Download A4Q AI and Software Testing Practice Exam and more Exams Technology in PDF only on Docsity!
Question 1. Which of the following best defines artificial intelligence (AI)? A) A set of hardware components that mimic human brain activity B) The study of algorithms that enable machines to perform tasks requiring intelligence C) A programming language specifically designed for robotics D) A database of human knowledge used for computation Answer: B Explanation: AI is the field concerned with creating algorithms that allow computers to perform tasks that normally require human intelligence, such as reasoning, learning, and perception. Question 2. The Turing Test is primarily used to evaluate: A) The speed of a computer processor B) Whether a machine can exhibit behavior indistinguishable from a human C) The accuracy of a statistical model D) The efficiency of a sorting algorithm Answer: B Explanation: Proposed by Alan Turing, the test measures a machine’s ability to generate responses indistinguishable from those of a human interlocutor. Question 3. Which period is known as the “AI winter” due to reduced funding and interest? A) 1950‑ 1956 B) 1960‑ 1970 C) 1980‑ 1990 D) 2000‑ 2010 Answer: C Explanation: The 1980‑1990 era saw disappointment with expert systems and a subsequent decline in AI research investment, termed an “AI winter”.
Question 4. In symbolic AI, propositional logic differs from predicate logic in that: A) Propositional logic can express relationships between objects B) Predicate logic includes quantifiers like ∀ and ∃ C) Propositional logic is used for neural networks D) Predicate logic does not allow variables Answer: B Explanation: Predicate logic extends propositional logic by adding quantifiers and variables to represent relationships among objects. Question 5. Knowledge‑based systems typically rely on: A) Large labeled datasets for training B) A rule base and an inference engine C) Genetic algorithms for optimization D) Convolutional neural networks for image processing Answer: B Explanation: Knowledge‑based systems store expert knowledge as rules and use an inference engine to draw conclusions. Question 6. Which learning paradigm uses feedback in the form of rewards and punishments? A) Supervised learning B) Unsupervised learning C) Reinforcement learning D) Semi‑supervised learning Answer: C Explanation: Reinforcement learning agents learn optimal actions by receiving reward signals from the environment.
Explanation: The “naïve” assumption is that each feature contributes independently to the probability of a class. Question 10. Support Vector Machines (SVM) primarily aim to: A) Minimize the number of support vectors B) Maximize the margin between classes C) Reduce the dimensionality of data D) Perform hierarchical clustering Answer: B Explanation: SVM finds the hyperplane that maximizes the distance (margin) between the nearest points of different classes. Question 11. In K‑means clustering, the value of K represents: A) The number of dimensions in the dataset B) The number of clusters to form C) The maximum number of iterations allowed D) The distance metric used Answer: B Explanation: K is the user‑specified number of clusters the algorithm will partition the data into. Question 12. The perceptron learning rule updates weights only when: A) The output matches the target B) The learning rate is zero C) The perceptron makes an error D) The input vector is all zeros Answer: C Explanation: Weight adjustments occur only on misclassification to reduce future errors.
Question 13. Which of the following is a common source of algorithmic bias in AI systems? A) Using a high‑performance GPU B) Over‑parameterizing a neural network C) Training data that under‑represents certain groups D. Using cross‑validation for model selection Answer: C Explanation: Biased training data leads the model to learn and amplify existing prejudices. Question 14. A “weak test oracle” in AI testing means: A) The expected output is unknown or difficult to define precisely B) The test environment lacks sufficient compute resources C) The test suite contains duplicated test cases D. The system under test is deterministic Answer: A Explanation: In AI, especially ML, the correct answer may not be known beforehand, making oracle construction challenging. Question 15. Metamorphic testing helps address the oracle problem by: A) Comparing outputs of two independent implementations B) Checking that certain input transformations produce predictable output changes C) Randomly generating inputs without checking outputs D. Using only synthetic data for testing Answer: B Explanation: Metamorphic relations define how output should change when inputs are systematically altered, allowing verification without a true oracle.
Question 19. Which testing level is most appropriate for verifying the interaction between a recommendation engine and the user‑profile service? A) Unit testing B) Integration testing C) System testing D. Acceptance testing Answer: B Explanation: Integration testing focuses on interfaces and data flow between components, such as a recommendation engine and profile service. Question 20. A/B testing in AI‑enabled products is primarily used to: A) Compare two different algorithms on the same dataset B) Measure user response to two variants in a live environment C) Validate the correctness of unit tests D. Generate synthetic training data Answer: B Explanation: A/B testing presents two versions to real users and measures differences in behavior or performance. Question 21. Which of the following best describes “self‑healing” test scripts? A) Scripts that automatically rewrite themselves when code changes are detected B) Scripts that use AI to locate UI elements even after layout modifications C) Scripts that execute faster on GPU hardware D. Scripts that generate random inputs for fuzz testing Answer: B Explanation: Self‑healing scripts leverage AI (e.g., computer vision) to adapt to UI changes without manual maintenance.
Question 22. In AI‑driven risk prediction for software testing, the primary input is: A) Historical defect data and code metrics B) Real‑time CPU usage C) User interface screenshots D. Network latency measurements Answer: A Explanation: Risk models use past defect patterns, code complexity, and change history to forecast future testing needs. Question 23. Which of the following is an example of unsupervised learning? A) Predicting house prices from labeled data B) Classifying emails as spam or not spam using labeled examples C) Grouping customers into market segments without pre‑defined labels D. Training a robot to navigate a maze with reward signals Answer: C Explanation: Unsupervised learning discovers hidden structures (e.g., clusters) without explicit labels. Question 24. The term “statistical significance” in AI testing indicates: A) The model’s runtime is below a threshold B) The observed performance difference is unlikely due to random chance C) The model uses a statistically based algorithm D. The dataset size exceeds 1 million records Answer: B Explanation: Statistical significance quantifies confidence that results are not due to random variation.
Explanation: Sample bias occurs when the training data is not representative, leading to skewed model behavior. Question 28. A back‑to‑back testing strategy for AI systems involves: A) Running the same test suite on two different versions of a model to compare outputs B) Executing tests sequentially without resetting the environment C) Using the same test data for training and testing D. Measuring performance on a single dataset only Answer: A Explanation: Back‑to‑back testing compares two implementations (e.g., old vs. new model) under identical conditions. Question 29. Which of the following AI techniques is most suitable for image classification tasks? A) Decision trees B) Convolutional Neural Networks (CNNs) C) Naïve Bayes D. K‑Nearest Neighbors with Euclidean distance only Answer: B Explanation: CNNs are designed to capture spatial hierarchies in images, making them the state‑of‑the‑art for visual classification. Question 30. In a regression problem, the appropriate evaluation metric is: A) Accuracy B) F1‑score C) Mean Squared Error (MSE) D. Confusion matrix
Answer: C Explanation: MSE quantifies the average squared difference between predicted and actual continuous values. Question 31. Which of the following best describes “constraint‑based solving” in symbolic AI? A) Using gradient descent to minimize loss B) Defining a set of logical constraints and searching for solutions that satisfy them C. Training a neural network with back‑propagation D. Randomly sampling the solution space Answer: B Explanation: Constraint solvers operate on declarative constraints, exploring the search space for assignments that meet all conditions. Question 32. The main advantage of using a validation set instead of cross‑validation for hyper‑parameter tuning is: A) Faster computation when the dataset is large B) Higher model accuracy automatically C) Elimination of overfitting completely D. Ability to use the test set for training Answer: A Explanation: Holding out a single validation set avoids the repeated training cycles of k‑fold cross‑validation, reducing computational cost. Question 33. Which AI ethical principle emphasizes that systems should be transparent about how decisions are made? A) Beneficence B) Explainability (or transparency)
C. Only CPU resources, no GPUs D. Absence of any monitoring tools Answer: B Explanation: Realistic AI testing depends on data that mirrors production conditions; quality and representativeness are crucial. Question 37. In component‑level AI test automation, generating unit tests automatically can be achieved by: A) Recording user actions manually B) Using program analysis to infer input‑output specifications C. Randomly selecting code lines to test D. Executing the whole system without isolation Answer: B Explanation: Static or dynamic program analysis can derive contracts and generate corresponding unit tests. Question 38. Which metric best captures the trade‑off between precision and recall? A) Accuracy B) Specificity C) F1‑score D. ROC‑AUC Answer: C Explanation: F1‑score is the harmonic mean of precision and recall, reflecting a balance between them. Question 39. Monkey testing differs from fuzz testing primarily in:
A) Monkey testing uses random clicks on UI, while fuzz testing provides random inputs to APIs or functions B. Monkey testing is deterministic, fuzz testing is not C. Monkey testing requires AI, fuzz testing does not D. Monkey testing only works on mobile apps Answer: A Explanation: Monkey testing simulates random user interactions; fuzz testing supplies random or malformed data to program interfaces. Question 40. Return on Investment (ROI) for an AI‑based testing tool is calculated by: A) (Cost of tool) / (Number of test cases) B) (Savings from reduced manual effort – Tool cost) / Tool cost C. (Number of bugs found) × (Tool price) D. (Tool’s CPU usage) / (Execution time) Answer: B Explanation: ROI measures financial benefit relative to expense; savings from automation minus the tool cost, divided by the cost. Question 41. Which of the following best illustrates “sample bias” in a facial‑recognition model? A) Training only on images of adults of a single ethnicity B) Using a high‑resolution camera for data collection C. Normalizing pixel values before training D. Applying dropout regularization during training Answer: A Explanation: Limiting the training set to a narrow demographic leads the model to perform poorly on under‑represented groups.
Explanation: The ROC curve plots TPR vs. FPR for varying decision thresholds, helping select an operating point. Question 45. In AI‑driven UI test automation, “object identification” typically relies on: A) Hard‑coded pixel coordinates only B) AI models that recognize UI elements based on visual features and DOM attributes C. Manual scripting of each element’s XPath D. Random guessing of element IDs Answer: B Explanation: Modern AI‑enabled tools use computer vision and attribute analysis to robustly locate UI components. Question 46. Which of the following is a characteristic of a well‑designed acceptance criterion for an AI feature? A) It specifies exact numeric thresholds for performance metrics (e.g., precision ≥ 0.90) B. It requires the system to be 100 % accurate on all inputs C. It does not reference any measurable metric D. It only mentions user interface colors Answer: A Explanation: Acceptance criteria for AI should be quantifiable, allowing objective verification (e.g., precision, recall thresholds). Question 47. A model that performs well on the training set but poorly on the validation set is likely suffering from: A) Underfitting B. Overfitting C. Data leakage
D. Concept drift Answer: B Explanation: Overfitting describes high training performance with low generalization to unseen data. Question 48. Which of the following best describes “concept drift” in deployed machine‑learning systems? A) The model’s parameters gradually decay due to hardware wear B. The statistical properties of the input data change over time, reducing model accuracy C. The source code of the model is refactored D. The model is retrained daily with the same data Answer: B Explanation: Concept drift occurs when the underlying data distribution evolves, requiring model updates. Question 49. When evaluating an AI‑enabled chatbot, which metric would directly measure the relevance of its responses? A) Latency B) BLEU score (or similar language‑generation metric) C. CPU utilization D. Disk I/O throughput Answer: B Explanation: BLEU and similar metrics compare generated text to reference responses, assessing relevance and fluency. Question 50. In a test‑automation framework that uses AI to generate test data, “data diversity” is important because: A) It reduces the size of the test suite
Question 53. Which of the following best explains why a confusion matrix is not directly applicable to regression problems? A) Regression outputs continuous values, not discrete classes B. Regression models never make mistakes C. Regression datasets are always larger than classification datasets D. Regression uses only precision as a metric Answer: A Explanation: A confusion matrix categorizes predictions into true/false positives/negatives, which requires discrete class labels. Question 54. A “reinforcement learning” agent for game playing receives a reward of +1 for winning and – 1 for losing. Which of the following learning objectives does it optimize? A) Minimizing cross‑entropy loss B. Maximizing expected cumulative reward over episodes C. Reducing mean squared error D. Maximizing precision Answer: B Explanation: Reinforcement learning seeks policies that maximize the expected sum of rewards (return). Question 55. Which AI technique is most appropriate for generating natural language summaries from large documents? A) Decision trees B) Recurrent Neural Networks (RNNs) or Transformers C. K‑means clustering D. Naïve Bayes classification Answer: B
Explanation: Sequence models like RNNs and Transformer‑based architectures excel at language generation tasks. Question 56. In AI testing, “oracle poisoning” refers to: A) Corrupting the test data used to validate model outputs, leading to misleading pass/fail results B. Adding extra CPU cores to speed up testing C. Using a faulty random number generator D. Encrypting test logs for security Answer: A Explanation: Oracle poisoning manipulates the expected outcomes (oracle) to hide defects or mislead testers. Question 57. Which of the following is a common method to mitigate overfitting in neural networks? A) Increasing the number of hidden layers indefinitely B. Applying dropout regularization during training C. Removing the bias term from each neuron D. Using only a single training epoch Answer: B Explanation: Dropout randomly disables neurons during training, preventing co‑adaptation and reducing overfitting. Question 58. In a system that uses AI for automated defect triage, the primary input to the model is: A) The source code repository URL B. Historical defect reports, severity, and associated metadata C. The size of the compiled binary
