The most-talked-about topic at a national political science conference was not what I expected

Every April, several thousand political science professors gather in the Chicago Loop for a weekend of sharing scholarship and advancing the academic cause at the Midwest Political Science Association meetings. In addition to preparing my own research for presentation, each year I anxiously await hearing my colleagues’ latest takes on the most important questions in the discipline. But the big takeaways from this year’s event were not what I expected.

Given that this is the first national political science gathering in the US since the V-Dem Institute downgraded the US from “liberal democracy” to “electoral democracy,” I went into MPSA expecting the status of US democracy would be a major topic.1 It was, but there was another topic that everyone wanted to talk about more.

A.I.

Nearly every conversation I had included some lamentation about the role of Artificial Intelligence in the academy and all the challenges it poses for research and teaching. The challenges are enormous, posing threats and opportunities to every aspect of a college professor’s job.

On the research front, using AI to supplement an academic’s workflow has the potential to dramatically increase the rate at which scholars can generate research. Whether one is a qualitative scholar processing field notes, documents, or interviews, or a quantitative scholar gathering and processing data for analysis, AI dramatically reduces the time and effort for scholars.

On one hand, this “skilled personal assistant” is exciting. It can level the playing field for scholars who do not typically have access to research assistants. It eases the burden of time-intensive tasks that can seriously bog down a project. It lowers the costs of investigating hypotheses, allowing one to advance their contribution more quickly.

On the other hand, this will produce an arms race in published research—even more intense than the one already in place. Academics are already under enormous pressure to publish peer-reviewed articles and books, and are threatened with job loss if they do not meet their institution’s standards. If everyone can more easily and quickly produce research, then the incentives to churn out work intensify for everyone.

But the system is not designed to carry this load. I had a conversation with the editor of a major journal who told me that the journal is not yet seeing an increase in submissions, but they know it’s coming, and expect to raise their standards in response. In the end, this increased competition is probably a net-win for science, but a net-work-stress-multiplier for individual scholars.

Assuming we see a big increase in the number of papers being submitted for peer review, and knowing that the number of available peers does not change much year-over-year, scholars should expect to be asked to review more work, which is unpaid labor. Exhausted and stressed people look for shortcuts, and A.I. is sitting right there. This will increase the incidence of scholars using A.I. to perform, or at least help with, the peer review process, which, if overdone, distorts the very purpose of peer review. It’s not called “A.I.-review,” and the journal editor I spoke to told me they have an explicit policy against reviewers using A.I. to review research, but enforcing this standard will be difficult.

And then there’s teaching, where A.I. is a total game-changer for instructors and students.

Applying A.I. to college-level teaching is complicated. Instructors are expected to both restrict A.I. usage sufficiently to make genuine assessments of student learning and incorporate A.I. into the curriculum so that students gain skills and experience with the new tools. These goals are often in conflict with one another.

One panel I attended focused on the challenges and opportunities of pedagogical applications of A.I. Here are some of the most important takeaways I got:

• Revisit (and perhaps revise) our learning objectives and assessments to ensure they are aligned. For example, that assignment you’ve been using for 10 years that you think is both helping students to evaluate quality academic sources and teaching them about the principles of collective action probably cannot meet both of those objectives anymore, as designed. Which learning objective is paramount? Does the assignment match the objective?

• A.I.-proof strategies for assessing student learning:

  • Cold calling in the classroom

  • Pop quizzes

  • In-class writing

  • Blue book exams

  • Require students to write in a platform with trackable histories (e.g., Google Docs) to ensure they are not pasting content.

  • Make assignments feel relevant (more cheating occurs on required assignments that feel like busy-work).

  • Use oral exams, or keep this as an option if content provenance is in question.

  • Require students to use journaling, reflections, and commenting on the process they use to create artifacts.

  • Allow students to use A.I. to help with writing tasks, but require them to submit both the original text and the revised text to document changes.

  • Require students to create appendices to describe and document processes and A.I. usage.

  • To guard against students reading A.I.-generated summaries of assigned readings rather than the original text, in class, give students quotes from the text and ask them to critically engage with them (e.g., describe the context, meaning, importance, contribution); or, ask students to identify which quotes are truly from the text and which are not, providing justifications. Closed book, of course.

• All syllabi should include a clear A.I. use policy that aligns with any university and department policies and clarifies consequences for policy violations.

• Add an A.I. literacy module to every course that students complete on their own. This might include:

  • What is A.I.? How does it work? What platforms are out there? What are they good for? Limitations, risks, and hallucinations.

  • Responsible and ethical use of A.I., including academic integrity, and clearly defining plagiarism, attribution, and citation expectations.

  • Awareness of bias in A.I. inputs and outputs.

  • Discussion of data privacy and the consequences of feeding personal information or copyrighted material to A.I.

  • Environmental impact of A.I.

  • Provide prompt guidance using principles (e.g., CLEAR: concise, logical, explicit, adaptive, reflective).

  • Develop habits for evaluating A.I. output.

  • Provide examples of appropriate and inappropriate applications of A.I.

  • Provide suggestions for how to use A.I. as a learning assistant (e.g., summarize notes, outlining, brainstorming, improving grammar, revising without using A.I.-generated prose)

I left with ideas that I will likely experiment with in my own classes, but still many questions, particularly around teaching political science students computational skills. Learning to code in Stata or R has been a challenge for social science students for many years. Might A.I. give us the freedom to teach social science research design and basic statistics without teaching coding? Rather, teach students to prompt A.I. to write code and evaluate its output, rather than generate the code itself? Maybe—but it’s awfully hard to evaluate the output if you don’t have a basic understanding of the code logic and functions in the first place.

To be clear, my observations at MPSA are far from a random sample. I am at a mid-to-late stage in my academic career and tend to spend more time at these things mentoring younger faculty than getting feedback on my own work. I attended few panels that didn’t already include my name in the program and had scores of conversations in the famed hotel lobby.

Leaving Chicago, I’m grateful for the solidarity with university professors who are all working through the same challenges right now—teaching and researching in an era with unprecedented threats to US democracy, the academy, and a fundamental shift in the means of knowledge production. There are many more conversations to come.

Share

Guest Submission Form