An Integrated Framework for Generative AI in Undergraduate Mathematics: Policy, Pedagogy, and Assessment
This conceptual paper introduces AI-EPE, a comprehensive framework that integrates policy, pedagogy, and assessment to guide the ethical and effective use of generative AI in undergraduate mathematics by defining key constructs like ethical competency and epistemic agency while offering level-specific guidelines and assessment strategies.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Mathematics has always been more than a collection of answers to be memorized. For students, the true value of the subject lies in the mental work of figuring things out: the struggle to understand a concept, the process of building a logical argument, and the responsibility of verifying a result. This intellectual effort is what educators call "productive struggle," and it is the engine that drives deep learning. However, a new kind of technology has arrived that can perform this mental work instantly. Generative artificial intelligence, the same type of system that can write essays or create images, can now solve complex math problems, construct formal proofs, and explain difficult concepts in seconds. This capability creates a profound dilemma for universities. If a machine can produce a perfect solution, how can a teacher know if a student actually understands the math? If a student uses the tool to bypass the struggle, they may graduate with a transcript full of correct answers but without the reasoning skills those answers are supposed to represent.
The challenge is not simply about detecting unauthorized use. It is about preserving the very thing that makes mathematics education valuable: the development of a student's own mind. When a student relies on a machine to do the thinking, they miss the opportunity to build the neural pathways required for logical reasoning. Yet, banning these tools entirely ignores the reality that students will encounter them in their future careers and lives. The question has shifted from whether to allow these tools to how to use them in a way that strengthens, rather than replaces, human thinking. Researchers Eunmi Joung, HyeJin Tina Yeo, Bitnara Jasmine Park, and Mia Kang have addressed this complex problem by proposing a new way to think about the relationship between policy, teaching, and testing in undergraduate math.
The authors introduce a framework called AI-EPE, which stands for Ethical Competency, Perceived Utility, and Epistemic Agency. These three ideas form the backbone of their approach. Ethical Competency is the ability to judge when it is right to use the tool and when it is not. It is not just about following a rule, but understanding the purpose of the learning task. Perceived Utility is the belief that the tool actually helps learning, rather than just making work easier. Finally, Epistemic Agency is the student's active responsibility to remain the captain of their own reasoning. It means that even when using the machine, the student must verify the steps, question the logic, and take ownership of the final result. The researchers argue that for mathematics education to survive this technological shift, these three conditions must be met at every level of the university, from the administration down to the individual homework assignment.
The paper outlines how these conditions should be applied through three distinct layers of governance. At the top level, the university administration sets the basic rules, such as protecting student data privacy and ensuring that no course requires students to pay for expensive software. This creates a fair baseline for everyone. The second layer belongs to the mathematics department. Because math is a cumulative subject where skills build on one another, the department must decide which specific skills in which courses must be practiced without any help. For example, in a basic algebra class, the department might decide that students must perform core calculations on their own to ensure they are ready for calculus later. In a statistics class, the rule might be that students must choose the model and interpret the results, even if they are allowed to use the tool to fix coding errors. This layer ensures that the learning goals of one course do not get undermined by the rules of another. The third layer is the classroom itself, where instructors translate these departmental standards into specific instructions for each assignment, telling students exactly how and when they can use the tool.
To make this work in practice, the researchers suggest changing the nature of the work students do. Instead of asking for a final answer, which the machine can generate instantly, teachers should ask for the thinking process. One effective method is to have students solve a problem on their own first, then use the tool to check their work, and finally write down what they learned from the comparison. Another approach is to have students act as critics, finding errors in a solution generated by the machine or explaining why a machine's answer might be misleading. In a proof-based course, a student might be asked to construct a logical argument from scratch and then use the tool to try to find a flaw in it. These tasks shift the focus from getting the right answer to demonstrating the ability to control, evaluate, and justify the reasoning behind the answer. The researchers emphasize that the goal is not to ban the technology, but to design tasks where the human mind must do the heavy lifting that the machine cannot replicate.
The paper also addresses how to grade this new kind of work. Since the correct answer is no longer the only proof of understanding, grading criteria must change. Teachers should look for evidence of decision-making, where the student explains why they chose a certain path. They should look for justification, where the student supports their choices with logic rather than just repeating what the machine said. They should also look for error detection, where the student spots mistakes in the machine's output, and reasoning transparency, where the student clearly explains their own thought process. By focusing on these elements, educators can assess whether a student truly understands the material, even if they used a tool to help. The researchers note that these strategies are most effective when practiced in low-stakes assignments before being used in major exams, giving students time to develop the habit of critical engagement.
Despite the detailed plan, the authors are careful to state that this framework is a proposal, not a proven solution. It is a conceptual model built by combining existing theories about teaching and technology with the specific needs of mathematics. It has not yet been tested in classrooms to see if it actually improves student learning or if it is practical for teachers to implement. The researchers acknowledge that their ideas were developed primarily within the context of the United States higher education system and may need adjustment for other countries or different types of institutions. They also point out that the boundaries between their three main ideas—ethics, utility, and agency—might overlap in ways that are not yet clear. Future studies will need to test whether these ideas work together as intended and whether they can be adapted to different student populations and course structures.
The ultimate goal of this work is to help universities navigate a future where artificial intelligence is a constant presence. The researchers suggest that the solution lies not in fighting the technology, but in redesigning education to ensure that students remain the primary thinkers. By clearly defining what must be done independently and what can be supported by a machine, and by designing assessments that make human reasoning visible, educators can protect the integrity of mathematical learning. The paper offers a starting point for departments and instructors to have these difficult conversations and to build a system where technology serves the student's intellectual growth rather than replacing it. While the framework is not yet a finished product, it provides a coherent structure for moving forward, ensuring that the next generation of mathematicians and scientists develops the critical thinking skills they will need to thrive in a world full of intelligent machines.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.