate how artificial intelligence can be applied to reduce poverty and inequality. The initiative will produce rigorous, evidence‑based assessments of which tools deliver measurable benefits, for which populations they work, and what safeguards are necessary before scale‑up.
PAIE brings together policymakers, technology platforms, nonprofit implementers and economists to design and run randomized evaluations and other robust studies. The program aims to move beyond broad promises about AI by generating practical findings that can inform policy and implementation decisions.
The program operates through targeted funding competitions that commission evaluations of real-world AI applications. These competitions prioritize research questions that governments and practitioners have already identified as urgent. The aim is to produce findings that can directly guide policy and operational decisions.
Structure, partners, and priorities
PAIE’s early portfolio funds studies in education, gender bias, employment, disaster response and environmental protection. Partners include philanthropic organizations, public agencies and private funders. The initiative stresses inclusive, responsible scaling of tools that demonstrably help vulnerable populations.
Project selection emphasizes practicality and demand-driven research. Proposals compete for funding on the basis of their ability to generate usable evidence for implementers. Results are intended to move debate from general claims about AI toward specific, actionable insights for policymakers and practitioners.
Funding and partnerships
PAIE is funded by a coalition of philanthropic, public and private partners.
The programme has received a grant from Google.org and philanthropic contributions from Community Jameel.
It also draws on funding from Canada’s International Development Research Centre (IDRC) and support from UK International Development, alongside a collaboration agreement with Amazon Web Services.
An additional grant from Eric and Wendy Schmidt, recommended by Schmidt Sciences, will fund research on generative ai in workplace settings in low- and middle-income countries.
This coalition is intended to combine domain expertise, implementation capacity and financial resources to support rigorous, policy-relevant evaluations.
The coalition is intended to combine domain expertise, implementation capacity and financial resources to support rigorous, policy-relevant evaluations. PAIE is chaired by Professor Joshua Blumenstock (University of California, Berkeley), J‑PAL Global Executive Director Iqbal Dhaliwal, and Professor David Yanagizawa‑Drott (University of Zurich). Their mandate is to prioritize research questions that policymakers are actively asking and to ensure results are both policy‑relevant and ethically grounded.
Research themes and initial studies
The coalition’s first round of funded evaluations focuses on areas where AI could change outcomes rapidly. The work follows a push for policy‑relevant, ethically grounded research from earlier stages of the initiative. Funded studies will test whether AI tools can improve measurable student outcomes and support teacher decision‑making in real classrooms.
Education and teacher support
In education, researchers will evaluate AI‑assisted systems designed to personalize learning and to inform teacher choices about instruction. Partners include Kenya’s social enterprise EIDU and India’s NGO Pratham. Both organisations are piloting classroom applications intended to identify learning gaps and adapt instruction at scale.
Evaluations will measure both learning gains and practical adoption barriers. Assessments will examine accuracy of gap detection, the relevance of recommendations for teachers, and classroom workflow impacts. Findings are intended to guide policymakers, funders and school systems considering wider deployment.
Findings are intended to guide policymakers, funders and school systems considering wider deployment. PAIE-funded teams will test whether AI tools increase teacher productivity and improve student outcomes when integrated into routine lesson planning and established pedagogical models.
Research teams, including Daron Acemoglu, Iqbal Dhaliwal and Francisco Gallego, will measure effects in real-world school settings. The studies will compare AI-augmented lesson preparation with standard practice. Analysts will track teacher time use, lesson quality and measurable learning gains. They will also assess implementation costs and scalability.
Addressing gender bias and labor-market access
Researchers will examine whether AI systems reproduce or reduce existing biases. They will test for differential impacts by gender, socioeconomic status and other demographic factors. The work will explore how AI-informed instruction affects students’ acquisition of skills linked to labor-market opportunities.
Key questions include whether AI tools broaden access to career-relevant content and credentials. Teams will evaluate effects on guidance and pathways to employment for historically underserved groups. They will monitor for unintended consequences, such as tracking into narrower curricular options or reinforcing stereotypes.
Studies will combine quantitative outcome measures with qualitative evidence from teachers, students and employers. Results will inform recommendations on design, oversight and mitigation strategies to limit bias and protect equitable access to future labor markets.
Building on findings that will shape recommendations for design, oversight and mitigation strategies, researchers are testing whether AI can reduce unconscious bias and expand job opportunities.
In partnership with Italy’s ministry of education, affiliates including Michela Carlana, Will Dobbie, Francesca Miserocchi and Eleonora Patacchini will trial tools that give teachers predictive and real-time feedback. The tools aim to help teachers identify and correct biased decisions that affect male and female students.
Separate work in Kenya, conducted with NGOs Swahilipot and Tabiya, will evaluate an AI-driven career guidance tool. Investigators such as Jasmin Baier and Christian Meyer will measure changes in job search behaviour and subsequent labour market outcomes for youth, women and people without formal qualifications.
Why rigorous evidence matters
Rigorous, independent evidence is essential to assess whether these tools deliver on their promises. Policymakers and school systems need reliable measures of effectiveness before scaling deployments.
Evidence also clarifies potential harms. Systematic evaluation can reveal where AI amplifies rather than reduces bias. It can also identify barriers to fair access for vulnerable groups.
Finally, robust findings inform practical oversight and mitigation strategies. Clear evidence guides funding decisions and technical design choices that protect equitable access to future labour markets.
Clear evidence guides funding decisions and technical design choices that protect equitable access to future labour markets. The Abdul Latif Jameel Poverty Action Lab (J-PAL) brings a long record of evidence-driven policy to this effort. Since 2003 its network has produced more than 2,500 randomized evaluations of social programs worldwide. PAIE applies that same experimental ethos to artificial intelligence to assess which applications improve outcomes and which require redesign.
J-PAL leaders say the initiative prioritizes careful scaling over rapid deployment. The emphasis is on identifying interventions that can be expanded responsibly while flagging those that present equity or effectiveness concerns. Funders and partners reinforce this mission. Google.org describes the collaboration as an effort to establish what works and why. The IDRC stresses research that is sensitive to local contexts to ensure AI innovations remain safe and relevant.
Next steps and collaboration opportunities
PAIE plans targeted field trials, partnerships with service providers, and capacity building for local researchers. These activities aim to generate replicable evidence on performance, distributional impacts and implementation challenges. Funders are expected to use results to prioritize investments and oversight mechanisms that protect vulnerable groups.
Researchers and implementers are invited to propose study designs and data-sharing arrangements that align with PAIE’s transparent evaluation standards. Collaborations that combine randomized evaluations with mixed-methods research are particularly valuable for diagnosing why interventions succeed or fail. The initiative will publish protocols and findings to inform policymakers, funders and developers about scalable, context-sensitive AI applications.
Project update and opportunities for collaboration
The initiative will run additional calls for proposals and expand its evaluation portfolio over coming years. It invites governments, non-governmental organisations and technology developers to collaborate on study design, implementation and translation of findings into policy guidance.
J-PAL seeks partners interested in responsible AI adoption and rigorous field evaluation. The team plans to publish protocols and findings publicly so policymakers and implementers can make informed choices backed by empirical evidence.
Organisations and researchers wishing to participate or obtain further information should contact the PAIE team at [email protected] or subscribe to the J-PAL newsletter for updates. By combining multidisciplinary partnerships with systematic field evaluation, Project AI Evidence aims to identify contexts where AI can deliver measurable social benefits and where additional caution is warranted.