Failure to override AI decisions

AI features in Flinn are intended to support — not replace — the user. If a user cannot override an AI decision, or has to fight the interface to do so, the result is continued reliance on a possibly-wrong AI conclusion. The risk grows when the AI's confidence is high but its output is incorrect.

Hazardous situation: Users continue to rely on incorrect AI outputs because the interface does not provide a clear way to override or correct them.

• Override is always available. AI suggestions in Flinn — screening recommendations, paper recommendations, similar-report detection — are always editable or dismissible by the user. See AI Screening Support, AI Paper recommendation and AI Detection of Similar incident reports.
• Traceable reasoning. AI-generated text is annotated with reference numbers so users can verify the underlying source and override individual claims; see AI Clinical Report Writing.
• Manual additions and deletions. Users can add, edit or remove results manually; see How can I delete a manually-added result? and the broader extraction workflow in the AI Extraction Guide.
• Audited overrides. Overrides are recorded in the Audit Trail, so the rationale for departing from an AI suggestion is preserved.
• Related guidance. See Misinterpretation of data for input-side guidance and Data exposure related to AI Training for the broader context around AI behaviour.

If you encounter a workflow where an override is missing or unclear, please Report a problem or a bug.