Generative AI at Work / Unit 6 GAI 141 worked example, 60 to 90 seconds
A Three-Level Review Rule
The idea
Matching the depth of review to the stakes is a course rule built from NIST's framework and the M-CaReM-AI method, and it has three levels.
In practice
Step one: rate the consequence of an error in the output: low, moderate or high. High means it touches safety, money, legal obligations or customer trust, the areas the method singles out. Step two: low-consequence output gets a light check by the person using it. Step three: moderate-consequence output gets a full claim-by-claim check, and an expert review where the domain is specialist. Step four: high-consequence output is a draft for a named person who makes the decision; the AI never decides. NIST's framework supports this last level: human intervention is needed where a system cannot detect or correct its own errors, and serious risks call for the most urgent attention.
Where people slip
Ratings creep downward under time pressure. Decide each task's level in advance and write it down, so the level is not chosen on a busy afternoon.
Your place on the trail
Every lesson rests on earlier ones and opens later ones. Tap any to jump.
Something wrong in this lesson? Flag it · Sources for Generative AI at Work