Why not use AI?

We don’t have access to the algorithms generative AI platforms use, so insofar as we know, they work like glorified search engines, searching the web, collecting information, and looking for common patterns*.  The result almost certainly reflects some median consensus of the loudest or most common voices on the internet, rather than the most expert or well-informed.  When generative AI is used for artistic works, the result is often loaded with tropes and termed “AI slop.”  Why should the written output of generative AI be any different?

Why not use generative AI?  Much of what we do in academia, and any career path in general, involves implicit claims about exceptionality.  If I submit a manuscript to a journal for consideration, I am making the implicit claim that my manuscript stands above the other hundreds of manuscripts the journal editor has received.  If I ask my supervisor for a raise or promotion, I am making the implicit claim that my work stands above and beyond that expected of my present situation.  But if I cannot do my work without generative AI, then I am making the implicit argument that my thoughts and my work are no better than some median consensus.

*I should be careful to make a distinction.  There are some tools, such as hill-climbing learning algorithms and various analytical tools, that arguably share some qualities with generative AI, and they, like pocket calculators, have immense utility for solving complex problems that would be too time- or resource- intensive for humans to solve unaided.  In such cases, these devices are not a substitute for the human mind; rather, they extend its capabilities.  Such uses of “artificial intelligence” seem distinct from uses in which algorithms are being used to accomplish tasks perfectly within the abilities of the human mind.

What is the role of generative AI in academia? All too often, the discussion is simplistic—should we use it or not—with people in the “not” camp dismissed as hopeless Luddites. AI is the future, and the future is here, so it should permeate every aspect of the academy. Or should it?

There is a difference between using AI to explore the unknown and relying on AI to answer questions we already know the answers to. Academics ask both types of questions. In my own discipline, biology, we have been using tools that are arguably AI for decades, as we seek to explore phylogenetic relationships (for a tree-exploring “learning” algorithm is machine-learning in nature). Moving forward, AI holds great promise in extending the frontiers of understanding the three-dimensional structures of proteins, for example. Researchers often use generative AI as a tool to push beyond the frontiers of present knowledge. That’s all a great use of generative AI as an exploration tool.

But in academia, we also ask our students to explore questions we already know the answers to, especially in our foundational courses. The purpose of asking these questions is not to find the answers, because we already know the answers and have for decades. The purpose of asking such questions in foundational classes is to instill in our students the background knowledge, vocabulary, and ways of thinking that they will need in order to be successful in the discipline. Foundational classes are not about exploration but are about preparing students to explore. Students need to know what questions are reasonable and how to ask them, so that some day, when they do press up against the edges of knowledge, they will know how to move forward—perhaps by creating effective AI prompts. Generative AI is not a useful tool in foundational classes, because it quickly becomes a substitute for doing the hard work of building a foundation in the discipline, since asking the computer is easier than thinking.

There is a role for foundational classes that explicitly teach AI usage, but those should be separate from foundational classes in the disciplines, because including such information in the discipline-specific courses takes space away from the subject matter of the courses and duplicates efforts elsewhere. It’s also important not to succumb to academic trends to make discipline-specific foundational courses more relevant or timely by throwing in a dose of AI. Let’s just be honest: A course such as Introductory Biology is roughly set in the period 1860-1960, because that is when the modern discipline of biology developed. Nothing will change the fact that in order to use AI to explore the evolution of protein structure in populations experiencing climate change, a solid foundation of long-dead researchers such as Gregor Mendel, Charles Darwin, Ronald Fisher, Ernst Mayr, Rosamund Franklin and the like is required.