Gains:
- Ability to understand and apply the ethical principles of using artificial intelligence in mathematics (being in the service of learning, transparency, human responsibility, justice, confidentiality).
- Ability to assess privacy risks and take appropriate precautions in sensitive content such as exam questions, student data and unpublished research
- Ability to combine all parts of the module (production + verification) into a single end-to-end workflow, completing each production step with a verification step and final human approval
In this final unit we combine two things: on the one hand, the ethical, academic integrity and confidentiality dimensions of the use of artificial intelligence in mathematics; On the other hand, to bring together what we learned throughout the module in a single end-to-end workflow. AI in mathematics is a powerful tool, but it comes with responsibility as well as power: who you use it for, what you use it for, and how you use it has ethical consequences. A student having his/her homework done by AI and a teacher speeding up course material with AI are very different things. This unit complements technical skill with responsible use.
A few definitions. Academic integrity means honestly showing who really owns a work and not shortcutting learning. Plagiarism is presenting someone else's (or a medium's) product as if it were your own original product. Transparency means clearly stating where AI is used. These concepts have become central to mathematics education in the age of AI.
Basic principles of ethical use
- Not in place of learning, but in service of it. AI can support a student in learning the concept (explanation, alternative explanation), but should not make him/her think instead. Having AI do the homework undermines learning.
- Transparency. Indicate where AI was used — in an article, course material, or report. Concealing is dishonest.
- Responsibility belongs to humans. Anyone who publishes, teaches, or delivers AI output is responsible for its accuracy. “The AI said so” is not an excuse.
- Justice. The use of AI in assessment should be governed by equal and clear rules for all students.
- Security. Do not enter sensitive information, such as exam questions, student data, or unpublished research, into AI tools whose confidentiality is not guaranteed.
Privacy: what does it mean in the context of mathematics?
Even though mathematics may seem like it has "no hidden data", context is important. An unpublished research result, an institution's exam question bank, students' personal performance data, or a company's proprietary algorithm—all are sensitive. Everything you input into an AI tool can be processed by that tool's provider and, in some cases, used in training the model. So for sensitive content, either use locally/institutionally secured tools or anonymize the content.
Data type
Risk
precaution
Published theorem/problem
low
freely available
Active exam questions
high
Do not enter or secure vehicle
Student personal/performance data
high
Anonymise/do not enter
Unpublished research
high
Secure/local tool
Lecture outline
low-medium
Generally suitable
Caution: Before going into AI to verify an exam question, consider whether that question should remain secret. Questions from an active exam could theoretically be exposed when entered into an AI tool whose confidentiality is not guaranteed. For critical exams, either change questions or use only corporate-approved, data privacy-assured tools.
End-to-end workflow: an example task
Let's say you need to prepare both course material and an exam for a subject within a week. The flow combining all parts of the module could be like this:
1. Planning. Clarify topic, level and objectives (Unit 7).
2. Drafting the material. Have AI produce level-appropriate lectures + solved examples (Unit 7).
3. Sample verification. Check each solved example with SymPy (Unit 4).
4. Visualization. Produce and render graphics to support the narrative with Matplotlib code and compare them with mathematics (Unit 6).
5. Notation. Clean index, render and verify formulas with LaTeX (Unit 10).
6. Question generation. Have multiple choice questions + distractors produced for the exam (Unit 8).
7. Bulk verification. Solve all the questions in bulk with SymPy and check the answer key (Unit 8, 9).
8. Ethics and transparency control. Is confidentiality appropriate? Have you stated the use of AI in the material when necessary? You have the ultimate responsibility (this unit).
9. Human approval. Review and confirm everything one last time. Any unverified piece will not go on air/test.
Tip: In an end-to-end workflow, the most efficient strategy is to position AI as the “speed layer,” deterministic tools (SymPy, render) as the “correctness layer,” and yourself as the “judgment and accountability layer.” Every production step should be followed by a verification step; Unverified output should not be carried forward to the next step.
three mini cases
Case 1 — Transparency built trust. One academic used AI in the first draft of a textbook chapter, then verified every mathematical claim and stated in the preface that "AI support was obtained in the production of the manuscript, all content was verified by the author." This transparency increased the trust of colleagues and maintained the standard of integrity.
Case 2 — Use that undermines learning. A student had the AI solve his entire homework set and submit it. He failed the exam in the same subjects because he had never learned the concepts. The teacher suggested using the AI to “get the solution checked and get an alternative explanation,” but to solve the problems himself first. AI should be in the service of learning, not in its place.
Case 3 — Risk of privacy breach. One teacher considered hacking into a general AI tool to verify questions for a central exam that had not yet been administered, but then decided against it; instead he solved the questions in SymPy by himself and a colleague. Entering active exam questions into an external tool was a serious privacy risk; local authentication provided the secure path.
Four copyable templates
1) Use that supports learning (for students):
I solved this problem FIRST, my solution is attached. DO NOT give me the solution; just show me where I went wrong and what concept I misunderstood so I can correct it. My solution: [here]
2) Generating a transparency note:
I prepared a [course material/report] and used AI in the draft stage, verifying all the math myself. Write a short, professional transparency note that states this honestly.
3) Privacy pre-check:
I would like to privacy-qualify the following content before entering it into an AI tool: [content type]. Is this type of content considered sensitive, what risks are there, what precautions (anonymization, local tool) should I take? Say it short and clear.
4) End-to-end verification control plan:
I prepare course material + exam for [subject]. List as a control plan what verification I need to do at each step of production: sample verification, graphic control, notation, answer key, confidentiality, human approval.
Weak prompt / Strong prompt
Weak: "Solve these homework questions and I'll turn them in." (student)
Result: Learning is skipped, academic integrity is violated, student fails to learn the concept. It is ethically and pedagogically objectionable.
Güçlü: "I tried to solve this homework question, but I got stuck; my solution is attached. Do not give the answer directly; show me with a clue which step and why I did wrong, remind me of the correct concept, then I will try again."
Conclusion: AI is used in the service of learning; The student continues to think, honesty is maintained.
Common mistakes
- Using it to skip learning. Having AI do the homework is easy in the short term, but destroys learning in the long term.
- Not being transparent. Concealing the use of AI is against academic integrity; specify when necessary.
- Putting the responsibility on AI. “The AI said so” is not an excuse; The publisher/teacher is responsible.
- Entering confidential content into external tool. Active exam, student data, unpublished research are sensitive.
- Leaving validation to the end of the workflow. Verification should come after each production step; Errors should not accumulate.
In summary
The use of AI in mathematics requires ethical responsibility as well as technical skill: it should be in the service of learning (not in place of it), be transparent, retain ultimate responsibility, adhere to fair rules, and respect confidentiality. In an end-to-end workflow, AI is the speed layer, SymPy and rendering tools are the accuracy layer, and humans are the judgment and responsibility layer. Each production step should be followed by a verification step; No unverified output should go into publication, lecture, or exam. That's the essence of this module: get faster with AI, but verify and take responsibility for each step.
Application task
Design a small-scale end-to-end task: a short narrative + 1 chart + 3 multiple choice questions for a topic. Follow the flow of the module: (1) plan, (2) draft with AI, (3) validate examples with SymPy, (4) render and check the graph, (5) clean and render the notation with LaTeX, (6) batch validate the questions with SymPy, (7) do a privacy and transparency check, (8) give final human approval. Record what verification you did and what you found at each step. At the end of the process, evaluate where AI saves time and where your judgment is indispensable.
checklist
- [ ] I used AI in the service of (not instead of) learning.
- [ ] I have transparently stated the use of AI where necessary.
- [ ] I have assumed responsibility for ultimate accuracy.
- [ ] I did not access confidential/sensitive content through unsecured tools.
- [ ] I followed each production step with a verification step.
- [ ] I did not carry any unverified output into the broadcast/lesson/exam.
Module Exam
1. Which of the following is the most accurate positioning for artificial intelligence in mathematics?
- A) Artificial intelligence is a strategy, blueprint and organizing tool; It is up to man to verify the exact calculation and result by deterministic means ✔
- B) Artificial intelligence is like a calculator and every numerical result it gives is precise
- C) Artificial intelligence is never wrong in the conclusions it confidently presents, so verification is unnecessary
- D) Artificial intelligence is only useful for writing text, it has nothing to do with any mathematical tasks
Description: A large language model is a tool that statistically generates text; It is not a calculator or proof checker. It produces correct mathematics most of the time, but it does not guarantee; so every output should be verified with deterministic tools (SymPy, calculator, manual validation) and the responsibility should remain with the human.
2. A student asks artificial intelligence to solve an integral and wants to get the result. What is the most reliable way to verify the result?
- A) Differentiating the result and checking whether it returns to the original function (SymPy or manually) ✔
- B) Ask artificial intelligence 'is this true?' to ask
- C) Accepting the result as it is because it is presented fluently and confidently.
- D) Moving on to another question and not checking the result at all
Explanation: The derivative-integral inverse is a built-in verification tool in mathematics: taking the derivative of the result found and checking whether it returns to the original function independently confirms the correctness of the integral. The same cross-checking can be done with SymPy.
3. What does the principle of 'the right result is not the right way' mean in problem solving with artificial intelligence?
- A) The right result always proves that the path is also right
- B) Path is unimportant, only the result should be scored
- C) Incorrect intermediate steps may lead to the correct result by chance; Therefore, both the result and each intermediate step must be inspected ✔
- D) It is impossible to find a solid path, so we only look at the outcome
Explanation: Sometimes wrong intermediate steps lead to the right result by chance; Even if the result is correct, the path may be invalid. That's why in mathematics it's just 'is the result correct?' but rather 'is each intermediate step correct?' The question is also essential and every step should be supervised.
4. What should you do when AI uses the phrase 'without loss of generality' in a proof?
- A) Accepting that the proof is valid because the statement is fluent
- B) Justifying for yourself that generality is not actually broken; ✔ not accepting without question
- C) Considering the proof valid without reading it at all
- D) Searching for this expression from the source as a theorem name
Clarification: The statement 'without loss of generality' is sometimes valid (if there is genuine symmetry) but sometimes it is a hidden error. Artificial intelligence uses this expression a lot; You must justify to yourself each time that the generality is not really violated, and you must not accept the statement without question.
5. You wrote a SymPy code to the artificial intelligence. How do you get the result correctly?
- A) Making artificial intelligence predict the output of the code
- B) Assuming the code is correct without reading it
- C) Trusting the AI's interpretation of the output
- D) Run the code yourself in a real Python environment and see the exact output ✔
Description: The power of SymPy is only revealed when the code is run in a real Python environment. Ask artificial intelligence 'what does this code produce?' asking is not verification, because AI can also make up the code output. You must run the code yourself.
6. If the solution of an equation found manually by artificial intelligence and the result given by SymPy are different, which one should you trust and why?
- A) SymPy result because it is deterministic and precise; Then have the artificial intelligence find the location of its mistake ✔
- B) The result of artificial intelligence because it explains more fluently
- C) Choosing one at random because they are both equally reliable
- D) Rejecting both and considering the question unsolvable
Description: SymPy is a deterministic symbolic computing system: it always gives the same exact correct result to the same input. Artificial intelligence produces text statistically and can hallucinate. In the dispute, SymPy should be taken as the basis and the artificial intelligence should be made to find the location of its mistake.
7. Why does the condition '0.1 + 0.2 == 0.3' not work as expected in a Python code and how to fix it?
- A) Addition is broken in Python; need to use another language
- B) Exact equality does not hold due to floating-point number precision; A tolerance such as 'abs(a-b) < 1e-9' should be used ✔
- C) The value 0.3 is incorrect; It is necessary to write 0.30000000004
- D) The condition always works correctly, there is no problem
Explanation: Computers store decimal numbers in floating-point format with limited precision; 0.1 + 0.2 is stored with a very small deviation, not exactly 0.3. Therefore, instead of the exact equality '==', a tolerance such as 'abs(a-b) < 1e-9' should be used.
8. What is the best 'common sense test' before applying a scalar root finding code to the real problem?
- A) Applying the code directly to the actual problem without testing it at all
- B) Check the length of the code and count it correctly if it is long.
- C) Testing the code with a simple input whose answer is known (e.g. x²-4=0 roots ±2) and checking that it gives the expected result ✔
- D) Confirming to artificial intelligence that the code is correct
Explanation: Testing the code with a simple input whose answer you already know (e.g. ±2 for x²-4=0 with obvious roots) is the cheapest and most effective verification. If the code gives the known situation correctly, the probability of it being reliable for problems you do not know increases.
9. In visualization, how do you check that a graph reflects the mathematics correctly?
- A) If the graph looks aesthetically pleasing, it is considered correct.
- B) Comparing the graph with known properties of the function and the true value of at least one point ✔
- C) Not looking at the axis labels at all
- D) Relying on the AI's verbal description of the graph
Explanation: A graphic that looks nice doesn't have to be accurate; AI may draw the wrong function, in the wrong range, or with a distorted asymptote. Comparing the graph with known properties of the function (oscillation range, symmetry, asymptote) and at least one point calculation catches errors.
10. What is the most critical verification point in a course material produced with artificial intelligence and why?
- A) Font and color scheme of the material
- B) Length of the material and number of pages
- C) Accuracy of solved examples; because students take them as models and teach the wrong example and wrong method ✔
- D) How quickly the material is produced
Explanation: Solved examples are taken as models by students; An incorrectly solved example directly teaches students the wrong method. Therefore, each resolved example must be verified individually, step by step, either with SymPy or manually. A small ambiguity in the explanation of the subject can be compensated, but the wrong example causes cascading damage.
11. What is the most efficient and reliable way to verify a multiple choice question set generated by artificial intelligence?
- A) Asking the artificial intelligence once again if the questions are correct
- B) Checking only a few questions at random and trusting the rest
- C) Using the answer key as it is without checking any of it
- D) Checking the answer key collectively and deterministically by printing and running a code that solves all the questions with SymPy ✔
Explanation: Artificial intelligence most frequently makes mistakes such as marking the answer key incorrectly and producing multiple/zero correct options. Writing and running a code that solves all the questions at once with SymPy is the most efficient way to check the answer key deterministically and in bulk.
12. Which of the following is the main reason why artificial intelligence makes mistakes in mathematics?
- A) Artificial intelligence is a language model; does not check logical validity, generates text statistically ✔
- B) Artificial intelligence has never learned mathematics
- C) Artificial intelligence is deliberately programmed to give wrong answers
- D) Artificial intelligence makes mistakes only in very difficult problems, never in easy ones.
Description: Artificial intelligence is not a logic engine, but a language model that produces text with statistical patterns; It has no internal mechanism to check the logical validity of the steps. 'A valid proof' with '3-digit multiplication' is for him the task of producing the same type of text, so its correctness cannot be guaranteed.
13. When the result of a probability problem is 1.4, how do you know there is an error without going into detail?
- A) The calculation is correct because the result is 1.4, the probability may exceed 1
- B) The sign of the probability is unimportant, the value is accepted regardless
- C) It is not understood that there is an error, the result is considered final.
- D) Probability must be between 0 and 1; ✔ Common sense testing indicates an error because 1.4 exceeds this limit
Description: The common sense and order of magnitude test checks within seconds whether the result is within a reasonable range. A probability must by definition be between 0 and 1; 1.4 is outside this limit, so it is obvious that there is an error without looking at the details of the calculation.
14. Why is it important to write 'e^{x^2}' instead of 'e^x^2' in LaTeX?
- A) They are exactly the same, the difference is only aesthetic
- B) Curly brackets determine the scope of the exponent; Without it the expression would be incorrect or ambiguous, so the multi-character exponent {} is essential ✔
- C) Curly brackets are useless in LaTeX
- D) 'e^x^2' is always true, using '{}' is unnecessary
Explanation: The most common error in LaTeX is the scope error: if the curly brackets '{}' are not used in a multi-character exponent, subscript or fraction, only the first character is covered and the expression becomes incorrect (or even ambiguous). 'e^{x^2}' makes it clear that the entire exponent is x²; The code should be rendered and verified by visual inspection.
15. You want to have the questions of an active central exam verified by artificial intelligence. What is the best approach in terms of ethics and privacy?
- A) Entering questions freely into any general AI tool because mathematics does not contain confidential data
- B) Using questions in the exam without verifying them
- C) Validating questions with a secure/native tool (e.g. SymPy) or using only privacy-assured enterprise tools ✔
- D) Sharing questions on social media and asking others for control
Description: Active exam questions are sensitive; It could theoretically be exposed when an artificial intelligence tool whose confidentiality is not guaranteed is hacked. The correct approach is to either validate questions with a secure/native tool (e.g. with a colleague in SymPy) or to use only corporate-certified, data privacy-assured tools.