Odpowiedzi na quiz z wiadomościami na portalu pacjenta opracowane przez sztuczną inteligencję

Odpowiedzi i wyjaśnienia pojawiają się w miarę rozwiązywania quizu.

1. What did the 2024 Stanford study by Garcia and colleagues find about AI-drafted inbox replies?
   • Drafts cut reply time in half
   • Clinicians used drafts for nearly every message
   • Task load and burnout scores fell, but there was no significant time savings
   • The program was shut down for patient harm
2. Why does the Ayers 2023 study not prove AI drafts beat physicians in real clinical inboxes?
   • It compared answers to public questions on Reddit, not messages from patients in a health system
   • It used only nurses as evaluators
   • It measured only spelling errors
   • It tested an image model
3. The UC San Diego study by Tai-Seale and colleagues found which pattern?
   • Shorter replies and faster sending
   • No drop in reply time, longer time reading messages, and longer replies
   • Complete elimination of nurse triage
   • Drafts were sent automatically without review
4. A physician leaves a fluent draft mostly unchanged even though it contains an error. What is this tendency called?
   • Alert fatigue
   • Retrieval failure
   • Disparate impact
   • Automation bias
5. A draft tells a patient to double a blood pressure medication, missing yesterday's high potassium result. What risk does this show?
   • Missing context, because the model sees only what the prompt includes
   • A problem with the patient portal's font
   • The model working too slowly
   • A billing code mismatch
6. Which safeguard sends high-risk messages away from AI drafting?
   • Making replies longer
   • Removing the start-blank option
   • Excluding categories such as test results and possible emergencies from drafting and routing them to triage
   • Sending drafts automatically after five minutes
7. What does California's AB 3030 generally require?
   • A ban on all AI in healthcare
   • Disclosure of AI-generated patient communications unless a licensed provider reviewed them
   • Federal FDA approval for every draft
   • Patients to write their own replies
8. Which monitoring practice best catches drafts that were sent without real review?
   • Counting total portal logins
   • Measuring server response time
   • Tracking the number of patients enrolled
   • Logging how much clinicians edit drafts and reviewing a sample of sent messages with minimal changes