Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

AI ni Idanwo ti Awujọ

Tests public accountability, context of use, challenge routes, accessibility, agency responsibility, and service continuity.

Jẹmọ itọsọna awọn ọnaAI Public Sector
Ibeere 1 ti 8

A public service uses a model to prioritize benefit cases. What must not happen?

Awotẹlẹ gbogbo ibeere8 ibeere

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

  1. A public service uses a model to prioritize benefit cases. What must not happen?
    • Staff document who reviews recommendations
    • A proxy score silently determines a person’s access to the service
    • The agency tests accessibility
    • The service measures incomplete applications
  2. Why are a triage model and an eligibility recommendation different contexts of use?
    • Both always make final decisions
    • Triage systems cannot affect people
    • Eligibility is only a data-format question
    • They can have different consequences and require different evidence and oversight
  3. Which feature gives a person a meaningful remedy for an incorrect outcome?
    • A generic explanation that names no factors
    • An internal benchmark dashboard
    • A clear route to challenge or correct it
    • A longer processing queue
  4. Why test accessibility and relevant populations before deployment?
    • A system may produce unequal error or access burdens not visible in an overall score
    • Accessibility changes model size only
    • An average error rate proves equal access
    • Public services need no user testing
  5. A pilot performs well on a benchmark. What additional check is needed before operational use?
    • Treat the benchmark as deployment authorization
    • Skip staff training
    • Assume vendors carry agency responsibility
    • Review procurement, security, records, privacy, and applicable authority independently
  6. What should an explanation identify when a person asks why a decision occurred?
    • A generic statement that AI was involved
    • The actual factors used in the specific decision
    • The vendor’s company history
    • Only the overall model accuracy
  7. Who remains responsible when an agency uses a vendor model?
    • The agency for the complete workflow, including updates and staff practice
    • The vendor alone
    • The system’s training dataset
    • The person affected by the result
  8. What should a public-service plan include for system failure?
    • A rule to deny every pending case
    • An unreviewed substitute score
    • A fallback route for people and staff when the system is unavailable
    • A hidden outage message