AI for Learning

Productive struggle is the point

The purpose of coursework is for you to learn, and that requires active engagement and, often, “productive struggle.” AI can be an effective supplement to deepen or expand your work, but any time you replace your own effort with AI's, you are missing a chance to get better at a task, fully internalize content, or improve your critical thinking. Be intentional about when and how you use assistance, and never hand over the keys to your education.

Learning happens when you do the work. The effort of retrieving something from memory, of holding a hard problem in your head until it yields, of writing a sentence three times before it says what you mean: that effort is not a tax on learning. It is the mechanism of learning. Work that feels smooth often teaches you less than work that feels hard.

AI removes friction. That is what makes it useful, and it is exactly where the risk lives. A tool that produces a clean paragraph, a working function, or a tidy summary in three seconds has removed the part of the task where the learning was going to happen. You still get the artifact. You do not get the capability.

So the question to ask before each use is not “can AI do this?” but “if AI does this, what do I lose the chance to practice?” Sometimes the honest answer is “nothing I need”: formatting a bibliography, converting units, tidying a data file. Sometimes the answer is “the entire skill this assignment exists to build.”

Course rules come first

Some instructors permit AI for brainstorming, feedback, or debugging; others prohibit it. Under Penn's Code of Academic Integrity, using AI without your instructor's permission can constitute unauthorized assistance or plagiarism. Before using AI for any coursework, check the syllabus or ask your instructor, and follow their instructions regarding usage and citation.

Consider intellectual property

Course materials—readings, slides, lecture notes, assignments—are not understood as public documents, and their creators may not have given permission for them to be shared, including uploading into LLMs. Do not upload anything without permission, and always prefer Penn-licensed tools, particularly PennChat, which is walled off for the Penn community, does not train any models, and is free for you to access.

AI as a thought partner

The most defensible way to use these tools in coursework is as an interlocutor rather than a producer: something you argue with, not something you delegate to.

A thought partner takes a position you have already formed and pushes on it. It asks what you meant. It names the counterargument you were avoiding. It tells you which step of your reasoning it could not follow. In every one of those exchanges the thinking stays yours; the AI supplies pressure, not content.

This is also, conveniently, where the tools are strongest. Models are unreliable narrators of fact but capable readers of structure: better at telling you that your third paragraph does not follow from your second than at telling you what your third paragraph should say.

A rough test: if you deleted the AI's output entirely, would your understanding still be better than before you started? If yes, it was a thought partner. If the output is the work, it was not.

Integrity is about the work, not the tool

Submitting work with inaccurate or fabricated citations, unattributed quotations, or unmarked paraphrase is a violation of academic integrity in itself, independent of whether you invented them or an AI did. “The model generated it” is not a defense, because the obligation was never about where the text came from. It is about you vouching for what you submit.

This matters more with AI than it used to, because these tools fabricate confidently. A citation with a plausible author, a plausible journal, a plausible year, and a DOI that resolves to nothing is a common failure mode, not an exotic one. Every reference you submit is a claim you are personally making about the scholarly record.

Verify every citation against the actual source. Read what you cite. Quote exactly and mark it. Paraphrase in your own structure and attribute it. These were the rules before; they have not changed.

You will have to talk about your work

Faculty are increasingly adding oral components: discussion sections that count, in-class defenses of a submitted argument, one-on-one conversations about your paper. This is a direct response to AI, and it is spreading.

AI-produced writing does not prepare you for any of it. You cannot defend an argument you did not construct, and the gap shows immediately: the essay is fluent and the conversation about it is not.

The same gap follows you out of the building. Job interviews are, almost entirely, exercises in talking about work you claim to have done, and so is much of the first year of a job. Being able to explain your own reasoning under questioning, in real time, to someone who knows the subject is not a bonus skill layered on top of the coursework. It is a large part of what the coursework was building.

The practical stakes

It is tempting to read all of this as a rule you might get caught breaking. It isn't. The reason to stay in control of your own learning is that the alternative leaves you with a transcript and not much else.

Coursework builds particular skills: reading something difficult and knowing what it says, constructing an argument that survives objection, debugging a problem nobody has written about, sitting with a question long enough to answer it. They remain valuable precisely because AI can produce a passable imitation of their outputs. When the output is cheap, the scarce thing is the judgment to tell whether it is any good, and that judgment comes from having done the work yourself.

Use AI to go further than you could alone. Don't use it to avoid becoming someone who could go there.