引導式測驗 · 中等 等級
臉部辨識與錯誤逮捕測驗
測試人臉搜尋錯誤逮捕風險:候選人名單、錯誤匹配、NIST 人口統計結果、羅伯特威廉斯、獨立證據和陣容
問題 1 的 8
A one-to-many face search returns a ranked candidate. What does the ranking mean?
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 one-to-many face search returns a ranked candidate. What does the ranking mean?
- The candidate committed the offense
- The image resembles a gallery candidate under the system’s comparison, not that the person is identified or guilty
- The candidate was present at the crime scene
- The system has established probable cause
- What did NIST’s demographic-effects testing show about face-recognition algorithms?
- Every algorithm has identical error rates across groups
- Laboratory results predict each police search exactly
- Face matching never produces false positives
- Error rates and demographic effects varied across algorithms and testing conditions
- In Robert Williams’s Detroit case, why was the face-search result not sufficient to support an arrest under the settlement safeguards?
- The result is an investigative lead and requires independent investigative steps and supervisory review
- The algorithm ranked him first, so no additional evidence was needed
- The settlement prohibits every police department from searching faces
- A face search can determine guilt only if the image is high resolution
- Why should a witness not be told which lineup image ranked highest in a face search?
- The ranking would improve the witness’s memory
- The witness is required to repeat the software output
- That information could influence the witness and make an identification procedure suggestive
- A lineup must show the candidate first
- Which follow-up is genuinely independent of the same face-search result?
- Running the same photo through a second vendor without other inquiry
- Evidence gathered through a separate investigative path, such as verified records or physical evidence
- Showing the candidate’s image repeatedly to the witness
- Increasing the match threshold after seeing the candidate
- Why does gallery size affect one-to-many search risk?
- Searching more images changes the opportunity for a false candidate to appear above a threshold
- A larger gallery proves the true person is included
- Gallery size eliminates image-quality problems
- The system’s rank becomes a court decision
- Which search records best support later auditing?
- Only the final arrest report
- The suspect’s name and the vendor logo
- A confidence score without the searched image
- Probe image, gallery and threshold details, ranked results, software version, and operator notes
- A candidate match is followed by a witness choosing the same person after being shown the match. What concern arises?
- The second step always validates the algorithm
- The witness’s choice proves the probe image was clear
- The witness may have been influenced, so the procedure may not provide independent confirmation
- The result eliminates the need to investigate further