An ethics dashboard can map topics, not a student's character
Can language analysis help instructors see how students reason about AI ethics? A study of 90 graduate students combined expert labels with a language model that identified moral-value themes in written responses. The system tracked topics such as fairness, privacy and accountability, as well as links between problems and proposed solutions. Participant-grouped cross-validation produced a macro F1 score of 0.83 for the coding task. Students' responses also changed after instruction, but the one-group design cannot attribute those changes to the module. Crucially, broader coverage or a more balanced distribution of ethical terms is not the same as better moral judgment. The analysis can help an instructor locate omissions to discuss; it cannot establish a student's character or replace evaluation of the argument.
Why it matters for leaders
Use any such dashboard formatively: ask what a response overlooks, then let a teacher examine the justification. Test on new scenarios and a separate cohort. Exclude automated moral rankings or disciplinary inferences from the purpose of the tool.
The evidence
Expert-labeled responses supported internal model evaluation and pre/post semantic analysis. Grouping by participant reduced leakage between validation folds; held-out predictions were not manually corrected.
Who was studied
90 science and engineering graduate students; one university AI ethics module.
What the evidence cannot tell us
One cohort, repeated prompts and no control group constrain conclusions. Rare labels were less stable, and internal cross-validation is not external validation in a different institution or age group.
Read the original sources
Summary and leadership analysis from the Week ending 2026-11-06 brief. The edition’s selection notes explain source access and evidence limitations.