Most schools are having the wrong argument about AI. They are debating how to catch it — detectors, lockdown browsers, honor pledges — as though the goal were to preserve a world in which take-home essays and lab reports still measure something. That world is gone. A student can now generate a polished, correctly formatted, entirely plausible writeup of an observation they never made. If your assessment can be defeated by a chatbot, the chatbot is not the problem. The assessment is.
This course takes the opposite posture, and it has two halves that only work together.
Half one: teach students to use AI well
We do not ban the tool. Banning it is both unenforceable and dishonest — the scientists and engineers these students will become are already using AI every day, and a school that pretends otherwise is preparing them for a world that no longer exists. So we teach AI as an explicit skill:
- How to ask it to explain a concept three different ways until one clicks — why the Moon shows phases, how parallax gives a distance — and how to check the explanation against the textbook and the sky, because it will confidently invent things.
- How to use it as a tireless practice partner: "give me ten sky-motion problems at increasing difficulty, then check my work and tell me where my reasoning broke."
- How to spot when it is wrong — a planet placed in a constellation it can't be in, a moon phase that can't occur at that time of night, a distance with the wrong units — which is itself a stiff test of whether the student actually understands the astronomy.
Used this way, AI becomes an accelerant for the Learn and Master stages. A student who can interrogate a model, catch its errors, and use it to drill themselves is learning faster than any previous generation could. We want that.
Half two: assess where AI cannot reach
And then — this is the part that makes the first half safe — the grade does not come from anything a student does alone with a screen. It comes from three in-person demonstrations a chatbot is structurally incapable of doing for them. Each maps onto one of the course's lab-notes essays.
- The observation-journal defense. The student presents their term-long sky-observation journal and defends it on the spot — what was the Moon doing this week and how do you know? where did the planet move, and why? what does this pattern in your own pages mean? AI can describe the night sky. It cannot make this student's nightly observations or answer for this student's record.
- Timed sky identification. Handed a star chart and a patch of sky — the real one, or a planetarium projection — and a finite amount of time, the student must name the constellations, find the visible planets, and reason about what should be up tonight and why. It is reasoning under a clock, against the actual sky, where a wrong call is visible overhead. There is no prompt to type.
- The oral journal walk-through. The student sits across from the teacher and walks through their own journal — this entry, this crossed-out sketch, this night the clouds rolled in — and explains the thinking behind it. Generated text has no memory of a Tuesday evening under the stars. The student's own journal does, and so does the student.
We are not trying to build an assessment AI can't help with. We are building one AI can't replace the student in — because the thing being measured is whether this person can actually do the astronomy.
Why this is the honest answer
The two halves need each other. Teaching AI without changing assessment just hands students a faster way to cheat. Changing assessment without teaching AI leaves them unprepared for the tools their field already runs on. Together, they resolve the tension completely: a student is encouraged to use every tool available to learn, precisely because the moment of accounting is a live one — their own eyes, the real sky, their journal, their voice.
This is what we mean when we call the course AI-proof by design. Not walled off from the future, but built so that the future's most powerful tool makes our students more capable rather than less — and so that, when it comes time to show what they know, there is no screen to hide behind. Only the sky overhead, and a person who has actually looked up.