AI 의사결정 퀴즈

퀴즈를 풀면 정답과 설명이 나타납니다.

1. Which sequence separates the stages of a decision?
   • Observed evidence, model estimate, action rule, accountable decision-maker
   • Model name, vendor, interface color, release date
   • Training data, model output, automatic approval in every case
   • Policy statement, confidence score, and no review
2. Why is a model’s most likely prediction not automatically the best action?
   • Predictions cannot be measured
   • A prediction is not a policy; goals, constraints, and consequences also matter
   • Every action must maximize model confidence
   • A human cannot use predictions
3. When should a decision workflow evaluate both false alarms and missed events?
   • Only when both have identical costs
   • Only after removing all human review
   • When either error can produce consequences that affect the action policy
   • Never; overall accuracy captures every consequence
4. What can influence a threshold for acting on a model score?
   • Only the color of the dashboard
   • The number of people who built the model
   • The model’s marketing category
   • The consequences of errors, available capacity, and score reliability
5. Why might a workflow defer some cases for review?
   • It can provide an alternative when consequences, capacity, or score reliability make immediate action unsuitable
   • Deferral guarantees a correct result
   • It removes accountability
   • It proves the prediction is causal
6. What can happen if future training data records outcomes only for cases a system selected?
   • The system automatically becomes causal
   • Those data may reflect past system choices rather than the full population
   • The model can no longer produce scores
   • All unselected cases become confirmed errors
7. What should accompany consequential decisions according to the guide?
   • A generic confidence statement alone
   • An unexplained automated outcome
   • Expert oversight, evidence-grounded explanations, and a way to correct mistakes
   • A guarantee that the model never errs
8. What should a team test before relying on a consequential workflow?
   • Only the model in isolation
   • Only whether the interface opens
   • A different task with the same vendor
   • The complete workflow under the conditions where it will be used