Generative AI at Work / Unit 5 GAI 122 review, 2 to 3 minutes
A Summary, Verified
The idea
The M-CaReM-AI method teaches with a composite case set in a hospital building a new incident-reporting system. Its scenes are illustrative, not a real event, and two of them show verification working.
In practice
In the first scene, an analyst asks AI to draft a summary of the incident record from the hospital's policies and notes. The draft is fluent and well organized, and it includes an invented risk-score field that appears nowhere in the source material. The analyst lists the draft's fields against the policy and deletes it. In the second scene, the team asks AI to draft the rules for which incidents must be reported to outside regulators. The draft states a clean rule: all severe incidents are reportable. It sounds right, and it matches what generic documentation often says. Validation with the hospital's own quality and risk staff shows the real rule is different: reportability depends on the incident type, the level of harm, and which version of the policy applies. The method's comment on the scene is that a confident, invented rule like that one could have had serious compliance consequences. In both scenes the error was caught the same way. Someone compared the output with the organization's actual policies, and people who knew the domain reviewed it before anything was built on it.
Where people slip
Neither error would have been caught by reading for quality. The invented field looked like good design, and the over-general rule looked like good practice. They were caught by checking claims against the real source and by expert review. The method's own conclusion from the case is that AI sped up the drafting but often got policy details wrong, and that it was no substitute for the team's domain judgment.
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