AI in Science Quiz

Answers and explanations appear as you take the quiz.

1. Before selecting a scientific AI method, what should a researcher define?
   • The scientific question and what observation would disconfirm the claim
   • Only the model architecture
   • The desired publication date
   • A result that cannot be challenged
2. Why might a random train/test split overstate generalization when readings are correlated?
   • The test set becomes larger
   • Data from the same subject, instrument, or experiment can appear across splits
   • Random splits remove all uncertainty
   • The model cannot train on numbers
3. What should a researcher record to support reproducibility?
   • Only the model’s final prediction
   • Only the paper title
   • Preprocessing, model versions, relevant random seeds, and evaluation material
   • The number of plots generated
4. What is required before treating an AI-generated hypothesis as a supported scientific claim?
   • As soon as the model explains it confidently
   • Whenever it matches a familiar theory
   • Only when the model’s score is high
   • After appropriate experiments or independent analysis support it
5. A model performs well on random readings from instrument A but poorly on instrument B. What should the team do?
   • Report both results and investigate calibration or measurement differences
   • Report only the better score
   • Claim general performance based on the random split
   • Remove instrument B from the record
6. Why separate what is measured from what is inferred?
   • It removes the need for replication
   • It clarifies which parts are observations and which are model conclusions
   • It guarantees a causal explanation
   • It makes missing data irrelevant
7. What should a researcher understand about calibration, confidence intervals, and prediction intervals?
   • They are interchangeable scores
   • Use only the largest value
   • They answer different questions
   • Replace them with the model name
8. What should be preserved from source data through a published figure?
   • Only a screenshot of the chart
   • The model’s general description
   • A list of software logos
   • The path from source data to figure, table, or manuscript