The feedback loop can reinforce historical bias: What is the clearest concern?
Decisions shape future records, so reduced observations may reinforce the historical under-recording rather than correct it.
The question
A customer-service model recommends faster escalation for customers whose earlier complaints were historically under-recorded. Those recommendations produce fewer later records for that group, and the model then learns from the reduced records. What is the clearest concern?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Human review removes the monitoring needHuman involvement may help, but it does not eliminate automation bias or the need to monitor outcomes and feedback effects.
- The feedback loop can reinforce historical bias ✓Model-driven decisions affect future data, potentially reproducing earlier under-recording and strengthening biased patterns over time.
- The model has improved because records decreasedFewer records may reflect reduced visibility caused by the system, rather than better service or fewer underlying problems.
- The training data became automatically representativeNew observations generated through biased decisions do not automatically correct historical underrepresentation or establish representativeness.
The trap
Trace how outputs change future inputs before treating improved-looking statistics as evidence of fairness. How to remember it
Decisions shape future records, so reduced observations may reinforce the historical under-recording rather than correct it.
How many of these would you get right?
One of 1581 AIGP questions on Certsqill. Take a free five-minute check and see your score per domain — not one number, but which section to open tonight.
Test your AIGP readiness — freeMore Understanding How to Govern AI Development questions
- Record the event as a near miss: What should the evidence support? →
- Document monitoring: What governance action is best supported? →
- Trace dependencies: What evidence is needed before decommissioning under the organization’s risk-management →
- All 426 Understanding How to Govern AI Development questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Development ·
Every answer, right and wrong, comes with its own explanation.