y’all I’m a PhD candidate and a TA and I genuinely don’t know where the line is anymore. and if I don’t know, I’m pretty sure most undergrads don’t either.
my program has a policy. it says ‘use of AI for submitted work is prohibited without explicit instructor approval.’ that’s the whole thing. no examples. no clarification about what counts as AI use. using ChatGPT to outline a section: prohibited? using Grammarly: probably fine but also maybe not? using a tool to help me understand a dense source before I engage with it: I genuinely don’t know.
I’ve asked around. other grad students are in the same position. we’re all guessing. some of us have quietly asked our advisors and gotten responses that ranged from ‘just don’t use it’ to ‘everyone uses it, just don’t get caught’ which are both unhelpful in different ways.
on the TA side it’s worse. I’m supposed to apply a policy I can’t define to students I’m supposed to treat fairly. not a great position.
what I actually want is some principles I can apply consistently rather than a vague rule I have to interpret every time. things like: thinking and understanding are yours, output that represents thinking should be yours too. using AI to learn something is different from using AI to produce something for evaluation. disclosure wherever you’re uncertain.
does anyone have a clearer framework that actually works in practice?
from a student perspective the inconsistency across teachers and classes is the main problem. one teacher says never, one says fine for drafts, one says use it but disclose it. so you’re making separate judgment calls for every class with different rules. that’s not really a framework, that’s just surviving
The principles you outlined at the end of your post are actually more coherent than most institutional policies I’ve seen. The real issue is that principles require judgment and institutions often can’t enforce judgment. They need bright lines. The mismatch between what would actually work and what can be administered at scale is where most of these conversations stall.
The distinction you’re drawing between learning use and output use is the most defensible framework I’ve seen in practice, and it maps onto what educational research says about where AI tools support versus undermine learning. The challenge is that most institutional policies haven’t caught up to that distinction, so applying it requires judgment that the policy doesn’t authorize.
The ‘everyone uses it, just don’t get caught’ response from an advisor is a policy failure, not advice. What it tells students is that the institution knows the policy can’t be enforced and has stopped trying. That’s worse than having no policy because it’s a policy that teaches people the rules are for others.
The framework I’ve been developing in my own research is grounded in what the work is supposed to demonstrate. If an assignment is meant to demonstrate your reasoning, your analysis, your argument, then AI tools that generate reasoning, analysis, or argument on your behalf undermine the point. AI tools that help you understand the material so you can reason, analyze, and argue more effectively do not. The question is always: what is this assignment for?