By Benjamin Bushong
The rapid development of generative AI like ChatGPT, Claude, and Gemini has ignited both excitement and concern within the academy. These tools purport to offer immense potential to reshape learning and teaching practices. I believe that this potential comes with large and inescapable risks, particularly when students rely on AI—the broad umbrella term I will use throughout this article—as a substitute for their own cognitive efforts. However, I also believe that research-active instructors can use themselves as a model for the appropriate inclusion of these tools in the classroom. An ostrich approach simply will not work. In this brief article, I thus argue for a balanced approach: AI should be used in the classroom, but in a fundamentally similar way to how researchers utilize it—as an aid to enhance human cognitive processes, not as a replacement.
Although its diffusion depends on field, AI has become a key input in the production function of academics, with applications ranging from coding assistance to editing and initial drafting of arguments. Although the use cases are broad, in a nutshell it seems to me that researchers are using AI to handle routine tasks, allowing focus to shift toward more complex, creative, and critical aspects of the work. For example, AI tools can help debug code or generate initial structures, freeing us to concentrate on, for example, refining algorithms and solving complex problems. Our students should be allowed to apply AI similarly.
This does not preclude some forms of tedious work. For example, an extreme view would be that since AI can program (for instance, it is a stronger Python programmer than the author of this article), we no longer need to train our students to program. Recently Nvidia co-founder and CEO Jensen Huang suggested that programming is no longer vital: “Over the last 10-15 years, almost everybody… would (say) that it is vital that your children learn computer science, everybody should learn how to program. In fact, it is almost exactly the opposite.” I humbly disagree. I believe our experience as researchers suggests a middle way wherein students learn to program with helpful guidance from AI tools. Understanding the basic logic of a “for” loop never gets old.
In writing, AI can catch minor errors and suggest revisions, improving the readability and clarity of a document. However, we as researchers still need to engage deeply with our content to ensure that the AI’s suggestions align with their intended meaning. Writing is a cognitive task that develops critical thinking and can be useful to structure our thoughts. A middle way would suggest a mixture of AI-guided writing and “free writing”. This approach may attenuate the ever-feared “fully automated essay”.
More broadly, I believe my experience in behavioral economics can offer a few simple insights into how students might interact with AI and the potential pitfalls.
A broad literature on present bias—the propensity to prefer pleasurable things now at the cost of the later—suggests we proceed with some caution. Students may prefer immediate rewards—such as the quick completion of assignments using AI—over long-term educational gains like developing critical thinking and writing skills. This bias can lead to a preference for AI-generated content, even when it’s not in students’ best interest. A middle way requires frequently addressing the longer goals in our instruction.
The related temptation to procrastinate, (gaining the pleasure of relaxation now at the cost of stress later), can lead students to turn to AI for last-minute completion of assignments. A middle way includes incremental deadlines, and structured assignments can help mitigate this risk by encouraging consistent engagement with the material.
Finally, AI can facilitate fast, intuitive thinking—famously described by Daniel Kahneman and others as System 1 in a two-system framework. But deeper learning requires slow, deliberate thinking: System 2. AI does not introduce a new problem here but can exacerbate the worst of students’ existing predilections. We should thus design assignments that encourage students to use AI as a tool to initiate System 1 thinking, followed by critical engagement that activates System 2 thinking.
When used judiciously, AI can be a powerful tool in education and it must be on the minds of the conscientious educator, lest we be perceived as even more luddite than our current assessment from the student body. And beyond our collective appearances, I believe that incremental inclusion of AI into pedagogy will improve quality of instruction and allow our students to transition to work life more effectively.
Benjamin Bushong is an Associate Professor of Economics in the College of Social Science.