Section 69 of 102
68. Uncertainty and Confidence Management
Stable section ID: S05-CON-010-SECTION-69 · 26 content blocks
System05 shall explicitly represent uncertainty in observations, estimates, diagnoses, predictions, and recommendations.
Uncertainty may arise from:
- Measurement error.
- Sensor degradation.
- Missing information.
- Time delay.
- Conflicting sources.
- Incomplete configuration.
- Model limitations.
- Environmental variability.
- Human-report uncertainty.
- Communication loss.
- Unknown physical changes.
- AI inference.
- Future prediction.
Confidence shall be associated with a defined claim and shall not be expressed as a general property of the agent. High confidence in detecting an abnormal condition does not necessarily imply high confidence in its cause or recommended correction.
Where practical, the system should identify:
- The uncertain value or conclusion.
- The source of uncertainty.
- The estimated confidence or range.
- The decision affected.
- The consequence of error.
- The evidence required to reduce uncertainty.
- The validity period of the estimate.
Low-confidence information may still justify precautionary action when the potential consequence is severe. Conversely, high model confidence shall not authorize an action that exceeds the agent’s permission or professional scope.
Uncertainty shall be communicated in language appropriate to the user. The system shall avoid both false precision and vague warnings that provide no basis for responsible action.