There is a growing sense of concern about the widespread adoption of generative AI among college students. As one UChicago student wrote in a Maroon article, “the purpose of generative AI is to avoid doing work,” further suggesting that its usage in classrooms was comparable to “giving elementary schoolers calculators while they’re learning arithmetic.” They ultimately argued that allowing AI in academic settings would produce graduates incompetent to tackle the world’s challenges. Written in the context of the Maroon themselves recently publishing a near-completely AI-generated article recapping a Chicago Bulls game, it’s reasonable to see how AI utilization can allow people to avoid having to learn how to approach work. However, the stakes of this argument hinge on treating all work completed with AI as otherwise valuable to an individual’s learning, a stance I believe overlooks an important distinction.​

Generative AI certainly replaces work, but I would argue its effects on the thought processes of minds faced with novel questions can be seen as more of a shift than a replacement. AI changes where thinking happens. Anthropic themselves brand their Claude LLM as “the AI for Thinkers & Problem Solvers”. When used properly, AI reduces the mental effort required to execute the steps of a task while also increasing the importance of abstractly framing information into answerable questions. Thus, it is possible for generative AI to make work more efficient without undermining one’s understanding of the processes employed to find solutions. For example, the hardest part about algebra questions is fitting the information of the world into solvable equations. From that point, given that we understand the order of operations, the use of AI in theory would increase work efficiency by solving equations without detracting from one’s understanding of algebra.

Technology has followed this pattern for a long time. Calculators removed the need to do arithmetic by hand. Spelling checkers reduced the effort required to write pristinely. Code editors now handle large amounts of repetitive syntax. In each case, the focus is moved away from execution and towards what is really trying to be accomplished, something that ought to require human judgment to be most impactful.​

Of course, this outcome depends on how AI is used; as easily as AI can increase one’s work efficiency and allow them to focus on higher-level thinking, it can be used to replace work altogether, thus eliminating all parts of organic thought. This fact gives some merit to the student’s argument that there are certainly many individuals utilizing AI in this manner: those who copy-paste questions directly into LLMs and copy-paste those answers back into assignments, or students who rely on AI to understand material without trying to grasp it themselves first.

​The recent incident with the Maroon illustrates this dynamic in a more public way. A now redacted article that was exposed to be largely AI-generated from start to end made it through editing and was published with sentences and ideas that were grammatically correct, but inhumanly hollow. What went wrong there was how AI was used. All thinking behind the piece had effectively been bypassed. When this foundational step, the most important step in fact, was skipped using LLMs, the product produced displayed the necessary aspects to be considered an “article” and be published, but really lacked what’s most important: a purpose.