They Had the Answer. They Just Didn't Understand It.

A few weeks ago I had one of those office hours sessions that showed me exactly what some students are doing with AI right now.

It started the way busy office hours usually do. I had been helping a student work through equilibrium problems, when two more walked in with their own questions. I got the new arrivals started and after a few minutes circled back to the first student, let's call them X. They'd been plugging away on RICE tables while I was across the room, so I expected them to be further along. Instead they looked up and said, "Dr. Coates, I don't understand why this reaction is going in the reverse direction."

Now, I could have just told them. It would have taken thirty seconds. But as a good instructor would I asked "Well what do you think? What did we do in class to figure out which direction a reaction needs to go to reach equilibrium?"

X barely thought about my question, they just looked at their laptop and then turned it toward me. They had typed the homework problem into ChatGPT, and the RICE table in the output showed the reaction going in reverse. They had the answer. They just didn't understand it.

I didn't say anything about the ChatGPT window. I prompted them to check their lecture notes and in-class activity from that Monday's class. As they were searching I prompted, "What were our steps for the last 3 problems on the activity?" They flipped through the activity sheet, and then I watched the lightbulb moment happen in real time. Their eyes went wide. "OH! We need to calculate Q and compare it to K!"

I then suggested they scroll up in ChatGPT. And there it was, Q calculated, direction determined, every step laid out similar to what we'd done in class. The answer to their question had been sitting right there above the RICE table. They just never scrolled up to see it.

I told them something I mean genuinely: if you're going to use ChatGPT, keep questioning it. Don't just accept the answer, understand it. And honestly I was proud of them. They were given the answer, but they actually wanted to understand. They could have plugged that final answer into our online homework system, gotten the points, and moved on with their day. They didn't.

But I'd be lying if I said most students are making that choice right now.

I know because our online homework systems track time on task and time to completion. A couple of years ago, one or two students would finish suspiciously fast (5 minute total completion time that should be at least an hour!). Now it's a consistent handful. And when exams come around, the gap is obvious, not because the material is too hard, but because homework is where students actually build the neural pathways for solving these problems. That's where the mastery happens. If AI is doing the practice for them, they're showing up to the exam having practiced almost nothing outside of class time.

This shortcut is not entirely new, to be fair. When I started teaching, the issues were students working in groups and just copying each other's answers, or guess-and-check cycles in online systems. The method has changed. The shortcut is the same.

What's harder to solve is the homework dilemma itself. What used to work was implementing "homework journals", requiring students to show their work on paper for the online homework. Low point value. But recently low compliance. I thought students were blowing it off as busy work. Now I think the real reason is simpler: they aren't doing the homework themselves, so there was no work to show. I've toyed with reducing homework point values so that cheating on it is less rewarding, but that just punishes the students who are doing it honestly and using it the way it's meant to be used.

And honestly, I don't want to penalize anyone. While I know students obsess over grades, that's not the point for me. I genuinely want them to learn. I want them to sit down in front of an exam and feel prepared. Every workaround I've tried has been in service of that, not catching cheaters.

I don't think the answer is banning AI. That ship has sailed, and frankly, these students are going to work in a world where AI is everywhere. What I want is for them to use it the way student X accidentally used it, as something to interrogate, not just accept.

That's the problem we built LabNotes.ai to solve. It uses the Socratic method, so instead of handing students the answer, it asks them questions back. It cares less about whether they can produce the final answer and more about whether they can understand the process to get there. The 24-hour AI tutor, not the 24-hour answer key.

I still think about student X, about how close they could have been to just copying that answer and never knowing they already understood the concept. They just needed someone to ask them the right question first.