Błąd automatyzacji w quizie dotyczącym sztucznej inteligencji klinicznej

Odpowiedzi i wyjaśnienia pojawiają się w miarę rozwiązywania quizu.

1. Radiolog przegapia subtelną asymetrię, ponieważ AI nie umieściło tam oznaczenia, choć sam by ją zauważył. Jaki to rodzaj błędu automatyzacji?
   • An error of commission
   • Algorithm aversion
   • Alert fatigue
   • An error of omission
2. Co wykazało badanie z 2023 roku w czasopiśmie Radiology, gdy radiolodzy otrzymywali celowo błędne sugestie oznaczone przez AI na mammogramach?
   • Accuracy was unchanged for every reader
   • Accuracy fell, and less experienced readers were affected most
   • Only very experienced readers were affected
   • Readers ignored every suggestion
3. According to the guide's summary of a 2023 JAMA vignette study, what happened when explanations were added to a systematically biased model?
   • Explanations did little to offset the drop in diagnostic accuracy
   • Explanations fully eliminated the bias
   • Clinicians refused to read the explanations
   • Accuracy exceeded the no-AI baseline
4. After weeks of false sepsis alerts, a physician dismisses all of them, including a correct one. What does this illustrate?
   • An error of commission caused by over-trust
   • Sequential reading
   • Under-trust, seen in hospitals as alert fatigue
   • Well-calibrated reliance
5. Which workflow design does the guide say best preserves independent judgment?
   • Showing the AI's answer before the clinician looks at the case
   • Hiding all confidence scores
   • Disabling overrides
   • Sequential reading, where the clinician forms a judgment before the AI output is revealed
6. A department finds that clinicians accept nearly 100 percent of an AI tool's suggestions. How does the guide suggest interpreting this?
   • It proves the tool is perfect
   • It may signal rubber-stamping and warrants auditing a sample of accepted cases
   • It means clinicians distrust the tool
   • It has no safety meaning
7. Why does the guide say that having a human in the loop does not guarantee safety?
   • If the human reliably defers to the system, the loop adds little protection
   • Humans are always less accurate than AI
   • Regulations prohibit human review
   • Humans cannot see AI outputs
8. Which personal habit does the guide recommend for staying independent of AI suggestions?
   • Treat the absence of an alert as proof that nothing is wrong
   • Accept high-confidence outputs without review
   • Write down your own hypothesis before looking at the AI output
   • Rely on explanations to decide whether the output is right