Red flags when evaluating an AI ethicist candidate yourself
If you are vetting without us, here is what tends to separate a real practitioner from someone who has learned the vocabulary.
Not every organisation vets a candidate through us, and that is fine. This is what we would actually look for, and it is less about interview technique than about whether someone can reason through a genuine value conflict rather than resolve it by pretending it is not there.
Can they name the actual conflict, not just the topic
Ask about a case with a real trade-off, a fraud model that is more accurate with more personal data, a moderation system that is safer with more false positives, a lending model more inclusive with less certainty.
A theoretical answer identifies the topic, "fairness," "privacy," "safety," and stops there.
A strong answer names the specific values pulling against each other and says plainly which one it would prioritise here, and why, rather than implying a clever enough technical fix would make the trade-off disappear.
Do they reach for a technical fix too quickly
Some tension between values is genuinely structural, not solvable by better tuning, more data, or a smarter architecture, see our piece on why fairness has more than one competing definition for a concrete version of this. A candidate who treats every ethical tension as an engineering problem waiting for the right fix has not fully grasped that some of these trade-offs have to be decided, not solved.
No answer for "what happens when you say no"
A candidate who has never had a recommendation rejected, or cannot describe what that felt like and what they did next, has either never been tested or is not telling the whole story.
The honest answer includes friction: an escalation that did not land well, a deployment that shipped anyway. Someone with experience carrying an ethical position under pressure has a specific, sometimes uncomfortable story here, not a smooth one.
Certainty where the field has none
AI ethics is a genuinely young field with unsettled questions, on measurement, on which fairness definition applies where, on what a good outcome even means in a specific context.
A candidate who answers every question with total certainty has either stopped updating their view against new evidence or is presenting more confidence than the underlying reasoning actually supports! Comfortable uncertainty, paired with a clear account of how they would work through it, is the more credible pattern, and the one worth hiring for.
See how we evidence it.
Every placement we make is evidenced against the same framework, with the reasoning shown.