Researchers from the University of Toronto and the University of Pennsylvania’s Wharton School conducted an experiment with more than 6,000 middle school students in Tennessee to evaluate how artificial intelligence affects learning outcomes in mathematics. The study tested four different approaches to teaching fractions, with students randomly assigned to receive either conventional computer-based instruction or the same software enhanced with an AI tutor. Within each group, half the students were required to answer practice questions correctly three times consecutively if they made a mistake, a strategy known as mastery learning.
Students engaged with the software for a single 50-minute session during math class and were assessed one week later with a 15-minute retention test. The results indicated that the combination of AI tutoring with mastery learning produced the strongest outcomes, with students scoring approximately 3 percentage points higher than those receiving conventional computerized instruction alone. Philip Oreopoulos, the lead researcher and an economist at the University of Toronto, characterized the findings as preliminary evidence suggesting AI may have some positive educational value, though he cautioned against overstating the significance of the modest gains observed.
The researchers attribute the effectiveness to how the AI tutor engaged with student errors. Rather than simply displaying step-by-step solutions that students could quickly skim without fully understanding their mistakes, the AI system provided interactive guidance tailored to each student’s specific work. Students in the AI-plus-mastery group spent more time on individual questions and demonstrated higher accuracy on subsequent problems after making errors.
However, the study revealed important limitations. The benefits were most pronounced on easier questions closely resembling the practice material and did not extend to more complex problems, suggesting the learning may not have transferred deeply. The research also involved only a brief intervention, leaving open questions about long-term effects over extended study periods. Oreopoulos emphasized that the research represents an early exploration rather than a definitive conclusion about optimal AI implementation in education, noting that future work should continue testing different features to identify more effective approaches.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI