AI In Education Excerpt 3: AI in Tutoring: The Benefits of Personalization
This article is the third in a series about AI and education by Andrew Rosston. See previous articles about AI as a Learning Aid and AI and Plagiarism.
Beyond use of AI by individual students and for classrooms, tutors can use AI to help students with unique challenges and learning styles. Personalization is a major potential benefit of AI, in adapting work to suit students’ strengths and weaknesses. This will likely build on previous work in earlier stages of AI and algorithmic development which could create personalized learning tracks such as those seen in some language learning programs. A creative tutor can make use of existing Large Language Models (LLMs) to generate materials and problem sets for students, but more specialized tools will greatly aid in reinforcing difficult concepts.
Though many AI programs are designed to act as personal assistants, there is an opportunity to construct education-specific programs, and existing LLMs may be converted at minimal cost. Work on Intelligent Tutoring Systems predates present-day AI, and such developments will continue parallel to other AI developments.
At the same time, there is reason for caution. In a 2024 study, Kestin et al. chose to test generative AI that had been specifically prompted to act as tutors, warning that LLMs are not specifically tailored for education and tutoring. The study also warns of the confidence in which AI LLMs deliver answers, despite their imperfect accuracy and tendency to hallucinate or improperly reference information.
While it is a relatively new field, some data now exists on the efficacy of AI tutoring. Some early research suggests that AI tutoring is more effective than human tutors in certain contexts.
One study of 70 medical students at McGill University found that a remote human expert was less effective than an AI in teaching surgical skills. This effect may be due to the AI’s ability to quantify criteria in learning such skills, which would apply to some fields (e.g. mathematics) more than others (e.g. writing). As this is a rapidly changing field, it is likely that the body of research will grow quickly in coming years, which may confirm certain uses as more or less helpful.
All but one student in the AI-tutored group in the McGill study indicated a preference for instruction by a combination of AI and a human instructor. This surgical training study notes that the AI cannot assist with interpersonal skills, such as “bedside manner” and cooperation with other medical personnel, and that comparison was not made with in-person tutors during the Covid-19 pandemic.
Some research has shown that AI tutoring can even outperform live lecture-based instruction, which may make it a key future learning aid. This may be especially important for families with fewer financial resources to gain the benefits of a human tutor at a lower cost.
University students also reported positive experiences with AI in Gyonyoru & Katona (2024) and Kestin et al. (2024), though with more neutral responses on data privacy and personalization in the former study. Use by undergraduate students, who are often more self-directed, may not be directly comparable to that of younger students or study of certain subjects.
Time spent on learning with AI did not affect test scores despite varying wildly (Kestin et al. 2024), implying that students benefit from learning at their own pace with personalized instruction, an effect which may apply outside of AI instruction. Notably, many of the benefits of AI over those of normal instruction appear related to personalization, including allowing students to learn at their own pace and developing personalized learning tracks.
Comparisons with human tutoring and personalization require further research to determine whether AI tutors are superior or can augment such instruction, as much research so far appears to focus on their benefits over generalized instruction such as lectures. This remains a topic for further research and experimentation, as well as a theoretical market opportunity for low-cost educational products, which will provide data in turn.
It should be noted that AIs specialized for use by ELL students and students with disabilities are their own distinct topics. These topics are touched on in the full paper but are deserving of specific discussion in their own right.
Andrew Rosston is a Business Analyst at OnlyMoso USA. He holds a B.A. in Business and Managerial Economics from Oregon State University.
