The student shuffles over to my desk, fighting back a grin and carrying a very meh proof for an optional challenge problem.
I have already seen this same example generated several times this week by Chat GPT. Like all the others, this was meticulously copied from a confident source without any evidence of understanding. There are none of the usual spelling or mathematical notation errors I would expect to see on a first draft of a student’s proof.
I give away nothing as I scan the page, as if reading for the first time with interest. The fact that he has engaged with the process at all makes me want to keep him engaged. This is a student who once turned in a photo of a cheeseburger and fries instead of a picture of his actual homework.
The page shows no erasures, hesitation marks, or U-turns across a full half-page of complex, multistep proof. His diagram has been painstakingly reproduced from the problem source. It bears no markings, analysis, color-coding, or any other evidence of reflection.
Out of the corner of my eye, I can see my student looking pleased with himself, watching as I appear to read his work. His eyes are bright beneath a tumble of brown curls, and a faint smile is forming under the first hint of dark shading over his upper lip.
I point with my pencil at his diagram in the upper right of the page. “What are two things you can tell me about segment DC in this figure?”
The color drains from his face. “Wh — what do you mean?”
I tap on the pristine, unannotated diagram. “Segment DC, over here.”
I look up at him and blink. Then I wait.
The adrenaline shooting through his veins is palpable. “It’s – it’s … ” He squinches up his brow. “… the midpoint …?”
I nod vigorously. “Good! The endpoint C is the midpoint of … what exactly?”
He frowns again, searching intently. He is fully present in the thinking. “… segment AR …?”
Carefully hand-copying an A.I.-generated artifact is not a long-term meaningful substitute for getting educated.
I nod again and hand him a purple colored pencil. “What does that mean about segments AC and CR?”
“They’re … equal?”
“They’re congruent, yes.” I tap on the diagram. “Mark those.”
He colors the two halves of the segment and draws tick marks to indicate equal length. I can see that he knows I know he was bullsh—ing me, but now he is here in the moment, learning.
I return to the text of his proof, near the middle of the page. “Tell me how you got from step three to step four.”
He shrugs and shakes his head. This is the most honest interaction we have had in weeks. We both know it.
“See if your table mates can help you to explain how this transition between steps three and four makes sense. Then come back.”
He takes the paper from my desk and starts to lay down the colored pencil, but I wave him away, indicating that he should take it with him. As he returns to his table group, I can hear him starting to explain their challenge. Another student leans into the conversation in the middle of the table, and somebody else points at the diagram.
This is one of the things young teens don’t understand about generative A.I. — you can only get out what you put in. And if you don’t have much understanding to begin with, the prompt you provide is not going to be very deep or precise … or even interesting.
A.I. is not magic, though the experience of using it can definitely feel magical.
Typing a prompt and receiving a lengthy and seemingly thoughtful response or a vivid image that seems well-crafted and thorough feels miraculous — at least until somebody knowledgeable starts analyzing and points out that there are seven fingers on the child’s left hand. But this is a child’s-eye view, believing that somebody else’s shortcut will replace your personal obligation to develop your own best human thinking.
Sooner or later, somebody a little more advanced than you are is going to point out there’s a circular argument short-circuiting your proof.
This is not an indictment of A.I. Far from it. It’s just evidence of the kind of unavoidable hiccups we have learned to expect in the evolution of any complex technology. There’s an old saying in the software engineering community: first we put the bugs in – then we take the bugs out.
At its best, generative A.I. can be a worthy dance partner for students. But they still need to learn how to do their own dance steps – that is, if they actually want to learn how to dance in the adult world.
So while I hate to be a downer, it needs to be said — and probably more than once. Carefully hand-copying an A.I.-generated artifact is not going to be a long-term meaningful substitute for getting educated.
No one is going to be excused from their basic obligation to learn how to think in this lifetime.
Sorry, kids.