AI Agents in Graduate Studies: Ethical Considerations

The emergence of GPT agent has me thinking about its use in graduate study. I have no answers, glib or otherwise, for what an ethical use of these technologies will be in academia, but I would argue that at least such considerations must recognize that we are living in a world of AI-generated junk but also super-charged AI-research productivity. From a university perspective it could feel like the tail is wagging the dog, but if STEM is to remain of value it cannot fall behind corporate powered research, which has far different ethical considerations.

This is mainly what it means to be a university in a neoliberal cultural and economy. The university has to compete in the market. Sure, as a state entity, UB plays a different role from a corporation or private university but we are still market role-players. So I don’t think it will be possible for academia to go in a substantially different direction from the rest of society in relation to AI use. That said, there are many decisions and negotiations to be made.

For example, how will our graduate students use AI agents to aid their ability to gather and analyze existing research? What are the best methods? How does this change our expectation of what our students should be able to do? How does it change our expectations of what we can do, of what our colleagues can do?

Is it two books for tenure now?

And doesn’t this just increase the “reading problem” of information glut? Do we just stop teaching things that are more than 5-10 years old? (just kidding) I mean who has the time or interest to read all this surplus knowledge?

Oh. right.

All that said, while the academic market is already leaning toward creating hyper-productive AI-enabled scholars. But we don’t need to respond by writing more (or at least I don’t). Putting that aside, agent mode is an interesting tool for learning, at least in my experience. I haven’t tried what is proposed below, but I think I might. It is describing a potentially more dialogic process than the type fire and forget use of AI.

Perfect — here’s a playbook you could follow for a safe and effective “agent mode + university library” workflow. I’ll break it into stages:

🔹 1. Roles & Boundaries

Your role (Professor/Researcher): Authorized user of your university library’s subscriptions. You are the one who retrieves full-text articles legally using your credentials.
My role (Agent Assistant): Research aide who discovers what’s available, flags paywalled material, and works with any full-texts you provide.
Boundary: I never use or store your credentials. I don’t attempt automated access to licensed systems.

🔹 2. Workflow When I Hit a Paywall

Discovery Phase
- I try to locate the article via open sources (publisher site, PubMed, preprint archives, arXiv, institutional repositories).
- If I detect it’s behind a paywall, I respond with:“This article looks restricted. If you have institutional access, you can fetch the PDF and upload it here, and I’ll analyze it for you.”
Intervention Phase (Your Action)
- You log in through your university’s proxy/VPN or library portal.
- Download the PDF or export the citation.
- Upload it directly to me (or store it in a synced folder I can search, if you set up one).
Assistance Phase (My Action)
- Once I have the file, I can:
  - Summarize it (abstract, key findings, methodology, limitations).
  - Extract references, citations, or data.
  - Compare it with other papers.
  - Build bibliographies, syntheses, or teaching notes.

🔹 3. Optional Enhancements

Reference Manager Integration:
Store all your downloads in Zotero (or Mendeley/EndNote). Export PDFs into a connected Drive/Dropbox folder. I can then search and cross-reference them for you.
Tagging System:
You could name or tag PDFs (e.g., “AI_ethics_2023.pdf”). When you ask me “summarize AI ethics papers since 2022,” I’ll know which files to scan.
Pre-Check Routine:
If I find a citation, I can first check:
- arXiv
- PubMed Central (PMC)
- institutional repositories (open access copies)
  Before ever prompting you. This minimizes how often you need to intervene.

🔹 4. Safe Practices

Always download via your university’s licensed portal — not through me.
Upload only what you’re allowed to use personally for research/teaching (no mass redistribution).
Treat me as your research analyst, not your access portal.

✅ This way, you stay compliant with license agreements, but you still get full use of me as your assistant once you’ve legally fetched the content.

Would you like me to draft sample “agent prompts” (little canned messages I could use automatically whenever I hit a paywall) so you don’t have to invent the workflow language each time? That way you could just plug them into your agent mode.

Alex Reid, professor of digital rhetoric, Media & Artificial Intelligence