Last updated:
The ongoing verification of AI outputs and processes against benchmarks and test cases to catch defects, bias incidents, or policy violations.
A continuous practice where sample outputs are tested against known ground-truth cases, policy-violation detectors, and fairness checks. QC pipelines run in parallel to production, flag anomalies (e.g., unexpected error patterns), and feed issues into defect-tracking systems. Governance defines QC sampling rates, validation criteria, and remediation workflows to keep system integrity high.
Real world example:
A content-moderation AI’s Quality Control system samples 1% of flagged posts daily and compares them to human-moderator judgments. Discrepancies trigger immediate retraining if error rates exceed 5%, ensuring ongoing alignment between automated and human standards.
