Back to School

The case for and against teacher AI use

Audio version available here.

It’s back-to-school season, and the AI concerns are mounting among my colleagues. The big question on every educator’s mind is, “What are we going to do about student AI use?” I look forward to hearing the strategies my colleagues have come up with to answer this. I have very little advice for them, since the answer is highly dependent on subject, grade level, expectations, etc. Instead, what I’ll discuss in this post is teacher AI use. Many educators are very resistant to using AI to help them with their jobs. In this post I’ll run through the arguments I’ve heard against AI use by teachers, and then make some suggestions for ways in which AI can be of incredible value in and around the classroom.

Common concerns

AI violates students’ privacy. That can be true, but it’s not inevitable. Most model providers have a way to set your AI interactions so they don’t use your data. And many third-party AI-powered services guarantee student records are kept private. For example, Canvas recently announced integration with OpenAI’s models, but those interactions are kept completely hidden from OpenAI.

Using AI means you are not doing your job. This is one I hear a lot. People worry that students are cheating by having their essays generated by AI. If teachers use AI to help grade those essays, then we get a pointless exercise of computers grading computers, which is completely antithetical to what education is supposed to be about. But automated grading systems are not new to AI, and there are clearly acceptable use cases. No one said teachers weren’t doing their job if they used paper scantron forms to grade multiple choice tests. A teacher using a spreadsheet to compute student’s final course grades is clearly acceptable. The point isn’t to replace judgment; it’s to automate the mechanical parts so human feedback can go deeper where it matters.

AI feedback isn’t going to be as good as that given by a professor. Having AI give feedback directly to students is only one way to use it, and is certainly not appropriate for all subjects. However, I have heard this criticism in places where TAs and undergraduate graders are common. In those cases the students aren’t getting professor feedback anyway, so the AI feedback would only have to be as good as what an average TA or grader might give.

Where AI can help

So how can AI help teachers? Again, that’s going to depend a lot on what grade level and subject you teach. Since there’s no single answer I’ll give several examples:

—Assignment design. Modern AI systems have a pretty deep understanding of most K-16 subjects now. With that understanding comes an ability to aid in novel assignment design which promotes student learning. Of course this is no substitute for teacher subject-matter expertise, but educators may at least find it useful as a brainstorming partner.

—Assessment creation. Any assessment device is always going to be the responsibility of the teacher. However, AI can be incredibly helpful to teachers in making first drafts of essay prompts, quizzes, exams, etc.

—AI tutoring. In some subjects it may be appropriate for teachers to lean into the ability of AIs to reinforce concepts. For example, a teacher may ask students to spend 20 minutes at night with a chatbot discussing some model of the atom, or some historical incident, or some mathematical concept. Of course, teachers must always be wary of the possibility that an LLM may “hallucinate” and just make something up. On well-established, K-16 subjects AIs are now generally trustworthy. With that said, educators have always had the responsibility of teaching students to fact-check what they read, and generally exercise good research skills.

—Creating activities. This is perhaps more useful in lower grades, but I believe all teachers can do this to some extent. I’ll give an example. A few months ago I gave a talk on neural networks for seventh-graders. To get the main idea across I created an activity with the aid of an LLM in which each student became a neuron, and the whole class worked together to become a multi-layer network which simulates an auto-regressive language generation model. Our whole-class network only generated simple five word sentences from a vocabulary of about 10 words, but it got the basic idea of a language-generating neural network across. The activity was an idea I had, but the AI assistance was invaluable for tweaking the design so it worked smoothly.

—Grading assistance. As I said above this is more controversial, but I believe there are acceptable use cases. AI-assisted grading platforms are in current development, and I expect they will become commonplace soon. They will likely take different forms for different subjects. AI generated feedback for creative writing assignments is very different than AI assisted grading in a Chemistry class. Not all teachers will benefit from this equally, but I believe we should all be open to this technology as it develops.

---

Whatever your personal feelings about AI, there’s no example in history of a powerful technology being invented and then abandoned. Choose whatever aphorism you want. The cat is out of the bag. Pandora’s box has been opened. You can’t put the genie back in the bottle. For better or for worse, AI is here to stay. And while powerful AI is going to affect everyone, few professions are more at the front lines than teaching. So if you are a teacher, why not use AI to help fight that battle? AI should never replace the judgment, creativity, and care that teachers bring. But teachers who learn to harness it will have more time, more energy, and more ways to connect with their students. That’s a future worth working toward.