引導式測驗 · 中等 等級
監獄和懲戒中的人工智慧測驗
在糾正中測試人工智慧:特定用途的風險、監控限制、敏感通訊、驗證、錯誤審查、人力權威和監督。
問題 1 的 8
An AI risk score is high for an incarcerated person. What does the score establish by itself?
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。
- An AI risk score is high for an incarcerated person. What does the score establish by itself?
- That a disciplinary violation happened
- That the person must be moved to a restrictive unit
- Only that the person’s data met the model’s scoring pattern, not that misconduct will occur
- That the score is calibrated for every facility
- Why is a maintenance-ticket prioritizer different from a tool affecting housing placement?
- Housing can affect liberty, safety, and services, so its errors carry different consequences
- Maintenance systems never make errors
- Housing decisions require no human review
- Both uses have identical impact
- A call-screening tool flags a conversation as threatening. Which step is essential?
- Use the flag as conclusive evidence
- Publish the audio to solicit opinions
- Expand monitoring to all communications without policy review
- Review the original conversation and context before treating it as a threat
- What should an agency evaluate before deploying an AI monitoring system?
- Only the number of calls processed
- False alarms, missed incidents, language variation, and impact on sensitive communications
- Only vendor response time
- Whether the user interface is attractive
- Why could historical incident data bias a correctional risk model?
- Historical data contain no human decisions
- A larger dataset removes all policy effects
- Records reflect staff observation and institutional practices as well as underlying behavior
- Risk models use only biological data
- Which safeguard is important when screening communications?
- Treat every message as available for any purpose
- Let the vendor decide which conversations are privileged
- Delete audit logs after each flag
- Protect privileged or confidential material through clear rules and restricted handling
- A staff reviewer can see only the score, not its source records. What is missing?
- Enough context to verify the flag and make an independent decision
- A higher score threshold
- More historical data
- A second copy of the same score
- How should a pilot be judged before expansion?
- Count how many alerts it generates
- Compare results and harms with the existing process using independent review
- Assume the system works if staff like it
- Expand first and collect error reports later