The Uncomfortable Questions · An AI FAQ for leaders

Question #10 · Trust

When your team says “the AI recommended it,” are they explaining a decision or avoiding ownership of it?

The short answer

“The AI recommended it” is becoming the corporate version of “the system made me do it.”

Often, people say it to avoid ownership. AI can inform a decision, but it can’t own one. The fix is one explicit rule: whoever acts on a recommendation owns the outcome, exactly as if they’d made the call alone.

Norm E. Murray

The psychology

Why it happens

Nobody sets out to dodge accountability. The phrase spreads because it solves a real psychological problem for the person saying it.

01

It spreads the weight

Psychologists call it diffusion of responsibility: the more parties involved in a decision, the less each one feels it belongs to them. An AI system never pushes back and never carries the consequences, so adding one quietly lightens everyone else’s load.

02

It works as a moral buffer

Researchers who study automation use that term for the distance a system puts between a person and the effects of their choice. When the model recommended it, a bad outcome feels like the tool’s error, not a judgment call that went wrong.

03

The incentives reward it

In most organizations, following the AI and being wrong is forgivable, while overruling it and being wrong is career-limiting. I call this the Override Penalty. Under those incentives, deferring is a rational way to protect a career.

04

Nobody learns from the miss

Decisions get made, but nobody owns them. When something goes wrong, there’s no one to learn from it, because everyone was only following the recommendation.

The bottom line: AI is an input, never an owner. If no human can explain why a decision was right, the decision isn’t finished.

Your move

What to do about it

  1. Make the ownership rule explicit.Say it out loud and write it down: whoever acts on a recommendation owns the outcome, exactly as if they’d made the call alone.
  2. Ask for the reasoning, not the source.When someone says “the AI recommended it,” ask: “And why do you agree?” If they can’t answer, the decision isn’t ready.
  3. Remove the Override Penalty.Judge people on the quality of their reasoning, not on whether they sided with the model. A sound override that turns out wrong deserves the same treatment as a sound deferral that turns out wrong.
  4. Put a name on every AI-assisted decision.In decision logs and board papers, record the person who made the call, not just the tool that suggested it.
  5. Model it yourself.When you use AI in your own decisions, say what it recommended and why you did or didn’t follow it. People copy what leaders do, not what leaders announce.