New research shows that artificial intelligence (AI) tutoring can help students improve their GRE test scores just as effectively as human tutors, but at a dramatically lower cost. This finding is significant because it suggests that AI-powered educational tools could make high-quality tutoring more accessible and affordable for a wide range of learners. The study, based on a large-scale evaluation named StudentBench, compared learning gains from AI tutors, human tutors, and no tutoring across thousands of students preparing for the GRE, a standardized test commonly used for graduate school admissions.
Key Takeaways
- AI tutoring produced learning gains equivalent to expert human tutoring on GRE Quantitative and Verbal sections.
- In five of seven GRE subject areas tested, the top AI tutor outperformed human tutors on average.
- One AI tutor achieved similar effectiveness at 918 times lower cost per percentage point of learning gained compared to human tutoring.
- Faster AI response times correlated with increased student engagement and better learning outcomes in quantitative GRE practice.
The researchers introduced StudentBench, a new platform designed to evaluate AI teaching methods on a large scale by collecting over 175,000 student-AI interactions. This system allowed them to rigorously compare the effectiveness of AI tutors with that of expert human tutors and no tutoring at all. Participants included 2,383 students who received either AI tutoring, human tutoring, or no tutoring while working on GRE practice questions.
To measure learning gains, the study focused on improvements in GRE scores across different domains such as Quantitative reasoning and Verbal reasoning. The phrase “statistically equivalent” means that, based on the data, AI tutors and human tutors performed so similarly in boosting student scores that any differences were not significant enough to declare one better than the other. Notably, in several GRE subtopics, the best AI tutors actually surpassed human tutors on average.
Beyond just test scores, the study also assessed how well AI systems could create lesson plans and practice problems. Expert tutors conducted thousands of pairwise comparisons to evaluate the quality of AI-generated materials. This helped the researchers understand not only how AI tutors performed in conversation but also how they planned instruction and designed exercises, which are crucial components of effective teaching.
One striking finding was the cost efficiency of AI tutoring. The study calculated that the best AI tutor delivered learning gains at less than one-hundredth of the cost of human tutors. This huge difference stems from AI’s ability to scale easily and respond instantly to students without the expenses associated with human educators. Additionally, faster AI responses encouraged more student interaction, more practice with correct answers, and ultimately better learning outcomes, especially in quantitative subjects.
These results highlight the potential for AI to democratize access to quality education by supplementing or even replacing traditional tutoring in some contexts. However, the authors note that AI tutoring is not a perfect replacement for all educational needs and human interaction still plays an important role. The StudentBench platform is publicly available, inviting further research and development in AI-driven education tools.
Looking ahead, this research opens the door to more affordable and scalable tutoring solutions that could benefit millions of students worldwide. As AI teaching systems continue to improve, they may become valuable partners in personalized learning journeys, helping to close gaps in educational opportunity and support lifelong learning.
Based on research published on arXiv by Curtis Northcutt, Inaara Hasmani, Kevin Feng et al..