A New Tool for Education

I spend my working life around supercomputers. One thing you learn fast in that world is that a new one doesn’t just let you run your code faster. It changes which questions are worth asking in the first place. Generative AI looks like that kind of enabler, and it has landed in the classroom whether or not any of us feel ready for it.

It’s not the first time a tool like this has shown up in a classroom. When the handheld calculator landed in the 1970s, plenty of teachers called it cheating. Why would a kid bother learning long division if the answer was a button press away. That worry wasn’t unreasonable at the time. What actually happened was less dramatic and more useful. The calculator did the arithmetic, which meant students could spend the same hour going further: working bigger and more challenging problems, checking whether an answer was even plausible, setting up the question in the first place. It wasn’t a shortcut around thinking. It was an amplifier. I suspect AI is going to follow the same arc, only on a much larger scale and on a much faster clock.

I want to be even-handed here, because I think the honest position is an uncomfortable one. AI in education carries real promise and real hazard, often inside the very same feature. So instead of reaching for a verdict, I want to work through the two questions I keep being asked: what is the role for students, and what is the role for faculty. I’ll take them in order.

The role for students

The most valuable thing AI offers a student, to my mind, is not answers. It is a way to explore. A capable model is a tireless companion for poking at an idea. What happens if I change this assumption. What would the strongest counterargument be. Where did this concept come from, and what am I not seeing. In the lab we’d call this the exploratory phase of a problem, the stretch of work before you even know what the real question is. Students have always had to do that phase mostly alone. They no longer do, and that is a genuine gain.

Close behind exploration is the role of tutor. A student stuck at eleven at night on a problem set now has something patient to ask. It will explain the same proof three different ways. It will not sigh. For a student who never had a parent who knew calculus, or a family that could afford a tutor, that is not a small thing. Used this way, AI can be a real leveler.

It is also a drafting and feedback tool, and here I’d ask for a little more care. Used well, it helps you get unstuck on a blank page and gives you a fast read on whether an argument actually holds together. Used poorly, it simply writes the thing, and you hand in work you could not have produced and do not fully understand.

That points straight at the hazard, and the hazard is real. The same tool that can accelerate a student’s thinking can also quietly stand in for it. If the slow, frustrating work of wrestling with a hard problem is the thing that actually builds a mind, then a tool that removes the wrestling can remove the education along with it. The fluent answer arrives, the struggle gets skipped, and something important fails to happen. No syllabus policy fully fixes this. It comes down to whether a student understands what the work is for, and teaching that understanding is partly our job. Which is a fair place to turn to the second question.

The role for faculty

For faculty, the most immediate benefit is mundane, and I think badly underrated: time. A great deal of academic life is not teaching and not discovery. It is drafting, summarizing, reformatting, wrangling logistics, and writing the same category of email for the hundredth time. AI is genuinely good at that kind of work. Every hour it hands back is an hour that can go to a student, a problem, or a piece of writing that truly needs a human mind. I’ve watched the same pattern in research computing for years. When the tool handles the boilerplate, the scientist gets to spend attention on the part that is actually hard.

The deeper opportunity, and the harder one, is rethinking courses and assessment. If a take-home essay can be generated in seconds, then the take-home essay is no longer measuring what we assumed it measured. That is uncomfortable, but it is also clarifying. It forces us to ask what we were really trying to assess, and then to design for that directly: a defense of an idea out loud, work shown in stages, problems that reward judgment over recall. This is more work for faculty in the near term. It is also, I’d argue, overdue.

AI can serve as a teaching aid in its own right. It can generate practice problems without end, adapt an explanation to a confused student in real time, and let a whole class examine where a model fails as a lesson in itself. Some of the best classroom uses I’ve seen treat the AI as the object of study rather than the oracle. Here is what it got wrong. Why do you think that happened. How would you check.

The hazard for faculty mirrors the one for students. Under time pressure, it is easy to let the tool lower the bar without noticing: to grade carelessly with it, to teach from material we never properly vetted, to stop catching the moments when rigor has slipped. The tool does not protect rigor. Only we do.

What does not change

Here is what I keep returning to. When the student and the professor have the same powerful tool sitting on the desk, the old picture of the faculty member as the holder and dispenser of information is finished. That picture was already fading. AI ends it. What remains is, I think, the better part of the job anyway. Education was never really about the transfer of facts. It was about building judgment: the ability to frame a good question, to tell a sound answer from a plausible-sounding wrong one, to know what actually matters. A national lab runs on precisely those skills, because a supercomputer will faithfully return a precise answer to a badly posed question. The tool is only ever as good as the mind directing it.

So the role for students is to use AI to think harder, not less. The role for faculty is to model what thinking harder looks like, and to build an education that AI can assist but cannot counterfeit. The promise is real and the pitfalls are real, and I don’t believe we get to choose just one. We have a new tool in the classroom. The question worth asking has not changed: what are we actually trying to teach.