A colleague bans ChatGPT in his course. He does it anxiously, and with reasons: he worries about plagiarism, about dependency, about watching his students ask a machine for what they ought to think themselves. The worry is legitimate, sharp, pedagogically significant. And he did not change a single line of his assessment. He still asks for the same essay, the same prompt, the same grading criteria as five years ago.
There is something here that moral psychology has known for decades and that teacher education has been slow to absorb. James Rest put it precisely: moral sensitivity — perceiving that a situation has ethical implications — is one thing, and moral competence — knowing how to act on that perception — is another. You can have the first without the second. My colleague has it. He perceives with total clarity that unrestricted AI use raises a question of equity. He does not know what to do with that perception.
When I validated the CDD-IAG model instrument with teachers from four countries, a finding appeared that first struck me as a statistical curiosity and turned out to be a pedagogical thesis. Ethical concern about AI behaves as a cognitive register of its own, with a life independent of the other competences. In plain terms: a teacher can be deeply uneasy about these tools and have no skill whatsoever to craft an effective prompt, critically assess a generated response, or design a sequence that promotes critical use. And the reverse. Someone technically fluent can be ethically negligent.
An analogy that does not illustrate but proves the point. A surgeon’s clinical competence and bioethical competence are not degrees on a single scale. A surgeon can be technically impeccable and ethically negligent, or ethically sensitive and clumsy with the scalpel. Medicine learned, through decades of error, that it must form both capacities separately and then integrate them into the same act. No one supposes that sensitizing a resident ethically makes them better at operating.
In teaching we still suppose exactly that. We suppose an “AI ethics” workshop settles the matter, or that teaching how to use the tool smuggles in the reflection on who benefits and who is harmed by its use. These are distinct interventions on distinct substrates. Ethical formation is not solved by teaching how to use the tool. Technical formation is not solved by sensitizing.
What does this mean for the colleague who bans? That his ban is a moral emotion in search of a pedagogical form, and it has not found one yet. Banning is easy and feels like action. Redesigning a task so that AI use becomes visible, discussable, and assessable as part of the process — that is competence, and it is another thing, and it costs.
Unease is the beginning. It was never the end. Between perceiving that something is wrong and knowing how to design the activity that addresses it lies a distance no ethical manifesto crosses for us.
The question, then, is not whether you worry. Almost all of us worry. The question is what you designed last Monday with that worry.
