Contents
AI can help you learn mathematics when you use it to deepen your own understanding. Judge its value by what you can explain and solve on your own afterward, not by how quickly it produces answers.
Key Points#
- Use AI when it helps you understand mathematical concepts more precisely.
- Work through examples yourself. Reading a solution does not necessarily mean you can reproduce its reasoning independently.
- Check your understanding by explaining every step to someone else without looking at your notes.
- Treat AI-generated proofs and examples as material to verify, even when they seem convincing.
Set Your Learning Goal#
The November 2026 essay presented here sets a clear goal for graduate mathematics education: to develop people who understand mathematics deeply and can conduct research. Producing more true statements is a separate goal.
This distinction matters even when results come with formal verification by systems such as Lean. Verification can confirm that a formalized statement follows from the specified definitions and assumptions. But that alone does not explain why the result matters, how it might be discovered, or what intuition students should develop.
Before using AI to learn, ask, “What will this help me understand?” Clarifying a missing step in a proof is a useful goal. Getting a complete solution so you can stop thinking works against that goal.
Try It Yourself First, Then Ask Focused Questions#
Working independently gives you something to compare with AI’s response. Without that preparation, a fluent explanation may feel easy to follow while leaving the real difficulty untouched. Recognizing the notation does not necessarily mean you understand why the argument works.
When you get stuck on a textbook proof, try these four steps.
1. Write down precisely the claim you cannot justify. 2. Review the definitions and assumptions available at that point. 3. Try a simple example or a direct derivation. 4. Show your attempt and ask AI about what remains unclear.
Ask for only the help you need, with questions such as “Which assumption justifies this implication?” or “Give me a hint without completing the proof.” Focused questions make it easier to assess the answer and continue reasoning yourself. If seeing an answer makes you lose the motivation to think, postpone using AI until you have finished your own attempt.
Check Examples Yourself#
AI can suggest examples and counterexamples that would take time to construct yourself. The essay’s author reports that this assistance helped them explore more mathematical patterns and resolve questions they had previously left open. This is a personal account, not evidence that all students will benefit in the same way.
The practical lesson is to use generated examples as exercises. Check whether the object satisfies the stated definition. Calculate small cases yourself. For a proposed counterexample, verify both that it satisfies the assumptions and that the conclusion fails.
You can also ask AI what happens when you remove one assumption. Investigate its answer on paper. This turns the response into a concrete mathematical task and helps reveal the role of each assumption.
Check Your Understanding Without Notes#
The essay proposes a demanding standard: explain the material to someone else, line by line, without looking anything up. A seminar is a good setting for this because listeners can ask questions at the exact point where your explanation becomes vague.
After studying with AI, close the conversation and reconstruct the argument on a blank sheet of paper. State the definitions, identify where each assumption is used, and justify each step. If you cannot reproduce a step, record it as a gap in your understanding rather than filling it with words such as “This is obvious.”
Reconstruction checks whether you have internalized the reasoning; questions check whether you can use it. Ask your listener to change an assumption or request an example. If your explanation stalls, return to that point and work through it again.
Verify Answers and Keep Your Learning Grounded#
AI-generated mathematics needs verification. A response may apply a theorem without satisfying its assumptions, change the meaning of a symbol, or offer an example that fails to meet the stated conditions. Clear writing does not resolve these problems.
For every response you intend to use, check four things: definitions, assumptions, every inference, and cited results. Where available, formal verification provides a different kind of assurance. You still need to check whether the formalized statement corresponds to the mathematical question you intended to ask.
Make a well-chosen textbook or course the foundation of your learning. Its choice of topics, sequence of arguments, and examples can teach ways of thinking that isolated answers cannot convey. Use AI to investigate questions within that framework, and agree with your advisor on how much you may use it in submitted work.
Next Step#
Choose one proof you recently studied with AI, close all reference materials, and explain it aloud from scratch. Mark the first step you cannot justify as your next learning task.
Source: https://www.ms.u-tokyo.ac.jp/~konno/grad_ai.html
Contents
