引導式測驗 · 中等 等級
人工智慧在社會安全殘障索賠測驗的應用
在 SSA 殘障索賠中測試 AI:決策支援、證據摘要、來源驗證、快速判定、錯誤審核和索賠人審查。
問題 1 的 8
According to SSA’s public descriptions, how should AI summaries support disability decisions?
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。
- According to SSA’s public descriptions, how should AI summaries support disability decisions?
- As a final determination of disability without a reviewer
- As a replacement for medical records
- As a guarantee that a claimant qualifies
- As decision support that helps adjudicators review evidence, not as a substitute for the individual file
- An AI summary lists a treatment date that conflicts with the medical record. What should the adjudicator do?
- Use the summary because it is shorter
- Check the source record and correct the summary or underlying file before relying on it
- Ignore the source record if the model is confident
- Deny the claim because the file is inconsistent
- What can a medical-record summarizer fail to preserve?
- Important limitations, chronology, attribution, or evidence that contradicts a conclusion
- Only the document’s file size
- The claimant’s legal right to appeal
- The program’s statutory definition
- Which measure is important alongside processing time?
- 模型參數數量
- Length of the generated summary
- Supported facts, omissions, attribution errors, and downstream decision or appeal patterns
- Vendor response latency alone
- Why should summary claims link to source pages?
- To make the summary legally binding
- To prevent a claimant from submitting more evidence
- To avoid preserving the original file
- So reviewers can verify statements and locate omitted or contradictory evidence
- A quick-disability tool flags a case as likely to qualify. What does that flag mean?
- The claimant is automatically awarded benefits
- The model has established every medical fact
- The case may fit a faster review path; eligibility still follows applicable program rules
- No adjudicator needs to inspect the file
- Which test set best checks an evidence-summary tool?
- Only clean typed records from the model’s training data
- Reviewed cases with scanned, unusual, and contradictory records representative of actual files
- Only cases that were allowed
- Examples selected solely by the vendor
- What does a claimant need if an AI summary misstates a fact?
- A usable way to identify the error and submit the correct source evidence
- Access only to the model’s source code
- A requirement to accept the summary
- A new prediction without record correction