Unlocking the Potential of AI in Education Through Coordinated Research And Development

Through the AI and Education Program, Renaissance Philanthropy is channeling talent and resources to spark AI-powered innovations in education. As part of this program, we incubate and operate the Learning Engineering Virtual Institute (LEVI).

LEVI is a coordinated research program that brings teams together to achieve a singular moonshot goal: to double the rate of middle school math learning for low-income students. LEVI teams are expected to reach 1.5 million students in the next three years.

We sat down with Kumar Garg, President of Renaissance Philanthropy, to learn about the potential he sees for AI in education, and how LEVI is taking a collaborative approach to tapping into that potential.

As you think about recent advancements in AI, what opportunities do you envision for its impact on student learning?

The power of AI is that it is a prediction engine — and armed with growing amounts of data and ever more powerful foundational models, those predictions are getting better quickly. In education, there are many tasks that are prediction tasks: estimating what a student knows based on student work, identifying misconceptions by giving the option of picking the most likely wrong answer on an assessment, giving teachers tools to identify which students may be losing motivation based on lack of classroom conversation.

One example is the power of individualized tutoring, which is well-documented by researchers as highly effective. Unfortunately, it has been expensive to deliver at scale. But with the steady improvements in core machine learning technologies, AI-based digital tutors are getting better and delivering at lower cost. They are also now being paired with in-person tutors and classroom instruction. This can be a boon to teachers, students, and parents — allowing teachers more time to target their time on areas of need, and also focus on the social-emotional dimensions of learning for each student (motivation, belonging, sense of real-world relevance).

The education sector hasn’t fully tapped into the possibilities of AI. What’s holding it back?

For one, education R&D is woefully underfunded. Unlike other sectors like health, defense, and energy, less than 2 percent of federal education spending is on R&D — meaning that education is a lagging adopter, rather than a leader in technical innovation.

A second reason is that education research has been lagging compared to the other sciences in its use of big data, and rapid experimentation. Until very recently, there was no pre-built R&D infrastructure and data where education researchers could run multi-arm trials on larger segments of students to see how different educational strategies compared with each other — including key sub-populations that have different needs.

This is why we have set the explicit goal to build the learning engineering community — which sits at the intersection of learning science, computer science (AI, big data), and an ethos of continuous experimentation and improvement. We need to nurture talent that is able to keep up with the cutting edge of AI, and be aware of the challenges and opportunities unique to how education is delivered.

The surge in AI capabilities and AI interest gives the education field a huge opportunity — to recruit that bi-directional talent, leverage AI’s growing capabilities to test educational ideas much faster, and put much educational data to use on behalf of parents, teachers, and learners. Our view is that philanthropy can play a critical role in jumpstarting this work – by creating more doors for talent to enter, by identifying big opportunities to go after, and by building the infrastructure for ongoing progress.

Tell us about the Learning Engineering Virtual Institute, or LEVI. Why is Renaissance backing coordinated research programs?

LEVI was started with an ambitious-but-straightforward goal: to drive the development of an educational intervention that could double the rate of middle school math progress for low-income students, do so at a price that was scalable (under $1,000 per child), and with wide enough reach to show that the model was replicable.

AI was a big reason we thought this ambitious goal was possible — whether pairing AI with well-studied tutoring approaches to drive scale at lower cost, or other ideas from the field. We designated seven teams in the effort, and have made strong progress in the two years since — with one of the teams showing early signs of the doubling effect.

The LEVI model is all about building an infrastructure for research and innovation that drives collaboration toward a moonshot goal. Through multi-year funding and an interdisciplinary framework, programs like LEVI provide both the flexibility and stability needed to adapt to field changes and focus on solving a single, complex challenge. Working within a defined period of time, LEVI teams have a sense of urgency and shared mission. We’ve seen that, in the virtual institute model, teams are driven to work efficiently and rigorously toward one overarching goal.

Carnegie Learning’s MATHstream platform | Video by Carnegie Learning, a LEVI team

Tell us more about the role of LEVI “hubs” and why collaboration is so important to the LEVI model.

Collaboration is at the core of the LEVI model. LEVI’s interdisciplinary nature allows the participating teams to exchange ideas and solve problems in ways not typically seen in traditional academic settings.

Hubs are also essential to LEVI’s structure. These are partners who provide specialized support and resources to accelerate the teams’ impact. For instance, the Evaluation Hub, run by J-PAL North America, helps with evaluations, ensuring consistency across interventions.

Teams and hubs work closely, sharing datasets, co-authoring research papers, hosting workshops, and more. Hubs also create public goods, like published academic research and publicly available strategies for procurement, that benefit both LEVI teams and the broader educational field.

Can you share a bit more about the challenges the current LEVI program aims to address and how you arrived at LEVI’s current moonshot?

Math is a strong predictor of academic success, like high school graduation, and future economic mobility. Despite this, math performance in the U.S. has dropped significantly — particularly among students who have been historically marginalized. The pandemic only made things worse. We’ve seen this most acutely in the decline in 8th grade NAEP math assessment scores.

This stood out as an urgent issue in the field of education that needed a fresh, more coordinated response — especially given that traditional interventions have struggled to reverse declining math proficiency.

NAEP eighth-grade math proficiency trends from 1990-2022 | Data from The Nation’s Report Card

What outcomes has LEVI generated?

In just two years, the majority of teams are already on track to achieve LEVI’s goal of doubling the rate of middle school math progress.

Rising Academies showed that, in a group of approximately 500 students, those in the treatment group improved their scores by 5 points from baseline to endline, more than doubling the improvement seen in the control group, which increased by only 2 points.

Carnegie Mellon University reported a 1.8x increase in learning for 8th graders at Moss Side Middle School, while Huntington East Middle School saw a 1.9x improvement on state tests with MATHia.

Together, LEVI teams have reached over 350,000 students, and more than half of these learners come from low-income backgrounds. All of the teams are making good progress on LEVI’s affordability goals.

The LEVI program also requires grantees to implement a public goods strategy. This is to ensure that aspects of their work are made open and accessible, benefiting both students and the learning engineering field at large. As one example, in 2023, one of the LEVI teams, Eedi, released a public dataset involving more than 5,000 students’ open-ended responses to math videos. That data could have a huge impact in helping others understand how different strategies affect learning online.

What insights would you share with others about running a coordinated research program?

The key is anchoring the program in a clear, strong, and ambitious goal statement. It's crucial that the goal be rooted in a real, evidence-based problem. It’s also important to take into account key dimensions of success and define them clearly — like affordability, scaling, and trialing.

Leading a program with international reach requires certain considerations. LEVI teams work with populations around the world, from Chicago to Ghana. This has allowed us to test and refine interventions in different contexts, but has also meant we’ve had to pay attention to applicability in assessment measures, scaling and procurement processes, and implementation costs.

Another important element to consider is how you source applications. LEVI was an invitation process to a smaller cohort of 50+ teams which were invited to apply, which set a high bar in terms of teams’ expertise. This is complementary to but different from the Tools Competition — another project of the AI and Education Program — which is an open process and has a lower barrier to entry.

How will the LEVI model continue to evolve?

We’re excited to take all the learnings gained from the first iteration of LEVI into more LEVI-like programs that set ambitious goals in education and drive coordinated R&D to unlock them. As part of that, we issued a public Request for Information for ideas from the field earlier this year, and are designing promising future directions from that input. Here are some key themes that came out of that RFI.

 Learn More

The first iteration of LEVI brought together seven teams – Carnegie Learning, Carnegie Mellon University, Eedi, Khan Academy, Rising Academies, University of Colorado Boulder, and University of Florida – to develop AI-based interventions designed to improve middle school math outcomes for low-income students.

Each team developed a five-year plan to achieve the institute’s goal, which includes deep collaboration across institutions and disciplines, and a strong ethos of building and experimentation. Learn more about each LEVI team here.