No, it’s not “okay to use it to get started”

The “It’s okay to use it to get started” story is fast becoming a trope in higher ed discourse on the relationship of AI¹ commercial products like ChatGPT to research education. And it’s irritating the hell out of me.

The story goes that it’s fine to use AI products in research to brainstorm ideas, refine topics, identify keywords for searches, and summarize readings. Such “appropriate” uses are typically contrasted with “inappropriate” uses: don’t use it to do the actual lit searching, to evaluate the results, or to write the paper.² To cite the familiar framing, students and other researchers are advised–indeed, often cheerfully encouraged–touse AI as a thinking partner, not a replacement.

This story is repeated over and again in workshops, library guides, tutorials, policies, course outlines, and other documents in higher ed. I’ve been trying to get my head around why this bothers me so much.³

Part of what irritates me about “It’s okay to use it to get started” is that it feels like a bit of a capitulation to the narrative that AI is inevitable. They all use it, they all want to use it, they all happily use it, so the best we can do is meet them half way by distinguishing between “appropriate” and “inappropriate” use.

Another part of it is its frequent appearance as part of the larger reduction of AI literacy to academic integrity and the reduction of integrity to follow the rules: using genAI to do the “important” work of searching, evaluating sources, and writing papers is “cheating.”

Now, it’s often and quite rightly noted that genAI products can be wrong, so you shouldn’t trust them to do the substantive bits of your work, the bits of your work that are important. You should think about the consequences of inaccurate information.

And it’s also often and quite rightly noted that if you’re using genAI products to automate your searches, your source evaluation, and the production of your paper, then you’re missing the learning that is embedded in these processes.

Inaccuracy. Lost learning. These are certainly good bases upon which to trouble the relationship of AI products to education. But do they not also apply to the processes of getting started with a new topic, brainstorming ideas, refining your topic, identifying keywords, and summarizing readings?

The “It’s okay to use it to get started” story seemingly abandons the idea that there is crucial learning embedded in these practices.

But it’s precisely these kinds of practices that the students with whom I work tend to struggle with the most, which is why I often spend a good chunk of teaching time addressing them.

We talk about what context means and about what, in turn, it means to contextualize something.

We talk about the intellectual and emotional value of surfacing, honouring, and qualifying what you already know and/or what you already assume about a subject, recognizing that you quite frequently do know and/or assume things about that subject even if it seems overwhelmingly unfamiliar.

We talk about and practice asking basic who-what-when-where-why-how questions as means of facilitating this initial work of identifying existing knowledge.

We also use these basic questions as a means of specifically identifying what you don’t know. And we talk about the practical value of articulating basic gaps in knowledge with as much specificity as possible and, in turn, the value of having smaller questions to pursue as a means of getting started.

We also focus on the relationship between different sorts of questions. What happened during this event?, Where and where did this event take place?, and Who was directly involved? might lead you to questions like What other significant events were happening in the same time and the same place? and To what other individuals and groups were those involved connected?

We apply similar kinds of strategies to the challenge of refining an existing research question and that of articulating the scope of a project.

We also talk about and practice the process of brainstorming potential topics as a means of actively focusing on what might be conceivable rather than only what definitely is. And we take the time to explore the logic behind these conceivable research directions: Maybe ____ because ____

As we are doing so, we take the time to underscore the value of imagination in research–the utility of replacing the daunting What are primary sources related to EVENT? with the more manageable Maybe EVENT was covered in the local newspapers? and Maybe there were posters advertising EVENT?

We explore the practice of identifying keywords for searches through a similar lens, situating it as a process of attending to the ways in which a phenomenon is and is not being framed.

And as we are doing so, we talk about the practice of navigating the political, analytical, and ethical dimensions of language, about the tension between the words you use to search and the words you use to write.⁴

None of this is earth shattering. None. I imagine that these are incredibly common and conscious practices in the work of instruction librarians and other educators all over the place. Maybe I’m wrong, but I imagine we all bring these things into our teaching because we think they bring important learning with them.

So does the concern about learning lost through the use of genAI products not extend to the kinds of learning embedded in these kinds of practices of “getting started”?

Or do those who are rehearsing these “It’s okay to use it to get started” narratives genuinely believe that these practices don’t have substantive educational value?

If we do accept that, say, the practice of brainstorming potential research questions and the practice of writing a paper both have educational value, then what’s the difference? Why is it okay to replace the one but not okay to replace the other?

The main difference I see is that those practices for which genAI product use is typically deemed inappropriate tend to be those that are about outputs–about demonstrations of the final form of a process: the final searches should be yours, the final analysis should be yours, the final paper should be yours.

As many, many others have pointed out, the idea that learning is about output is a defining and ever-deepening feature of higher education under capitalism. Still, I suspect that there are very few educators who would consciously and explicitly endorse the idea that education is about output, who would use their chest voice to say, yes, what matters in education is the final thing you produce, not the process that led you there–it is about the destination, not about the journey. But that’s ultimately the message that’s coming across, regardless of its embeddedness in a narrative that seemingly expresses the opposite.

There are connections to be made here to the trope of workplace readiness circulated in AI discourse in higher ed.

There are also connections to be made to the ways in which such discourse is intensifying the erosion of teaching and learning as social, of knowledge creation as necessarily relational: Here’s a brand new group study space! Here are some fun ways to get involved and meet people on campus! Here’s a memo circulating a white paper on the benefits of group work! Here’s a new evidence-based wellness initiative to address the problem of isolation and the growing number of mental health crises we’re seeing among students on campus! Don’t be alone. Also, check out these new AI “study buddies”! We paid a lot for institutional licenses, so we’d encourage you to use them rather than finding human “thinking partners” and therefore needing to make connections with other people, which is hard, right?

And, at a deeper level, there are connections to be made to the larger pedagogies of socioecological devastation inherent to AI as a project. Through it all runs a thread of ruling class extraction of wealth through the reduction of living relations to problems of data, productivity, technical efficiency.

But that’s for another day.

The above is a slightly expanded version of text originally posted as a thread on Bluesky.

Footnotes

Here’s what I mean by AI.
The exact sorting of research practices into categories of appropriate and inappropriate differs from one instance of the story to another, variation that is indeed telling.
I mean, beyond the more general assertion that it’s fine to use genAI and the utter lack of criticality that tends to accompany such assertions.
The practice of summarizing a source falls outside the scope of what I teach, but I take it as a given that the practice integrates, among other things, exercises of critical reading and contextualization.