Human review for important decisions

Unit ID: AIT-M11-U03 Estimated active time: 15-20 minutes

The main idea

Some outputs need human or expert review before use.

Not because the output is likely to be wrong, but because the consequence of it being wrong falls on someone who did not choose to rely on a tool.

Why this matters

The categories are predictable: anything affecting rights, money, health, education, employment or safety. Deciding them in advance is what makes the rule usable.

Safe example

AI can help prepare questions for a policy review, but it should not make the policy decision.

Preparation is assisted. The decision is not, and the line between them is worth writing down.

Better working habit

Define when manager, expert, legal, financial, medical, HR, or academic review is required.

A named category and a named reviewer. "Someone should look at this" means nobody will.

What to watch for

Do not let AI make decisions that affect rights, money, health, education, jobs, or safety.

In these areas the tool can help you prepare, gather and draft. It cannot hold the accountability, and neither can a document it produced.

Mini practice

Create human-review triggers for five task types.

For each, name the reviewer — an actual role, not a department.

Check yourself

Before moving on, answer:

• Does this affect rights, money, health, education, jobs or safety?
• Who is the named reviewer, and have they agreed?
• Am I asking the tool to prepare or to decide?