Learning objectives
After this lesson, you should be able to
- Classify low- and high-consequence AI tasks
- Require citations and uncertainty for regulated guidance
- Design approval gates for external actions
Capability is not authority
AI may summarize, classify, draft, and recommend without being authorized to contact, negotiate, book, pay, alter regulated records, or clear compliance exceptions.
- Ground answers in versioned approved sources
- Display uncertainty and jurisdiction
- Require approval for consequential external actions
- Log input, source, output, reviewer, and outcome
Learning needs governance
User feedback should create a review signal, not silently rewrite policies. Published guidance changes only through versioned editorial and expert review.
Apply the decision protocol
Use a fictionalized or fully permissioned operating scenario. Build five columns: observed facts, supplied facts, assumptions, controlling sources, and unresolved questions. Do not advance a consequential action while a required fact, authorization, qualification, or safety condition remains unresolved.
- Demonstrate: Classify low- and high-consequence AI tasks
- Demonstrate: Require citations and uncertainty for regulated guidance
- Demonstrate: Design approval gates for external actions
- Name the decision owner, evidence standard, stop condition, and next review time
Practice with evidence
Create a one-page decision record and ask a peer to challenge the source, version, applicability, missing facts, incentives, and proposed communication. Revise the record rather than defending the first answer. Preserve the initial and corrected versions so an editor can see what the exercise actually taught.
- Cite every externally verifiable claim
- Separate uncertainty from error
- Escalate beyond the lesson's stated competence boundary
- Remove private, proprietary, or personally identifiable information
Correct and transfer the learning
After the scenario, compare the decision to the current source and the stated objective. Record the misconception, the evidence that corrected it, the operational control that would prevent recurrence, and the conditions that would require the answer to be researched again.
Knowledge check
Should repeated user feedback automatically rewrite a compliance rule?
Reveal the answer
No.
Feedback should enter a governed review process with source verification and accountable approval.