Homework Is the Rehearsal: Why AI Education Needs to Measure the Process, Not the Output
Case of the Week, Min Wu, PhD · ai-public-health.com
Two students in my health informatics class submitted the same AI-assisted assignment last semester — a high-level design for a drug-drug interaction alert in a hospital’s medication ordering system. Both produced correct, well-organized designs. On paper, they looked identical.
I asked each of them the same question: walk me through how you got here.
The first student could tell me everything. She had started by listing where she expected the AI to fail before opening the tool — alerts that would fire too often, drug pairs the rule would miss, patients the alert would mishandle. She had distrusted the AI’s first pass because it produced a clean design without thinking about whether nurses and doctors would actually pay attention to the alerts in a busy shift, and she had pushed back with a follow-up prompt. She could describe the moment her own thinking changed because of a case the AI surfaced that she had not considered.
The second student could not tell me much beyond “I used ChatGPT.” The output was correct. The process was invisible — to me, and I suspect to him.
On my grading rubric, these two students earned the same score. That is when I knew my rubric was broken.
If you teach, you have already felt this. You are grading AI-assisted work and wondering whether the student learned anything or simply submitted what the tool produced. If you supervise students or junior staff using AI in practice, you have watched someone present a polished output and not be able to explain the reasoning behind it. The grading rubric says the work is fine. Your instinct says something was missed. That instinct is correct — and the problem is not the student. The problem is what we are measuring.
Why Outcome-Based Education Fails in the AI Era
Most homework in public health education is outcome-based. Did the student produce the right answer? Is the needs assessment complete? Does the policy brief cover the required elements? For decades, this worked well enough, because producing the right answer required doing the thinking.
AI broke that link. When a tool can produce outputs that meet every outcome-based criterion, the outcome no longer tells you whether the student thought at all. A student who iterated carefully with AI, questioned its assumptions, and corrected its errors gets the same grade as a student who pasted the prompt and submitted the result. Outcome-based homework has become unfalsifiable in the AI era. It cannot distinguish learning from delegation.
Worse, it teaches the wrong lesson. Students trained on outcome-based AI assignments learn that AI is a shortcut to the answer. That is exactly the opposite of what public health practice will require of them.
Consider what the AI cannot reason about, even on a task it appears to do well. A drug-drug interaction alert designed by AI looks clean on paper. The recorded medication list it operates on also looks clean. Both check out by every outcome-based criterion we have. But every clinician knows the gap between the recorded medication list and what the patient is actually taking — the doses skipped, the prescriptions filled but never started, the herbal teas and traditional medicines that the patient does not think to report, the over-the-counter drugs and supplements the chart was never designed to record, the old bottle the patient is still using from a previous prescription. That gap is not a data problem the AI can solve from the inside. It is a feature of how medication management actually works in patients’ lives. A student who builds an interaction alert without thinking about that gap has produced a correct deliverable for the wrong problem. A student who has thought about it has rehearsed the question that matters: what is the rule actually operating on, and what is it missing?
Public health education that remains outcome-based will produce graduates who have the degree but not the technique.
What Meta-AI Homework Actually Is
The alternative is process-oriented education — homework that measures not what the student produced, but what the student thought about while producing it. I call this meta-AI homework, and it is built on a tradition in cognitive psychology that goes back to 1979: metacognition, the study of how people think about their own thinking.
The AI-era update is to apply that tradition directly to how students use AI. Meta-AI homework has three phases, each aligned to a core metacognitive skill:
Pre-task self-assessment. Before using AI, the student asks: what do I already know about this problem? Where do I expect AI to help me, and where might it mislead me? What will I be watching for?
In-task guided questioning. While using AI, the student asks: is this output consistent with what I already know? What assumptions is the AI making that I would not? Where am I tempted to just accept the output without checking?
Post-task reflection. After using AI, the student asks: what did I learn from this iteration that I will carry to the next one? Where did my own judgment change the AI’s output, and where did I let the AI change my judgment?
This is rehearsal. Every assignment that asks these questions installs a little more of the capacity to use AI well under pressure — not by following a checklist in the moment, but by having internalized the questions so deeply that they run automatically.
Why Public Health Education in Particular
Public health has a distinctive educational challenge: graduates enter a field where just-in-time learning is the norm. Novel pathogens, unfamiliar community contexts, sudden climate-driven emergencies, new data sources that did not exist during their training. The content changes. The technique for thinking with new tools under uncertainty does not.
That diagnostic is useful — but only if the workforce has practiced applying it before the emergency arrives. Meta-AI homework is where that practice happens. It is the low-stakes setting where students rehearse the judgment they will need when the stakes are real.
The next pandemic, the next climate crisis, the next novel exposure — these will be met by whoever we trained. A workforce educated through outcome-based AI assignments will default to using AI fast, at every peripheral step, and wonder why the outcomes are unchanged. A workforce educated through process-oriented assignments will know where to plant and what kind of pole to use, because they have rehearsed it dozens of times before the moment came.
The Observer’s Insight
Education is how we install technique at scale. When what we measure changes, what gets installed changes. Outcome-based education in the AI era installs the habit of letting AI produce answers. Process-oriented education installs the habit of thinking with AI as a partner. Two different measurement choices. Two different workforces. Two different public health systems.
The decision is being made right now, in every syllabus, in every grading rubric, in every assignment a faculty member designs this semester. Homework is the rehearsal space. Change what homework measures, and you change what the workforce can do when it matters.
I hold one uncertainty openly: whether process-oriented homework scales to large classes without losing the metacognitive depth that makes it work. Grading a student’s thinking is harder than grading their output. I do not have a clean answer to this yet. But I am more worried about the alternative — graduating a generation of public health professionals who can produce correct AI-assisted outputs and cannot tell you why.
If you teach or supervise in public health: look at one assignment you give this semester. Is it measuring the output, or the process? If you mentor students using AI in practice, are you asking them about the answer they produced, or about how they thought while producing it? I would love to hear what you are seeing — reply or leave a comment.
The full meta-AI homework framework, including worked examples, a teacher guidance document, and grading rubrics, is available in Chapter 10 and Appendix of my textbook.
As a member of the Springer Nature Author Affiliate Program, I may earn a small commission from purchases made through this affiliated link to my book. Support the author by checking out my textbook, Artificial Intelligence in Public Health: https://tidd.ly/4mH9389
