引導式測驗 · 中等 等級

公平清洗與誤導性人工智慧解釋測驗

發現超越證據的公平主張,區分解釋與公平衡量,並狹義地描述法律解釋義務。

問題 1 的 8

What makes a fairness statement an example of fairwashing?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. What makes a fairness statement an example of fairwashing?
    • It reports uncertainty clearly
    • It offers reassurance stronger than the supporting evidence
    • It names the metric and sample
    • It invites review
  2. Does an explanation of a model output prove the decision is fair?
    • Yes, if it uses a chart
    • Yes, for every population
    • No, because explanations are never useful
    • No; explanation and fairness are distinct qualities
  3. Which details help readers assess a fairness claim?
    • Population, data, metric, subgroup results, uncertainty, and limitations
    • A broad claim of neutrality
    • Only the vendor name
    • Only aggregate accuracy
  4. What might a global feature-importance summary fail to show?
    • The average importance ranking it reports
    • That an analysis was performed
    • Group error differences or why one individual received a result
    • Any model input
  5. If available data are too limited to evaluate one group, what should an organization report?
    • That no risk exists
    • That equal performance is proven
    • Nothing, because a model is neutral by default
    • That the evaluation is limited or unavailable
  6. What does AI Act Article 86 provide?
    • A technical explanation for every AI output
    • A conditional explanation right for qualifying decisions within its statutory scope
    • A general right to source code
    • A fairness certification
  7. Why repeat fairness measurement after a material update?
    • Fairness can be measured only once
    • Updates always improve every subgroup
    • Model behavior, data, and deployment conditions may change
    • The Act sets one permanent score
  8. Which public statement is most evidence-bounded?
    • These metrics were observed on this test population; this subgroup was too small to assess reliably
    • Our model is unbiased, with no supporting detail
    • Explainable means fair
    • No complaints prove no harm