引導式測驗 · 中等 等級
臨床人工智慧測驗中的自動化偏差
測試遺漏與委託的錯誤、乳房X光檢查和小插圖試驗的研究證據以及保持臨床醫生獨立的習慣。
問題 1 的 8
A radiologist misses a subtle asymmetry because the AI placed no mark there, though she would have caught it on her own. Which type of automation bias error is this?
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。
- A radiologist misses a subtle asymmetry because the AI placed no mark there, though she would have caught it on her own. Which type of automation bias error is this?
- An error of commission
- Algorithm aversion
- Alert fatigue
- An error of omission
- What did the 2023 Radiology study find when radiologists received deliberately incorrect AI-labeled suggestions on mammograms?
- Accuracy was unchanged for every reader
- Accuracy fell, and less experienced readers were affected most
- Only very experienced readers were affected
- Readers ignored every suggestion
- According to the guide's summary of a 2023 JAMA vignette study, what happened when explanations were added to a systematically biased model?
- Explanations did little to offset the drop in diagnostic accuracy
- Explanations fully eliminated the bias
- Clinicians refused to read the explanations
- Accuracy exceeded the no-AI baseline
- After weeks of false sepsis alerts, a physician dismisses all of them, including a correct one. What does this illustrate?
- An error of commission caused by over-trust
- Sequential reading
- Under-trust, seen in hospitals as alert fatigue
- Well-calibrated reliance
- Which workflow design does the guide say best preserves independent judgment?
- Showing the AI's answer before the clinician looks at the case
- Hiding all confidence scores
- Disabling overrides
- Sequential reading, where the clinician forms a judgment before the AI output is revealed
- 某個部門發現臨床醫師幾乎百分之百採納 AI 工具的建議。指南建議該如何解讀這種現象?
- 這證明該工具非常完美
- 這可能代表臨床醫師只是在「蓋章」敷衍了事,有必要針對採納的案例進行抽樣稽核
- 這意味著臨床醫師不信任該工具
- 這不具備任何安全性意義
- 為什麼指南說「人在迴路中」(human in the loop)並不能保證安全性?
- 如果人類可靠地依賴系統,那麼這個迴路幾乎無法提供額外的保護
- 人類的準確度總是低於 AI
- 法規禁止人類進行審查
- 人類無法看到 AI 的輸出結果
- 指南建議採取哪種個人習慣,以保持對 AI 建議的獨立判斷?
- 將沒有警示視為一切正常的證明
- 不經審查就接受高信賴度的輸出結果
- 在查看 AI 輸出結果之前,先寫下自己的假設
- 依賴解釋來決定輸出結果是否正確