This is not a replacement for the official study guide. Please read the official study guide first — use this exam prep only to reinforce or practice your understanding.
Free AAISM exam practice questions with answers and explanations, organized by domain and part.
Domain 3: AI Technologies and Controls
Q531 An image recognition system must detect objects such as defects on a production line. Which neural network type is MOST commonly used?
- A. Naive Bayes text classifier
- B. K-means clustering routine
- C. Convolutional neural network
- D. Linear regression model
Answer: C
Convolutional neural networks are designed to detect spatial patterns in images. Linear regression, Bayes text classifiers and clustering do not model spatial features as effectively.
Domain 3: AI Technologies and Controls
Q532 Which architecture MOST commonly underlies modern large language models?
- A. Transformer architecture using attention mechanisms
- B. Single-layer perceptron with a fixed threshold
- C. Rule-based expert system with if-then logic
- D. Hidden Markov model for speech states
Answer: A
Transformers use attention to model relationships across a sequence and underpin most modern LLMs. Perceptrons, expert systems and HMMs predate or lack this capability.
Domain 3: AI Technologies and Controls
Q533 A model memorizes its training data and performs poorly on new data. Which problem does this BEST describe?
- A. Overfitting
- B. Underfitting
- C. Concept encryption
- D. Data drift
Answer: A
Overfitting occurs when a model fits training data, including noise, too closely and fails to generalize. Underfitting is failure to capture patterns even in training data, and drift is a change in data distributions after deployment.
Domain 3: AI Technologies and Controls
Q534 A model performs poorly on both training and new data because it is too simple. Which problem does this BEST describe?
- A. Underfitting
- B. Overfitting
- C. Model inversion
- D. Data poisoning
Answer: A
Underfitting occurs when the model cannot capture the underlying patterns, producing poor results on both training and new data. Overfitting performs well on training data only.
Domain 3: AI Technologies and Controls
Q535 Which metric is MOST relevant when the cost of missing a true fraud case is much higher than flagging a legitimate one?
- A. Recall
- B. Latency
- C. Throughput
- D. Uptime
Answer: A
Recall measures the share of actual positives found and matters when missed positives are costly. Throughput, latency and uptime measure system performance rather than detection quality.
Domain 3: AI Technologies and Controls
Q536 A spam filter flags too many legitimate emails. Which metric has declined MOST directly?
- A. Precision
- B. Recall
- C. Availability
- D. Latency
Answer: A
Precision measures how many flagged items are truly positive, so many false positives reduce it. Recall relates to missed positives, and availability and latency are system measures.
Domain 3: AI Technologies and Controls
Q537 A company wants an AI model to create synthetic customer records for testing without exposing real individuals. Which consideration is MOST important?
- A. Verify the synthetic data does not leak or allow re-identification of real records
- B. Use the largest possible dataset regardless of its source
- C. Assume synthetic data carries no privacy risk because it is artificial
- D. Share the output publicly to confirm its realism
Answer: A
Synthetic data can still reveal real records if the generator memorizes them, so re-identification risk must be tested. Assuming it is risk-free, ignoring source quality or publishing results are unsafe.
Domain 3: AI Technologies and Controls
Q538 A team stores training data and model artifacts in a shared cloud bucket. Which control is MOST important for the storage design?
- A. Encryption, access control and versioning with monitoring of access
- B. Monitoring of access logs while keeping data in plain text
- C. Versioning of artifacts with broad read access for all employees
- D. Encryption and access control for the bucket, without versioning of model artifacts
Answer: A
Encryption, least-privilege access, versioning and access monitoring protect data and model artifacts. Public access, shared credentials and disabled logs create exposure and remove accountability.
Domain 3: AI Technologies and Controls
Q539 Which practice BEST ensures reproducibility and rollback capability for AI models in production?
- A. Keeping only the latest model on a developer laptop for quick use
- B. Overwriting the production model with each new training run to save space
- C. Documenting changes only after incidents have occurred in production
- D. Version control for code, data, models and configuration with documented releases
Answer: D
Versioning code, data, models and configuration supports reproducibility, audit and rollback. Overwriting artifacts or local-only copies prevents either.
Domain 3: AI Technologies and Controls
Q540 Before an AI model replaces a legacy decision process, which regulatory consideration is MOST important?
- A. Whether legal and societal requirements such as explainability and non-discrimination can be met
- B. Whether the interface of the new tool is visually appealing to staff
- C. Whether the vendor offers a volume discount for the licence term
- D. Whether the model uses the newest algorithm available on the market
Answer: A
Regulatory and societal impact must be assessed, including explainability and fairness duties. Novelty, discounts and design do not address obligations.