Good AI, Bad AI

At the start of the school year I asked my students to let me know how they are using AI in my courses. I’ve seen some Good AI, and some Bad AI.

Good AI

I make mathematical computing a part of my year-long Linear Algebra course. The varying levels of computer programming experience among my students makes this a challenge: some could intern at Google and some can’t remember how a for loop works. AI Coding Assistants, used appropriately, provide invaluable support for students with limited experience. If they can’t successfully write a program to add two rows of a matrix together, they can have the Coding Assistant do it, check to see if it works, and then review the code themselves and try to learn something. Good AI! Here, “appropriate use” means making an honest effort to complete the challenge yourself first: This builds context for learning from whatever code the AI produces, and it also better positions the student to evaluate whether or not the code actually does what they asked it to do.

Bad AI

At the beginning of my Calculus course I ask students to write about a “mathematical observation” they’ve had. I am intentionally vague about what a “mathematical observation” is. Some students write about analyzing their commute to school, some about optimizing a video game strategy, some about a number theory course they took in a summer program. One goal of the assignment is to learn about my students as individuals and as mathematicians, so knowing what they think constitutes a “mathematical observation” tells me something about them.

One student began their paper by disclosing that they first asked ChatGPT to define “Mathematical Observation” for them. This immediately struck me as Bad AI. Thinking about what constitutes a mathematical observation was the point. Not only did the student ask an AI tool to do their thinking for them, but doing so undermined the very purpose of the assignment: Instead of learning what the student thinks as an individual, I got some averaged-out sentiment from a non-random group of authors.

For the record, the student did write a lovely and thoughtful mathematical observation, but afterwards we had a good conversation about the role of AI tools. “Don’t use AI to do your thinking for you,” I said, which seems like a good place to start in navigating this new landscape.