Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ
Xu hướng tự động hóa: Tin tưởng vào máy móc quá nhiều Câu đố
Xác định các lỗi thiên vị tự động hóa, phân biệt sự tin cậy với sự phụ thuộc và chọn các biện pháp bảo vệ cho các quyết định được AI hỗ trợ bằng các bước xác minh thực tế.
Câu hỏi 1 của 8
A reviewer follows an AI recommendation despite a conflicting detail in the source document. Which error pattern is this?
Xem trước mọi câu hỏi8 câu hỏi
Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.
- A reviewer follows an AI recommendation despite a conflicting detail in the source document. Which error pattern is this?
- A commission error from following an incorrect recommendation
- An omission error because the system had no display
- A privacy breach caused by metadata
- A calibration score from model evaluation
- A decision aid fails to flag a required field, and an operator overlooks it. Which error pattern fits?
- A commission error from following correct advice
- A prompt-framing effect
- An omission error associated with missed information
- A benchmark contamination problem
- Participants report similar trust in two systems but rely on one more when it is accurate. What does this illustrate?
- Trust and use are always identical
- Self-reported trust and observed dependence are distinct measures
- Automation cannot change human decisions
- Accurate systems should not be monitored
- Which interface design could raise overreliance risk?
- Displaying source evidence and allowing an override
- Showing which information is missing
- Making high-impact actions require review
- Showing a recommendation before the user can inspect the supporting evidence
- Why is “be careful with AI” an incomplete safeguard?
- It gives no concrete way to detect errors or check the underlying evidence
- It proves the tool is unreliable
- It prevents users from overriding the system
- It replaces workload analysis
- A company reviews only cases where workers rejected the AI suggestion. What evidence could it miss?
- Only correctly accepted suggestions
- No evidence, because rejection logs are complete
- Cases where workers accepted an incorrect suggestion
- The system’s interface color
- Which safeguard is most appropriate when an AI recommendation affects a consequential decision?
- An automatic action with no appeal path
- A named person checks the original evidence and can override the recommendation
- A confidence badge without supporting information
- A requirement to follow the top-ranked result
- Why can average system accuracy be insufficient for evaluating safe use?
- It guarantees equal performance on every case
- It measures human attention directly
- It makes a review process unnecessary
- It does not show whether errors occur in the specific decision context or whether users catch them