引导式测验 · 中等 等级

人工智能在金融领域的测验

将财务人工智能视为决策工作流程：错误成本、审计跟踪、真实的不利行为解释以及对后续行动的确认。

相关引导路径人工智能金融
问题 1 的 8

A model returns a risk score of 0.72. What should the team remember before treating it as a decision?

预览每个问题8 问题

当您参加测验时，就会出现答案和解释。

  1. A model returns a risk score of 0.72. What should the team remember before treating it as a decision?
    • The score is separate from policy, human review, and any final action
    • The score automatically denies the application
    • The score proves an explanation is legally sufficient
    • The score removes the need to inspect data
  2. What costs should a team define when evaluating a financial decision model?
    • Only compute cost
    • Consequences of false positives and false negatives
    • Only average accuracy
    • Only the number of features
  3. Which items belong in the guide’s audit trail?
    • 只看最终成绩
    • Only customer names
    • Data, features, model version, threshold, and human action
    • The model’s marketing description
  4. What should an adverse-action explanation for a credit decision identify?
    • A generic feature list invented after the fact
    • Only the model’s risk score
    • A general statement that the model was complex
    • Actual factors used by the decision process
  5. Why can historical decisions be a poor target label for a financial model?
    • They may encode past selection and not be an appropriate target
    • They always contain too few rows
    • They are automatically more fair than new data
    • They make auditing impossible
  6. What should a team monitor after deploying a fraud-flagging model?
    • 仅模型的名称
    • Drift, missing values, unusual behavior, review queues, and complaints
    • Only the number of servers
    • Only the average risk score
  7. What safeguard should apply to a proposed account transfer or account change?
    • Allow the score to act without limits
    • Delete the audit trail
    • Require confirmation and verify the resulting state
    • Use a generic explanation afterward
  8. How does a fraud flag differ from declining a credit application?
    • They are equivalent because both use a score
    • The fraud flag requires no human review
    • A credit decline is only a prediction
    • A flag can send a transaction for investigation; a decline is a consequential decision