Talk it through with Aurelius
Library›Aurelius›The problem
Aurelius · Work & Leadership
Knowledge + Guidance

When should I override the AI risk model instead of trusting it?

You sit with a model that scores risk, and something in you distrusts it. Good — that distrust is often correct. But distrust alone decides nothing. The real question is not whether the model is biased. It is. The question is what you do about it, today, on this one case in front of you. Most people in your seat do one of two things. They defer to the model because disagreeing invites scrutiny. Or they override on instinct because overriding feels like judgment. Both are evasions. One hides behind the machine. The other hides behind your gut. Neither is a choice. A choice has a reason attached — one you could say aloud to your supervisor and mean it. Build that habit now, before the stakes are higher than you think they are.

◆ How this problem reads on the two dials
GuidanceKnowledge
More coaching
Some to learn
1:1 with AureliusWith others (a Pod)
Mostly you & the coach
A little with peers
The core gap is a habit of disciplined judgment, not missing theory about bias, so coaching to build an override practice does more work than further explanation.
How the two dials adapt to you →
What’s really going on

You override when you hold specific knowledge the model lacks — not because the output feels wrong. Name that knowledge in writing before you choose to act. If you cannot name it, you are not overriding; you are guessing. The bias you fear in the machine is easier to see than the bias in your own judgment.

🔒 What you’ll build togetherUnlock by starting
A moveBefore you override, write down the specific fact the model did not have access to. If you cannot write it, do not override.
A moveKeep a log of every override for a month — what you changed, why, and what happened after. Read it alone, honestly, at the end.
A moveWhen you feel pulled to defer because overriding feels risky, name that feeling precisely — it is fear of scrutiny, not confidence in the model.
A moveSet your rule for what counts as sufficient reason to override before you are under pressure in a live case, not during it.
A moveWhen you see bias in the model but do not understand its cause, say so in writing rather than quietly working around it.

What changes unlock by starting

  • You build a written record that separates real judgment from guessing or fear.
  • You stop treating the model as either an oracle or an enemy.
  • You can explain every override you make, to yourself and to anyone who asks.
  • Your trust in your own judgment rests on evidence, not on mood.

Related problems

One object, two jobs: a public answer to a real problem, and — the moment you start the chat — Aurelius’s live plan for your version of it.