The Learning Engineering Virtual Institute, or LEVI, is a collaboration of researchers, learning engineers, and educators striving to develop, scale, and implement new tutoring platforms that can double the rate of math progress among middle school students, especially those from low-income backgrounds. Since 2022, seven teams have been working to achieve this goal. Their ideas range from an AI-powered chatbot that provides personalized math tutoring to AI video technology that digitally replicates the experience of having a personal tutoring session.
One team, PLUS–Personalized Learning Squared, blends human and AI tutoring to support middle school math students during the school day. Human tutors provide motivation, guidance, and math help when needed, while AI tutors offer personalized practice. Launched by Carnegie Mellon University, in partnership with Carnegie Learning and Stanford University, PLUS connects college tutors with classrooms virtually, reaching thousands of students each week across Pennsylvania, West Virginia, and New York. In this “5 Questions” feature, research team leader Dr. Danielle Thomas shares what makes PLUS work – and what’s next.
What was your “ah-ha” moment when you knew you were on to something workable?
We know that AI tutors can improve math learning, but many students don’t use them enough. We also know that human tutoring works but is costly. At PLUS, we combine human and AI tutors to offer effective and affordable tutoring to students who need it most. After piloting a small-scale human-AI tutoring program right before the COVID pandemic and observing a near-doubling of math learning among middle school students in a low-incoming serving urban school, we knew we were onto something–sparking the development of PLUS. Now, in the middle of year 3, we’ve provided affordable tutoring to nearly 3,000 middle school students and are making progress towards learning improvements at scale.
Have you made any significant shifts or course corrections since you started working with LEVI?
In the beginning, we were concerned about our ability to scale up our program to meet our ambitious goals in terms of the size of our student footprint. As it turns out, the strong interest from partner schools and the unexpected enthusiasm from Carnegie Mellon University students to work as tutors enabled us to expand the size of our tutor corps and the scale of our student footprint ahead of schedule, with ample momentum for further growth. That success has enabled us to redirect our efforts toward raising the quality and consistency of our students’ learning experience and ensure that our tutoring makes a real difference in student learning–all while keeping the costs low. This shift is helping us achieve both the desired scale and the desired impact.
What’s been the most surprising thing your team has learned thus far?
The most surprising thing we’ve learned so far is how powerful the combination of human tutors and AI-driven learning tools has been, especially for students who need the most help. We knew our approach could make a difference, but the impact we’ve seen on students below grade level and those with disabilities is particularly promising. For example, in one middle school, students who were behind in math showed nearly twice the learning gains compared to their peers who didn’t receive tutoring. In another case, students with disabilities who worked with our tutors practiced more skills and achieved proficiency at a much higher rate than those who used only the math software. These results show us that we’re on the right path to significantly improve learning outcomes for students who often face the greatest challenges.
When did you see your tool working for students and teachers?
One of the benefits of combining human tutors with AI-powered learning tools is that these systems constantly track students’ learning effort and progress. Thus, we can quickly identify when a student needs more support–well before classroom assessments and standardized test scores are available. The data from these tools helps us compare how students who have both a human tutor and AI-driven edtech are doing compared to those who are just using the technology alone. By the middle of our second year, we had sufficient data that we could statistically quantify positive effects of our program on students’ learning progress. We see evidence that our intervention is especially impactful for disadvantaged student groups, and could therefore contribute to the narrowing of opportunity gaps. Although exciting, we are dedicated to establishing the credibility of these results through further experiments. Importantly, we’ve also learned that supporting tutors in relationship-building strategies is critical to this work, and that social-emotional learning is intertwined with math learning.
What do you anticipate your project accomplishing in five years?
We are increasingly confident that the PLUS approach can achieve large gains in learning outcomes at scale – in other words, the ambitious goals we set in our proposal of a doubling of learning for 10,000 students for less than $1K per student per year will be achieved. Advancing technologies, particularly with generative AI, bring new possibilities. For example, our tutors are now receiving real-time AI feedback during training to prepare for working with students and to continuously improve their skills. Within two years, we made that possible. While it’s hard to predict exactly what PLUS and other tutoring providers will develop in the next five years, we know it will center on supporting teachers in their practice and increasing student access to high-quality learning experiences at scale. That’s at the core of what we do, and we’re excited for the possibilities ahead.
Kent Fischer
Director of Communications