Most ethics training in science is useless. Not because the ideas are wrong, but because it treats ethics as a sermon when it should be a workflow.
We tell grad students not to fabricate data. We don't redesign the grant deadline that makes fabrication the rational choice. We lecture about "responsible AI" while the review process rewards speed over rigor. The incentive structure is the problem; the ethics module is the window dressing.
Hallucinated references, flexible data definitions, p-hacking shortcuts — these aren't moral failures. They're rational responses to broken reward systems. Publish or perish doesn't just pressure people to publish. It pressures them to publish what gets cited, which often means publishing what people want to hear.
The fix isn't more ethics seminars. It's redesigning the workflow: open peer review, replication incentives, team-based credit models that don't privilege first authorship on flashy results.
If your ethics initiative doesn't change what gets rewarded, it won't change what gets done.
What other incentive problems are we just accepting as "how science works"?