AI in Insurance Quiz

Answers and explanations appear as you take the quiz.

1. Why is a predictive score not a complete underwriting decision?
   • It does not by itself establish fair use, lawful relevance, explanation, or review
   • A score cannot rank records
   • Every insurer uses identical data
   • Only human decisions can be measured
2. A claims model reduces payments but appeals rise. What should the insurer investigate?
   • Whether the model’s name changed
   • Whether more images were stored
   • Whether the benchmark score improved
   • Whether wrongful denials or delays increased, not just the payment total
3. What should be checked before using third-party or credit-related features?
   • Only the feature’s predictive strength
   • Whether the field is numeric
   • Permitted use, provenance, and whether the feature acts as an irrelevant or protected proxy
   • If the vendor provides a dashboard
4. Why monitor appeals, overrides, and complaints after deployment?
   • They prove the score is accurate
   • They can reveal errors or harms not visible in model metrics
   • They replace group outcome analysis
   • They are unrelated to customer outcomes
5. What does the NAIC Model Bulletin imply about insurer AI-supported decisions?
   • The bulletin approves all insurer models
   • A model removes the need for governance
   • All state requirements are identical
   • They remain subject to applicable insurance laws, including unfair discrimination rules
6. Which record helps explain an AI-supported claim outcome?
   • Model version, vendor data, reasons, human review, and corrective action
   • Only the final payment amount
   • The model’s parameter count
   • A copy of the customer’s unrelated policies
7. Why evaluate across relevant claim conditions and groups?
   • A single average guarantees fairness
   • Errors can vary by claimant or claim context despite acceptable overall accuracy
   • Different conditions cannot affect model error
   • Group checks replace case review
8. What should a policyholder be able to do where required?
   • Change the model’s training data directly
   • Access other customers’ claims
   • Ask questions about and challenge an outcome through an appropriate path
   • Approve the insurer’s vendor