MIT Lab Launches AI Testing to Fight Global Poverty

A new MIT research initiative is putting artificial intelligence to the test where it matters most: helping people escape poverty.

The Abdul Latif Jameel Poverty Action Lab at MIT just launched Project AI Evidence, funding eight studies to figure out which AI solutions actually work for people in need. The goal is simple but powerful: scale up what helps, shut down what doesn't.

Since 2003, J-PAL researchers have evaluated over 2,500 social programs worldwide. Now they're bringing that same rigor to the AI boom, partnering with governments, tech companies, and nonprofits to answer the questions policymakers are already asking.

In Kenya, researchers are testing whether an AI teaching tool can help educators spot where students are struggling and adapt lesson plans on the fly. Personalized learning works, but teachers rarely have the resources to pull it off alone.

Another study in Kenya explores whether AI can help career counselors find job opportunities for people without formal education. The tool identifies overlooked skills that might unlock employment, especially for youth and women.

In Italy, researchers are working with the Ministry of Education to test if AI can reduce gender bias in classrooms. Two tools will be evaluated: one helps teachers predict student performance more fairly, while another gives real-time feedback on whether they're treating students equally.

The initiative tackles climate challenges too. Studies will examine whether early-warning flood systems powered by AI can better protect communities from natural disasters, and if machine learning can help reduce deforestation in the Amazon.

The Ripple Effect

Project AI Evidence represents a shift in how we think about AI adoption. Rather than rushing to implement every new tool, the initiative creates a testing ground where evidence comes before scale.

Google.org, Community Jameel, Canada's International Development Research Centre, and Amazon Web Services are all backing the project financially. Eric and Wendy Schmidt contributed funding specifically to study generative AI in workplaces across low and middle-income countries.

Alex Diaz from Google.org explains the urgency: "AI has great potential to benefit all people, but we need to study what works, what doesn't, and why."

The research focuses on making sure AI tools don't just work for some people—they need to work for everyone, including those with the fewest resources. That means testing whether an AI tutor helps both wealthy students and those in underfunded schools, or if it widens existing gaps.

J-PAL will run regular funding competitions in coming years, inviting researchers to propose evaluations of AI tools addressing education, health, economic opportunity, and climate challenges. Each study will measure real outcomes: Are students learning more? Are people finding jobs? Are communities safer from disasters?

The message is clear: AI's promise is enormous, but only if we're willing to test it honestly and share what we learn.