A high schooler writes about AI tools in the classroom

IMO getting too worried about this sort of homework “cheating” feels like the wrong way of looking at it. Although, there are lots of processes that accept and reinforce this wrong viewpoint.

For k-12, getting the parent and the student to sit down outside of school and “cheat” by having the parent teach the kid is… victory! You’ve reinforced the idea that learning can happen outside schools.

For college, having students get together and “cheat” by doing their homework together is… victory! You’ve gotten the students to network with their peers. That’s… like, the main value proposition of a university, to some.

The problem is when undue grade weight is put on these processes. It is a hard balance to strike, because you need to offer enough grade to incentivize the stuff, but not enough that it feels unfair to those who go individually.

As far as LLMs go, it offers an alternative to learning to collaborate with other humans. That’s bad, but the fix should be to figure out how to get the students to get back to collaborating with humans.

I don't think they were trying to prevent parents from working with children; I think they were trying to prevent parents doing the homework for children, or the kids farming it out to someone else online, or getting someone else to do it for them, period.

Same with college; it wasn't exactly networking when someone I knew paid someone to do their homework for them.

Right, that's delegating.

This is a far too charitable interpretation of the problem. Students who cheat in these circumstances aren’t working together with their peers or LLMs to understand the subject matter.

They’re using the LLM to bypass the learning part completely. Homework problem gets pasted into ChatGPT. Answer is copied and pasted out.

This is analogous to a student who copies a peer’s homework answers without trying to understand them.

This isn’t “learning to collaborate” or networking. It’s cheating.

In practice, it catches up to students at test time. This is the primary problem for my friend who teaches a couple classes at a local community college: Students will turn in LLM work for the assignments and then be completely blindsided when they have to come in and take a test, as if they’ve never seen the material before.

One time he assigned a short essay on a topic they discussed with a generic name. A large number of the submissions were about a completely unrelated thing that shared the generic name. It would not be possible for anyone to accidentally make this mistake if they were actually parsing the LLM output before turning it in. They just see it as an easy button to press to pass the course, until it catches up with them later and they’re too far behind to catch up to people who have been learning as they go.