The Uncomfortable Questions · An AI FAQ for leaders
Question #24 · GovernanceWhat’s the one AI question you wish your board would ask, but never does?
The short answer
How do we know when our AI is wrong, and how fast do we find out?
That’s mine. Most boards ask about cost, compliance and competitors. Few ask about detection and recovery. That’s where the real exposure sits.
Norm E. Murray
The psychology
Why it happens
Boards ask the questions that are familiar, visible and comfortable. How an AI system fails, and how long it fails before anyone notices, is none of those things.
The agenda follows the headlines
The availability heuristic leads us to judge what matters by what comes to mind easily. Cost, regulation and rivals fill the business press, so they fill the board agenda, while quiet failure modes never make the list.
Directors see what management shows them
Economists call it information asymmetry: one side knows far more than the other. Management decides what reaches the boardroom, and AI updates tend to report wins and adoption, not the errors caught quietly last quarter.
Complex systems fail in ways nobody planned
Sociologist Charles Perrow argued in Normal Accidents that in complex, tightly connected systems, failure is not a rare exception but an expected feature. The question is never whether the AI will be wrong, only when and for how long.
Asking how it fails feels like doubting the decision
Barry Staw’s research on escalation of commitment shows that once we have backed an investment, we resist evidence against it. A board that approved the AI budget can find a question about failure uncomfortably close to second-guessing itself.
Your move
What to do about it
- Put the detection question on the agenda.At the next board discussion of AI, ask it plainly: how do we know when this system is wrong, and how fast do we find out? Notice whether anyone has a ready answer.
- Ask for a time to detect.For each AI system that touches customers, money or people, ask management how long an error would run before someone caught it. If nobody knows, that is the finding.
- Ask for the error log, not just the success story.Request the mistakes that were caught, who caught them and how. A report with no errors in it is a report, not a record.
- Rehearse a failure.Run a short exercise: the system has been wrong for a month. Who notices first, who decides what happens next, and what do you tell customers?
- Name who can switch it off.Every critical AI system needs a named person with the authority to pause it, and the board should know who that is before the day it matters.
Keep going
Related questions
Get a sharp take like this every morning
Norm’s Edge is a free daily briefing for leaders: one AI story, one decision, one minute.
Free. One email a day. Unsubscribe anytime with one click.
Prefer to read first? See Norm’s Edge