AI & общество Викторина

Отговорите и обясненията се появяват, докато отговаряте на теста.

1. Why consider people who do not directly use an AI system?
   • Only direct users can experience impact
   • Institutions can change their access, information, or work even for non-users
   • Non-users have no rights or interests
   • System effects stop at the login screen
2. A public service is faster on average but inaccessible to a language group. What does this suggest?
   • Average speed proves fair access
   • Language coverage is separate from system impact
   • The service has no social consequence
   • Efficiency does not capture who can access or contest the service
3. When should affected communities be consulted?
   • Only after deployment ends
   • Only if they operate the system
   • When defining the problem and evaluating the proposed solution
   • Only when the model is open source
4. What can generated content do in an information environment?
   • Increase access to explanations while also amplifying misleading material
   • Guarantee accurate public information
   • Remove the need for provenance
   • Ensure all readers see corrections
5. What helps readers evaluate a generated claim without blind trust?
   • A model’s confidence phrase
   • The number of generated pages
   • A polished voice
   • Provenance, corrections, and transparent evidence
6. Why should governance continue after deployment?
   • All social impacts are known before launch
   • Actual outcomes, complaints, and changed uses can reveal new impacts
   • Model accuracy prevents new effects
   • Governance is only procurement paperwork
7. Does technical accuracy on a test set establish that a system is socially suitable?
   • No; access, contestability, and institutional fit also matter
   • Yes, if accuracy exceeds a benchmark
   • Only if users agree with the answer
   • No, because accuracy cannot be measured
8. Who shapes the distribution of AI benefits and harms?
   • Model weights alone
   • Only individual users
   • Institutions and policy choices as well as technical capabilities
   • The benchmark designer alone