Teaching Thursday: Chasing AI

As readers of this blog know, I’m intrigued by how AI has become the latest raw material for the outrage factory (on social media). To be clear, I’m not an AI apologist nor some kind of fanboy. I have found some of its capacities useful — especially for analyzing very large bodies of unstructured text — but I am acutely aware of its limitations and the ethical problems associated with its use. 

That said, I’ve can also appreciate how AI has become a scapegoat for certain broader trends in social and in academia. There are lots of reasons to dislike and distrust the behavior of large tech companies recognizing the potential for abuse encourages us to remain vigilant. As goes society, so goes academia. The rise in the use of generative AI in scholarship and in the classroom has caused a good bit of alarm, led to a rash of new policies, and more than a little pious moralizing.

What has intrigued me the most lately is the close relationship between what AI can do in the classroom and in academia and the standards we often hold as instructors and researchers. This makes sense, of course, not only is AI trained on texts produced under certain standards, but we often reinforce those standards by providing rubrics and guidelines that serve as guardrails (or form the basis for Reinforcement Learning from Human Feedback [RLHF]) for AI responses. Recently, a colleague was decrying an AI application that promised to produce accurate citations for academic work. He noted that such an application would hinder students ability to do proper research; in fact, he offers a “pro-tip”: “the only honest and acceptable way to generate citations is to do research, write from your research, and cite the sources you use.”) There is much to agree with in his perspective.

What struck me however is that using Ai to generate citations is a solution only because academia has fetishized citations as an indication of academic rigor. I have to admit that I sometimes fall into this trap. When students ask me how many citations they need in, say, a research proposal, I usually tell them that citations are a way to show me that you’ve done the work. In other words, citations authorize one to make statements on a given topic. While this is pedagogically useful — especially for students who struggle to understand the value of the research process — it feeds into a culture where citations are less about giving credit, showing the genealogy of ideas, or engaging in a conversation and more about “showing receipts.”

This culture is nowhere more manifest than in the peer review process where (and I speak here as someone who has done just this) it is only too common to be told to cite this or that work. Sometimes this is a welcome suggestion, but as often, it is a nudge to include a vaguely related article that seems “not inappropriate to cite.” A cynic can sometimes see the nudge as a political gesture to cite a colleague’s (or a reviewer’s) publication or a kind of desultory comment designed to show a vague attentiveness to the manuscript at hand. In few cases does the reviewer make the case for why a particular citation is valuable. As a result, the citation becomes performative in an intellectually empty way.

This, of course, coincides with the rising use of citations to measure scholarly productivity and “impact.” I10-index and H-index are perhaps the most common examples. Both of these serve mainly to track the number of times publications cite a particular work and this serves as a surrogate for “impact” in the field. The main promoters of these kinds of measurements are tech services company that overlap nicely with firms promoting and developing generative AI. Indeed, the kind of structured data present in citations is easy to extract and quantify. More to the point, for publishers having an oft-cited article or journal improves their rating, attracts subscriptions (and submissions), and pads their bottom line. It creates a metric that is easy for administrators to understand as well especially in an era where vaguely benevolent efforts to support and reward “scholarly productivity” has given way to the more coercive demand for “research products” and “high impact practices.”

The rise of generative AI that provides citation (accurate and otherwise), then, is not an entirely unwelcome arrival of a disruptive force in academic culture, but as a natural extension of diverse commitments to an ecosystem that both supports citational politics and that seeks to quantify citations as evidence for impact and productivity. In other words, we brought this shit on ourselves.