Download AI Testin Practice Exam and more Exams Technology in PDF only on Docsity!
Question 1. Which term describes the phenomenon where AI systems appear to perform better as they become more integrated into everyday tasks, making their capabilities seem invisible? A) AI Winter B) AI Effect C) Automation Paradox D) Black‑Box Phenomenon Answer: B Explanation: The AI Effect refers to the tendency to consider AI achievements as “normal” once they become commonplace, thus diminishing the perception of their intelligence. Question 2. A system that can perform any intellectual task a human can, with the ability to improve itself autonomously, is known as: A) Narrow AI B) General AI C) Super AI D) Reactive AI Answer: C Explanation: Super AI (or Artificial Superintelligence) surpasses human intelligence across all domains and can self‑improve beyond human capabilities. Question 3. Which of the following best differentiates a rule‑based system from an AI‑based system? A) Use of neural networks for decision making B) Dependence on handcrafted if‑else logic C) Ability to learn from data D) Both B and C Answer: D
Explanation: Rule‑based systems rely on explicit if‑else logic, whereas AI‑based systems learn patterns from data; both aspects differentiate them. Question 4. Which AI technology is primarily used to handle uncertainty by representing degrees of truth rather than binary true/false values? A) Genetic Algorithms B) Fuzzy Logic C) Support Vector Machines D) Decision Trees Answer: B Explanation: Fuzzy Logic models reasoning with partial truth values, allowing handling of uncertainty and vagueness. Question 5. Which framework is most commonly associated with dynamic computational graphs, facilitating rapid prototyping in deep learning? A) TensorFlow (static graph) B) PyTorch C) Scikit‑learn D) Caffe Answer: B Explanation: PyTorch offers dynamic computation graphs, enabling intuitive model building and debugging. Question 6. For training large transformer models, which hardware accelerator provides the highest throughput per watt? A) CPU B) GPU C) TPU
C) AI Act D) HIPAA Answer: C Explanation: The EU AI Act proposes a risk‑based regulatory framework for AI systems, focusing on high‑risk applications. Question 10. In AI systems, flexibility primarily refers to: A) The ability to run on any operating system B) The capacity to adapt to new data or environments without extensive re‑engineering C) The speed of inference on edge devices D) The number of programming languages supported Answer: B Explanation: Flexibility denotes how easily an AI system can incorporate new data, tasks, or contexts. Question 11. Autonomy in an AI system is best defined as: A) The system’s ability to operate without any human oversight at all times B) The degree to which the system can make decisions within predefined operational bounds C) The speed at which the system processes inputs D) The number of layers in its neural network Answer: B Explanation: Autonomy concerns decision‑making capability within specified limits, not unlimited independence. Question 12. Managing evolution for self‑learning systems involves: A) Freezing the model after deployment B) Continuous monitoring, validation, and updating of the model as data drifts
C) Removing all feedback loops D) Disabling online learning features Answer: B Explanation: Evolution management ensures the model remains accurate over time by handling concept drift and updating accordingly. Question 13. Algorithmic bias most commonly arises due to: A) Faulty hardware components B) Inherent assumptions in the learning algorithm that amplify existing data patterns C) Random number generation errors D) Overly large training datasets Answer: B Explanation: Algorithmic bias stems from how the algorithm processes data, potentially magnifying existing inequities. Question 14. Which technique is used to detect bias in a classification model by comparing performance across demographic groups? A) Cross‑validation B) Confusion matrix decomposition by group C) Gradient checking D) Hyperparameter tuning Answer: B Explanation: Evaluating metrics (e.g., false‑positive rate) per demographic reveals disparate impact, helping detect bias. Question 15. An ethical principle that requires AI developers to disclose how data is collected and used is:
Question 18. Which characteristic makes AI testing in safety‑critical domains particularly challenging? A) Deterministic behavior of models B) Closed‑form mathematical proofs of correctness C) Non‑deterministic outputs and learning from new data D. Absence of any external regulations Answer: C Explanation: Safety‑critical AI systems often exhibit non‑deterministic behavior, making verification and validation difficult. Question 19. In reinforcement learning, the term “policy” refers to: A) The loss function used for training B) The mapping from states to actions that the agent follows C) The dataset used for supervised pre‑training D. The reward discount factor Answer: B Explanation: A policy defines the agent’s action selection strategy based on observed states. Question 20. Which of the following is a key factor when selecting a machine‑learning algorithm for a problem? A) The color of the developer’s IDE theme B) The nature of the target variable (categorical vs. continuous) C) The brand of the GPU used for training D) The number of coffee breaks taken during development Answer: B Explanation: The type of prediction (classification vs. regression) heavily influences algorithm choice.
Question 21. The standard stages of an ML workflow, in correct order, are: A) Deployment → Training → Validation → Data collection → Testing B) Data preparation → Training → Validation → Testing → Deployment C) Testing → Training → Deployment → Validation → Data preparation D) Validation → Training → Deployment → Data preparation → Testing Answer: B Explanation: The typical pipeline starts with data preparation, then model training, validation, testing, and finally deployment. Question 22. Overfitting in a machine‑learning model is best described as: A) The model performing equally well on training and unseen data B) The model capturing noise from the training data, leading to poor generalization C) The model having too few parameters to represent the data D) The model being unable to converge during training Answer: B Explanation: Overfitting occurs when a model learns spurious patterns (noise) from the training set, harming performance on new data. Question 23. Which dataset split is primarily used to fine‑tune hyperparameters without contaminating the final performance estimate? A) Training set B) Validation set C) Test set D) Production set Answer: B
Explanation: Precision measures the proportion of positive predictions that are correct (True Positives / (True Positives + False Positives)). Question 27. Which metric is most appropriate for evaluating imbalanced binary classification problems? A) Accuracy B) Recall C) F1‑Score D) Mean Squared Error Answer: C Explanation: F1‑Score balances precision and recall, making it useful when class distribution is skewed. Question 28. The ROC‑AUC metric evaluates: A) The probability that a randomly chosen positive instance ranks higher than a randomly chosen negative instance B) The average absolute error of predictions C) The percentage of correctly clustered points D) The speed of model inference Answer: A Explanation: ROC‑AUC measures the model’s ability to discriminate between classes across all thresholds. Question 29. For regression tasks, the coefficient of determination (R²) indicates: A) The proportion of variance in the dependent variable explained by the model B) The average absolute difference between predictions and actual values C) The number of hidden layers in the neural network
D) The probability of overfitting Answer: A Explanation: R² quantifies how much of the total variation is captured by the regression model. Question 30. The Silhouette Score is used to assess: A) Classification accuracy B) Clustering cohesion and separation C) Regression residuals D) Neural network weight sparsity Answer: B Explanation: Silhouette Score measures how similar an object is to its own cluster compared to other clusters. Question 31. Benchmark suites for ML performance, such as MLPerf, primarily help with: A) Generating synthetic training data B) Providing standardized tests to compare hardware and software efficiency C) Deploying models to mobile devices automatically D) Visualizing model architectures Answer: B Explanation: MLPerf offers a set of standardized workloads to evaluate and compare performance across platforms. Question 32. In a feed‑forward neural network, a “neuron” performs which mathematical operation? A) Convolution of input with a kernel B) Weighted sum of inputs followed by an activation function C) Sorting of input vectors
D) Uniform test case generation Answer: B Explanation: AI specifications often involve probabilistic outcomes, making exact requirement statements difficult. Question 36. Which test level focuses on verifying that the model’s predictions meet functional performance criteria? A) Input Data Testing B) ML Model Testing C) Component Testing D) Acceptance Testing Answer: B Explanation: ML Model Testing evaluates the model itself, checking metrics like accuracy, precision, etc. Question 37. Concept drift refers to: A) The gradual increase in model size over time B) Changes in the statistical properties of the input data distribution after deployment C) The shrinking of GPU memory during training D) The shift from supervised to unsupervised learning Answer: B Explanation: Concept drift occurs when the underlying data generating process evolves, potentially degrading model performance. Question 38. To test for automation bias in a decision‑support AI system, a tester should: A) Disable the AI and compare outcomes with human decisions B) Increase the number of hidden layers in the model
C) Reduce the dataset size by half D) Use only synthetic data for validation Answer: A Explanation: Comparing AI‑assisted decisions with unaided human decisions helps reveal over‑reliance on automation (automation bias). Question 39. When documenting an AI component for testing, which artifact is essential for defining test objectives? A) Source code comments only B) Model card or datasheet describing intended use, performance, and limitations C) License agreement of the development framework D) CPU temperature logs Answer: B Explanation: Model cards provide transparent documentation of model purpose, metrics, and constraints, guiding test planning. Question 40. A “back‑to‑back” test for an AI system typically involves: A) Running the same test case on two different hardware platforms simultaneously B) Comparing the AI system’s output against a trusted legacy system or reference model for the same inputs C) Executing tests in reverse chronological order D) Using adversarial examples only Answer: B Explanation: Back‑to‑back testing contrasts outputs from the new AI system with those of a known, reliable system. Question 41. In A/B testing of a new recommendation algorithm, the primary metric to evaluate is:
Question 44. Which of the following is a typical requirement for a test environment that supports AI model training at scale? A) Single‑core CPU only B) Distributed storage and compute resources with GPU/TPU support C. Manual data entry interfaces D) No network connectivity Answer: B Explanation: Large‑scale AI training demands distributed compute (GPUs/TPUs) and high‑throughput storage. Question 45. Virtual test environments for autonomous vehicles commonly simulate: A) Only static road maps B) Sensor streams (LiDAR, camera, radar) and dynamic traffic participants C. Textual user interfaces D. Email communication protocols Answer: B Explanation: Autonomous vehicle testing requires realistic sensor data and dynamic scenarios to evaluate perception and control algorithms. Question 46. Which AI technology is most suitable for automating the generation of test cases from natural language requirements? A) Computer Vision B) Reinforcement Learning C) Natural Language Processing (NLP) D) Genetic Algorithms Answer: C
Explanation: NLP can parse textual requirements and generate structured test cases automatically. Question 47. AI‑driven test case prioritization typically relies on: A) Random ordering of test cases B) Predictive models estimating the fault‑detection probability of each test case C) Manual ranking by developers D) Sorting by file size Answer: B Explanation: Predictive analytics estimate which tests are most likely to uncover defects, enabling efficient prioritization. Question 48. Defect prediction using AI often uses which type of input data? A) Source code metrics (e.g., cyclomatic complexity, churn) and historical defect logs B. Number of coffee cups consumed by developers C. Length of commit messages only D. Operating system version only Answer: A Explanation: Code metrics and past defect information serve as features for machine‑learning models that predict future defects. Question 49. In UI testing, computer‑vision‑based AI can assist by: A) Generating HTML code from sketches B) Detecting visual regressions via pixel‑wise comparison and object recognition C. Compiling source code faster D. Encrypting UI assets Answer: B
Explanation: Validation data guides hyperparameter selection and provides an early estimate of generalization performance. Question 53. Which loss function is most appropriate for binary classification problems? A) Mean Squared Error B) Hinge Loss C) Binary Cross‑Entropy (Log Loss) D) Kullback‑Leibler Divergence Answer: C Explanation: Binary Cross‑Entropy measures the difference between predicted probabilities and actual binary labels. Question 54. When evaluating a regression model, a high RMSE relative to the target variable’s range suggests: A) Excellent model performance B) The model’s predictions deviate substantially from actual values C. That the model is over‑parameterized D. That the dataset is perfectly linear Answer: B Explanation: Root Mean Squared Error quantifies average prediction error; a large RMSE indicates poor accuracy. Question 55. In clustering evaluation, a high Silhouette Score close to 1 indicates: A) Overlapping clusters B) Well‑separated, cohesive clusters C. Random assignment of points D. Presence of outliers only
Answer: B Explanation: Silhouette values near 1 mean points are close to their own cluster and far from neighboring clusters. Question 56. Which of the following is a key advantage of using genetic algorithms in AI testing? A) Guarantees optimal test suite size B) Ability to explore large search spaces and evolve test cases towards higher coverage or fault detection C. Requires no fitness function definition D. Works only on linear models Answer: B Explanation: Genetic algorithms iteratively improve a population of test cases based on a fitness metric, suitable for complex test generation problems. Question 57. Pairwise testing applied to AI input parameters primarily aims to: A) Reduce the number of test cases while still covering all possible pairs of parameter values B. Test every possible combination exhaustively C. Randomly select test inputs without regard to coverage D. Validate only the first parameter in the list Answer: A Explanation: Pairwise testing ensures that every possible pair of parameter values appears in at least one test case, balancing coverage and effort. Question 58. An adversarial example in computer‑vision AI is created by: A) Adding large, visible noise to the image B. Subtly perturbing pixel values so that the model misclassifies while the change is imperceptible to humans
