Are uses of GenAI in Education limitless?
Generative AI ("genAI") apps have begun to be used by staff and students of universities faster than most institutions are able to study, regulate, or even teach it well. GenAI like ChatGPT, Gemini, Claude, and Co-Pilot can draft an essay, explain a theory, translate a paragraph, invent (i.e., hallucinate) references, and imitate feedback now sits in the same digital space as lecture notes, library databases, and assessment portals.
Academic staff should neither ban it reflexively nor welcome it uncritically. The more defensible position is narrower:
Like calculators and the world wide web before them, students must realise that they may use AI for a small set of low-stakes learning tasks, but not as a substitute for scholarship, reading, interpretation, evidence, or original argument.
This distinction marks the difference between AI as a study aid and AI as a ghostwriter. It also recognises that some students can benefit from carefully bounded use, especially when they are learning terminology, testing their understanding, or improving the clarity of their writing, and that generative AI poses serious risks to universities, including the propagation of disinformation, weakened scholarly habits, privacy breaches, data sovereignty concerns, and the use of copyrighted work without meaningful consent.
The classroom question, then, is not “Should AI be allowed?” It is sharper than that: Which uses are educationally justified, and which uses damage the conditions that make education possible?
The case for limited encouragement
There is a weak version and a strong version of the case for AI's use in higher education.
The weak version has it that students will use it anyway, so lecturers may as well accept it. That argument is practical but thin. It treats academic policy as a surrender to convenience.
The stronger case is educational. If a tool can help students rehearse ideas, identify gaps in understanding, or prepare for human discussion, then limited use may support learning. The keyword here is limited. AI is more defensible when it helps a student prepare to think, and it is less defensible when someone believes that AI performs the thinking for them.
This matters because university learning is not only about producing a final answer. It involves slow and sometimes uncomfortable activities: reading difficult texts, noticing ambiguity, comparing evidence, accepting uncertainty, and revising one’s view. These habits are central to scholarship. They are also exactly the habits that automated text can weaken when used carelessly.
GenAI does nothing more than predict what the next word would be, given a particular question. GenAI does not have knowledge in the human sense. GenAI does not read with disciplinary judgement, understand a course context, or bear responsibility for a claim. AI responses often sound fluent before they are accurate. That makes it useful for some early-stage support and dangerous when treated as authority.
A student asking for a plain-language explanation of a difficult concept may receive something useful, but the student is not in the position to judge whether the genAI's output is useful or accurate. A student asking for a literature review on a specialised topic may receive confident nonsense, missing sources, or fabricated citations. The same tool can help or harm, depending on the task.
That is why academic staff should encourage only uses that meet three tests:
The use supports learning rather than replaces it.
The student can check the output against reliable sources.
The use is declared where assessment rules require it.
On that basis, genAI can have a place in the classroom. It cannot be given the place of the teacher, the library, the laboratory, the archive, or the student’s own judgement.
AI can help students learn the language of a discipline
Every discipline has its own vocabulary. First-year students often struggle not because they lack intelligence but because they are meeting a new language. Terms such as “epistemology”, “statistical significance”, “validity”, or “phenomenology” carry histories and debates. A textbook definition may be correct but hard to enter.
GenAI can help here if its output is treated as a first pass. A lecturer might encourage students to ask for a simple explanation of a concept, then compare that explanation with lecture notes, course readings, and scholarly sources.
For example, a philosophy student might ask a genAI tool to explain the difference between a valid argument and a sound argument. The tool will likely produce a basic distinction: validity concerns the structure of the argument, while soundness requires both validity and true premises. That is a helpful starting point, but that is not a substitute for working through examples, studying the assigned text, or learning why the distinction matters.
The uses of genAI are modest. That is their strength. They resemble asking a peer to explain a term after class. They may lower the barrier to entry without pretending to settle the matter.
The risk is that simple explanations can become simplistic explanations. GenAI tends to smooth over conflict, minority positions, and unresolved debates. In many disciplines, the disagreement is the point. A good lecturer can address this by requiring students to bring genAI explanation back into the classroom and ask four questions of the output:
What does this explanation leave out?
Which terms need more precision?
Does the assigned reading agree with it?
What would an expert in this field challenge?
Used this way, genAI becomes an object of critique, not an oracle.
GenAI can support low-stakes practice
Learning improves through practice before assessment. Students need safe opportunities to test understanding, make mistakes, and receive feedback. AI can provide some of this practice, especially when human feedback is scarce or delayed.
A lecturer might allow students to use AI to generate practice questions from their own notes. A student could paste a short paragraph they wrote themselves and ask for possible quiz questions. They could ask for flashcards on key terms, then check them against the course materials. They could request a practice debate prompt before a tutorial.
This use has limits. AI-generated questions can be badly framed, too easy, too broad, or simply wrong. The tool may focus on surface recognition rather than deeper analysis. For that reason, lecturers should not outsource assessment design to AI without expert review.
Still, student-led practice can be useful. It is similar to making one’s own revision cards, except the tool offers a quick draft that the student must test. The educational value lies not in the output alone, but in the checking process. Students learn when they ask, “Is this a good question?” and “What would count as a strong answer?”
This is especially helpful for students who do not yet know how to study effectively at university. Many arrive with habits formed around memorisation or last-minute writing. GenAI may be used to model better study questions, if lecturers show what a good question looks like.
For instance, compare these two prompts:
“Make me a quiz on this lecture.”
That prompt invites shallow output.
A stronger version would be:
“Using only the notes below, write five short-answer practice questions that test comparison, application, and evaluation. Do not answer them. After each question, say which part of my notes I should reread.”
This still needs checking, but it makes the task more educational. It directs attention back to the student’s own material. It also discourages the tool from becoming a substitute teacher.
GenAI can help with clarity without writing the argument
There is a legitimate use of genAI as a language support tool. This is especially relevant for students writing in English as an additional language, students with disabilities, and students who are still learning academic style.
A student may know what they want to argue but struggle to express it clearly. Asking genAI to identify unclear sentences, flag repetition, or suggest plainer wording can support communication. This resembles using a grammar checker, consulting a writing centre, or asking a classmate whether a paragraph makes sense.
The line must be drawn carefully. There is a difference between improving clarity and generating substance.
A limited and defensible request might be:
“Here is a paragraph I wrote. Identify sentences that are unclear. Do not add new ideas. Do not rewrite the paragraph. Explain what makes the sentences difficult to follow.”
A less defensible request would be:
“Rewrite this paragraph to make it more sophisticated and add stronger arguments.”
The first prompt supports learning. The second risks replacing the student’s prose and reasoning with machine-generated slop. It may also blur authorship. If the final submission contains arguments, examples, or phrasing that the student did not create or understand, academic integrity is at stake.
Universities already recognise that writing support exists on a spectrum. A tutor may explain structure. A proofreader may correct minor errors if policy allows it. A supervisor may ask hard questions about evidence. AI complicates this spectrum because it can generate whole paragraphs instantly.
That means policies should focus less on whether genAI was “used” and more on how much intellectual work it performed. If genAI corrects a typo, the issue is minor. If genAI supplies the thesis, the structure, the evidence, and the language, the submitted work no longer shows the student’s learning.
A useful rule for students is simple: if a lecturer asks, “Can you explain and defend every sentence in this work?”, the answer must be yes.
GenAI can be used to teach critical judgement
Perhaps the strongest classroom use of genAI is not asking it for help, but asking students to examine it critically.
Lecturers can bring genAI output into seminars and ask students to mark it up. Where is it vague? Where does it overstate? Where does it invent certainty? Which sources would be needed? Which disciplinary assumptions shape the answer? What kind of question produced this response?
This turns genAI into a teaching object. It helps students see why scholarly standards exist. A polished paragraph can still be weak. A confident answer can still lack evidence. A general explanation can still erase the complexity of a field.
This use is valuable in philosophy, education, law, health sciences, history, commerce, engineering, and the creative arts. In each case, students can compare AI output with expert sources and learn to recognise the gap between fluency and knowledge.
For example, in a philosophy class, students might ask an genAI system to explain utilitarianism. The answer will probably mention happiness, consequences, and the greatest good. A seminar can then ask what is missing: the distinction between act and rule utilitarianism, objections about justice, questions of measurement, or the difference between classical and contemporary accounts.
In a history class, students might compare a genAI-generated summary of the Treaty of Waitangi with primary documents and reputable scholarship. The point would not be to let the tool teach the topic. The point would be to expose the difference between a broad summary and historically grounded interpretation.
This approach also helps students understand why artificial intelligence, classroom use, pedagogy, philosophy, genAI, OpenAI and related debates cannot be reduced to a simple yes or no. The educational question is not whether a tool can produce text. It is whether students become more capable readers, thinkers, and writers after using it.
The risks are not side issues
A narrow defence of genAI in education must face the strongest objections. These objections are not technophobic. They arise from core academic values.
First, universities depend on trust, evidence, attribution, privacy, and intellectual independence. GenAI can put all of these under pressure.
AI produces errors with confidence
GenAI can fabricate references, misstate theories, confuse authors, and invent details. These errors are sometimes called hallucinations, but the plain point is enough: the system can be wrong while sounding right.
This is especially dangerous for students who lack the background knowledge needed to check the answer. A postgraduate student may notice that a citation looks suspicious. A first-year student may not. A fluent response can create false confidence.
In technical fields, the danger is sharper. A small error in code, law, medicine, engineering, or statistics can change the meaning of an answer. Academic staff have criticised genAI for giving overly general replies to specialised questions, and that criticism is well founded. The tools often flatten disciplinary complexity into plausible prose.
This does not mean they are useless. It means they must remain outside the chain of authority. The proper chain is still: evidence, method, disciplinary expertise, peer review, and reasoned argument.
GenAI can weaken academic integrity
Academic integrity is not only about cheating. It is about the relationship between work submitted and learning achieved.
If students use genAI to draft essays, solve problem sets, summarise readings they did not read, or generate discussion posts, the assessment no longer measures their understanding. This harms the student, the lecturer, and the qualification.
Detection tools cannot solve this problem on their own. They can misidentify human writing, miss genAI-assisted text, and create unfair anxiety. A stronger response is assessment design that asks students to show process as well as product.
That may include annotated bibliographies, oral checks, in-class writing, staged drafts, reflective notes on source use, and tasks tied closely to local course materials. None of these measures is perfect. Together, they make hidden substitution harder and genuine learning more visible.
GenAI may normalise shallow reading
One of the quiet dangers of genAI is that it makes not reading feel efficient. A student can ask for a summary of a chapter, then a summary of the summary, then a paragraph about the chapter, all without meeting the text itself.
This matters because scholarly reading is not just information extraction. It is an encounter with an author’s method, evidence, style, assumptions, and limits. Students learn by struggling with that encounter. They notice how claims are built. They learn what counts as proof in a field. They see how scholars disagree.
GenAI summaries can have a place after reading, as a comparison tool or revision aid. They should not replace the reading. If universities allow that replacement, they train students to consume conclusions without understanding the work that produced them.
Data sovereignty cannot be an afterthought
The privacy risks of genAI are often described too narrowly. The issue is not only whether a student enters a name, an email address, or an unpublished essay into a tool. The deeper question is where data goes, who controls it, how it may be stored, and whether it may be used to train future systems.
In Aotearoa New Zealand, this issue carries particular weight. Universities operate under the Privacy Act 2020, and many also recognise the importance of Māori data sovereignty. Māori data sovereignty is concerned with the rights and interests Māori have in relation to data that concerns Māori people, knowledge, resources, and communities. This is not a minor compliance matter. It is about authority, consent, and collective rights.
If a class activity asks students to upload interview transcripts, community knowledge, clinical notes, unpublished research, or culturally sensitive material into an external AI system, the lecturer may create risks that students cannot fully assess. Even when a platform promises privacy protections, academic staff should ask whether the use is necessary, proportionate, and approved by the institution.
The safest classroom rule is conservative: do not enter personal, confidential, unpublished, culturally sensitive, or community-owned material into public genAI tools. This includes student work unless there is clear consent and a clear institutional basis for doing so.
This rule applies to well-known systems and institutional versions alike. Tools such as Microsoft Co-Pilot, Google Gemini, ChatGPT, Anthropic Claude and similar services differ in settings, contracts, and data handling. Those differences matter, but they do not remove the lecturer’s responsibility to ask where the data goes and who benefits from it.
Data sovereignty also affects research practice. Scholars may work with participant interviews, archival material, Indigenous knowledge, commercially sensitive information, or early findings not yet published. Feeding such material into genAI systems can breach ethics approvals, funder rules, community agreements, or basic scholarly trust.
A university that treats genAI only as a study tool will miss this wider issue. GenAI policy belongs not only in teaching and learning committees, but also in research ethics, library services, information technology, Indigenous governance, and legal review.
Copyright and the problem of extraction
Generative AI systems are trained on vast collections of text, images, code, and other material. Much of the public debate has centred on whether copyrighted works were used without permission, payment, or proper acknowledgement. Writers, artists, publishers, software developers, and news organisations have raised legal and ethical objections. Some disputes have entered courts in major jurisdictions.
The legal position differs by country and remains unsettled in many areas. The ethical concern is already clear. Universities teach students to cite sources, respect authorship, and avoid plagiarism. It would be inconsistent to ignore the possibility that the tools themselves were built through large-scale use of other people’s work without meaningful consent.
This does not mean every classroom use of AI is automatically unethical. It does mean academic staff should avoid presenting these systems as neutral utilities. They are commercial and institutional products shaped by labour, data, copyright disputes, and unequal power.
The scholarly community has a special duty here. Researchers rely on the integrity of the knowledge commons, which includes books, articles, archives, datasets, creative works, and software. If universities benefit from systems that may draw value from that commons while weakening attribution, they should at least speak plainly about the trade-off.
Students should also learn the difference between citation and imitation. A genAI output rarely tells them which works shaped a particular sentence. It may produce a smooth answer that carries traces of many sources without naming them. This makes it a poor model for academic writing, where claims should be traceable.
A responsible AI policy should include copyright literacy. Students should understand that generated text is not automatically free of ethical concern simply because it is new in form. They should also know that AI-generated references may be false, incomplete, or misleading.
What lecturers should encourage
The question is not whether lecturers should mention genAI. They should. Silence leaves students to learn from advertising, peers, and guesswork. The better question is what lecturers should actively encourage. The answer should be narrow. Lecturers can reasonably encourage students to use AI for.
Students may ask for plain-language explanations of difficult concepts, then verify them against course materials. Students may generate practice questions, flashcards, or revision prompts from their own notes, provided they check the content. Students may ask for feedback on clarity, grammar, or structure without allowing the tool to add new ideas or arguments.
These uses have a common feature. They keep the student in charge. They also keep the tool away from the core assessed work of forming an argument, selecting evidence, interpreting sources, and producing original analysis.
Lecturers should discourage or prohibit students from using genAI to write assessed work or substantial parts of it, produce references or literature reviews without verification, summarise readings instead of reading them, analyse confidential, personal, unpublished, or culturally sensitive data, generate answers in closed-book or unauthorised assessment conditions, imitate the voice of another person or conceal authorship, and create content that the student cannot explain.
What universities should change
The rise of genAI should prompt universities to revisit long-standing assumptions about assessment. Many common assessment tasks were designed for a world where producing fluent prose required sustained effort. That assumption no longer holds.
This does not mean essays are obsolete. Essays remain valuable when they require close reading, independent judgement, and careful argument. But lecturers may need to make the process more visible.
Practical changes include:
requiring short research logs that show how sources were found and evaluated
asking students to submit annotated extracts from key readings
including oral defence or short viva-style checks for major assignments
using in-class writing for parts of an assessment
designing questions that refer to specific course discussions
asking students to reflect on how their argument changed during drafting
requiring accurate citation trails that can be checked
These measures do not punish students. They defend the value of their work. They also help lecturers see learning as it develops, rather than only the final polished submission.
Universities should also invest in human support. If students turn to AI because they cannot access timely feedback, writing help, disability support, or language support, then the institution should address those gaps. AI can assist at the margins, but it should not become a cheap replacement for teaching.
Staff need support too. Many lecturers face pressure to redesign assessments, police misconduct, learn new tools, and answer student questions without enough time or guidance. Institutional policy should be clear, practical, and consistent enough to reduce confusion. It should also respect academic freedom and disciplinary difference.
A chemistry lab report, a creative writing portfolio, a te reo Māori assignment, a clinical reflection, and a philosophy essay raise different issues. One policy cannot answer every case. It can, though, set principles.
The classroom culture matters more than the tool
Rules alone will not settle the issue. Students also need a culture in which academic work has visible value. If assessment is framed only as a hurdle, genAI will look like an efficient shortcut. If education is framed as the formation of judgement, genAI’s limits become easier to see. Lecturers can help by explaining why certain tasks matter. Close reading is not busywork. Citation is not clerical fussiness. Drafting is not inefficiency. These practices teach students how knowledge is made, tested, and shared.
Students also deserve honesty. Some academic tasks are dull. Some assessments are poorly designed. Some feedback arrives too late. AI did not create these problems, but it exposes them. Universities should respond by improving teaching, not by outsourcing the difficulty of learning to machines.
A mature classroom conversation about genAI might begin with these propositions:
genAI can produce useful outputs, but outputs are not the same as understanding.
genAI can support revision, but it cannot bear responsibility for a claim.
genAI can explain a concept, but it cannot replace disciplinary study.
genAI can assist access, but it can also deepen inequality if some students receive better guidance than others.
genAI can save time, but education often requires time spent carefully.
The last point is unfashionable but essential. Some slowness is not a defect in scholarship. It is the condition under which judgement forms.
A defensible policy of limited use
Academic staff should encourage students to use genAI only when the educational purpose is clear, the risks are controlled, and the student’s own work remains central. That position will disappoint both extremes. It will seem too permissive to those who want a ban. It will seem too cautious to those who see genAI as a general solution. Yet it fits the reality of university learning better than either response.
The limited-use position says yes to study support, practice, accessibility, and critique. It says no to hidden authorship, source avoidance, data extraction, and automated scholarship. It accepts that genAI exists, but refuses to let it define what education is for.
The most useful classroom question is not “Did genAI help?” but “What did the student learn, and can they show it?” If genAI helps a student understand a term before reading the assigned chapter, the use may be justified. If it helps them spot unclear wording in a paragraph they wrote, it may be justified. If it generates practice questions that lead them back to their notes, it may be justified. If it writes the argument, replaces the reading, invents the sources, or processes sensitive data, it should be rejected.
Universities should teach this distinction explicitly. They should not leave students to infer it from inconsistent rules or anxious warnings.
GenAI in education should be treated like a powerful but unreliable study aid. It can be invited into the classroom only under supervision, with clear limits, and with constant attention to the values it may erode. The task of the university is not to keep every new tool outside the gate. It is to protect the forms of thought, evidence, care, and responsibility that make scholarship worth pursuing.
Comments