"The AI Tutor That Won't Give Answers": MathGPT.ai Expanding to Over 50 Universities
As the use of AI in education advances, concerns regarding student "cheating" and the increased burden on faculty have emerged. An educational support AI platform tackling these issues head-on is MathGPT.ai. Following a pilot implementation at 30 domestic universities last year, the number of institutions using the service is expected to exceed 50 this autumn. We will explain this service—which is being utilized as a tutor that "doesn't give answers but encourages thinking" for students, and as an efficient teaching assistant tool for faculty—by looking at its background, features, and educational impact.
1. Background and Overview
1-1. The AI Dilemma in the Classroom
With the spread of AI, dilemmas in the educational field are becoming more serious, such as the "dilution of the learning experience" where students dump assignments onto AI, and the diversion of faculty resources to prevent academic dishonesty. Amidst this, MathGPT.ai was designed to support faculty work without "relying too much" on AI. Following the success of the pilot program, the number of universities adopting it is expected to increase to over 50 this autumn.
1-2. Basic Structure of MathGPT.ai
MathGPT.ai functions as an "anti-cheating" AI tutor and teaching assistant. Even if students ask for the answer, the AI is designed not to provide direct responses, but to promote thinking through Socratic questioning. It primarily handles university-level mathematics (algebra, calculus, trigonometry, etc.), and the faculty tool allows for assignment creation, automated grading, and the development of teaching materials.
2. New Features Considering Both Faculty and Students
2-1. Strengthening Faculty Discretion and Easing the Learning Environment
Setting Usage Timing: Faculty can control when students are allowed to interact with the AI.
Limiting Number of Attempts: It is now possible to set the number of attempts allowed to reach the correct answer.
Unlimited Practice Problems: Provides practice problems that do not affect grades, improving the "sense of security as a place to learn."
2-2. Further Improvements in Reliability and Accessibility
Image Submission Feature for Work: Ensures integrity by requiring students to submit images of their answer sheets as proof that they solved the problems themselves.
LMS Integration: Integrates with major learning management systems such as Canvas, Blackboard, and Brightspace.
Features for Students with Disabilities: Introduces screen reader support, voice mode, and AI narration with subtitles (e.g., voices in the style of Benjamin Franklin or Einstein). Compliance with the ADA (Americans with Disabilities Act) is also claimed.
3. Stance on Ensuring Safety and Countermeasures Against Incorrect Answers
3-1. Restricting Inappropriate Conversations
While Meta AI and other chatbots have been criticized for inappropriate interactions with young users, MathGPT.ai is explicitly designed "not to respond to conversations about romantic relationships or life philosophy." Chairman Peter Relan states, "It is important not to have those kinds of conversations."
3-2. Efforts Toward Accuracy Improvement and Transparency
Incorrect Answer (Hallucination) Reporting System: Students who discover errors are rewarded with gift cards. The service boasts a reduction record of 5 reports in the first year of implementation, 1 in the following year, and zero this year.
Human-in-the-loop monitoring: A specialized team of humans checks AI outputs, explicitly aiming for "100% accuracy."
4. Future Outlook and Impact on the Educational Landscape
4-1. Expansion Plans—New Subjects and Mobile Support
Future plans include expanding subjects to include chemistry, economics, and accounting. Additionally, development of a mobile app is underway, aiming to provide an environment where learning is accessible anytime, anywhere.
4-2. Resonance with Educational Reform: Criticism and Potential
Economist Tyler Cowen argues that AI will not be a "tool to hide cheating," but rather a "mirror that highlights the limitations of current education." He suggests that "we need a shift from robotic education to mentor-based education, rather than just anti-cheating measures." From this perspective, thought-provoking AI like MathGPT.ai could serve as a litmus test for the future of education.
Summary
MathGPT.ai is evolving from an "AI that prevents cheating" into an "educational support AI that draws out thinking." Its design, which respects teacher discretion and encourages student autonomy, can be seen as an ideal form of future AI education. At the same time, it is clear that AI is not a panacea and that teacher involvement is essential.
Moving forward, MathGPT.ai may initiate a transformation from education that "relies on AI" to learning that "collaborates with AI to deepen thinking." It is attracting attention as a new model of educational infrastructure that offers high value not only to educators but also to policymakers and technology developers.