AI-Based System Testing Fundamentals CT-AI Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   In the specification of an AI-based system, what does the 'data requirement' typically describe?

   • AThe cost of storing the data in the cloud
   • BThe number of users who will access the data
   • CThe characteristics of the data needed for training and operation, such as format, volume, and quality
   • DThe programming language used to implement the data pipeline
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A team is testing an AI system for autonomous vehicles. They have a test dataset of 10,000 images, but only 100 images contain pedestrians at night. What is the best way to ensure the test set adequately covers this critical scenario?

   • AOversample the nighttime pedestrian images by duplicating them in the test set.
   • BAdd synthetic images of nighttime pedestrians to the test set to increase coverage.
   • CRemove the nighttime pedestrian images from the test set to avoid bias.
   • DUse the test set as is, because it reflects the real-world distribution of scenarios.
   Reveal Solution Discussion Select an answer first
4. 23application · easy

   A team is testing an AI system for medical diagnosis. The test data was collected from a single hospital and includes only patients from that region. What is the main limitation of this test data?

   • AIt is not labeled correctly.
   • BIt contains too many missing values.
   • CIt is too small to provide statistically significant results.
   • DIt may not be representative of the broader patient population, limiting the generalizability of test results.
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   An AI system for predicting customer churn uses a dataset with many missing values for the 'income' feature. The team decides to impute missing values with the mean income. What is a potential risk of this approach?

   • AIt may reduce the variance of the income feature, affecting the model's ability to distinguish customers.
   • BIt will increase the accuracy of the model.
   • CIt will make the test data more representative of the training data.
   • DIt will eliminate the need for testing the model's robustness.
   Reveal Solution Discussion Select an answer first
6. 25application · medium

   A fintech startup is developing a machine learning model to approve small personal loans. The model will be deployed in a regulatory environment where decisions must be explainable and auditable. The team has limited data science resources and needs to launch within three months. Which testing approach is most appropriate?

   • ADefer testing until after deployment, then monitor live performance and adjust the model based on real-world outcomes.
   • BPrioritize testing for bias, robustness, and explainability, using a test set that reflects the expected applicant distribution, and document results for audit.
   • CFocus primarily on model accuracy using a held-out test set, and rely on the model's built-in feature importance for explainability.
   • DUse a simple rule-based system instead of ML to avoid the need for extensive testing.
   Reveal Solution Discussion Select an answer first