To explore the latest in artificial intelligence applications in education, consider the work of Jing Liu, director at the University of Maryland’s Center for Educational Data Science and Innovation. Two October studies provide contrasting insights into AI’s impact on learning. One study suggests that a bit of AI-driven math practice could boost middle schoolers’ math skills, while another reveals that an AI assistant might reduce college students’ engagement and lower their grades.
Liu intentionally released the studies together to illustrate that AI in education is a mixed bag. The ongoing challenge is to develop tools that enhance learning continually. This involves analyzing not only the technology but also student interactions, including unexpected outcomes.
The first study involved Goblins, an AI math practice website, introduced to 6,000 middle school students in Deer Valley, Arizona, during the 2025-26 school year. Teachers assigned math problems, while an AI tutor provided step-by-step assistance through text, drawings, and voice chat. Although the tool was mainly for struggling students, some teachers applied it to entire classes, while others avoided it.
Students solving at least two problems weekly—over 50 in the school year—scored higher on year-end standardized math tests than those who used Goblins less. Accelerate, a nonprofit that commissioned the study, equates this improvement to three extra months of learning. However, the study can’t confirm Goblins as the cause of these gains, as motivated students might naturally perform better. A randomized controlled trial is planned for further investigation.
Notably, 50% of students engaged with Goblins, addressing the “five-percent problem” where only a small fraction of students typically use educational tools as intended. This problem is well-documented in platforms like Khan Academy and IXL. Goblins’ success included engaging English learners and students with disabilities, partly due to its interactive avatar and multilingual capabilities.
Teachers retained control over assignments, choosing which problems to assign, echoing successful strategies from platforms like ASSISTments, which also showed positive outcomes with teacher-assigned practice and instant feedback.
When AI Access Backfires
The second study evaluated an AI assistant at the University of Maryland. Before a campus-wide rollout, Liu conducted a trial during the fall semester of 2025. Half of the 30 professors used the AI assistant in their classes, while the other half did not, spanning subjects like English, math, biology, and social sciences.
Professors uploaded syllabuses, readings, and assignments to the university’s AI chatbot, accessible 24/7 through Canvas. Despite its availability, students in AI-assisted courses had lower grades, a 0.11-point decrease on a four-point scale, and interacted less with course materials and faculty. The assistant’s presence possibly encouraged students to use external AI tools like ChatGPT, affecting their learning negatively.
Provost Jennifer King Rice noted, “AI tools could exacerbate inequities in learning outcomes,” particularly concerning for first-generation students. The lesson isn’t to avoid AI but to recognize unintended consequences. Maryland’s transparency in sharing these findings offers valuable insights for the educational field.
—
Read More Kitchen Table News
