NCA-AIIO - Essential AI Knowledge (38% of the exam) - Section 1.5
Explain key AI use cases and industries.
Describe representative AI use cases across industries such as healthcare imaging analysis, financial fraud detection, autonomous vehicles, and natural language processing for customer service. Recognise which use cases are compute-intensive enough to require GPU-accelerated infrastructure rather than CPU-only deployments.
Practice question for this objective
lock_openFree sampleEssential AI Knowledgeeasy
A retail bank's fraud team notices that rule-based systems miss novel attack patterns as soon as fraudsters adjust their behaviour. Which AI capability is best suited to detecting previously unseen fraud without requiring manual rule updates?
AA machine learning model that learns normal transaction patterns and flags statistical outlierscheck_circle Correct
BA collaborative filtering recommender system trained on purchase history
CA large language model that summarises customer support tickets
DA graph database that stores historical transaction records for audit queries
Match real-time fraud detection in financial services to machine learning anomaly detection rather than static rule engines or unrelated AI capabilities. Machine learning fraud detection systems are trained on labelled transaction data to distinguish legitimate from fraudulent activity. Because the model learns statistical decision boundaries rather than hard-coded rules, it generalises to novel fraud patterns automatically as it is retrained, giving it a significant advantage over brittle rule-based approaches that require continuous manual maintenance.
Why A is correct: Supervised and unsupervised ML models build a statistical baseline of legitimate behaviour and score deviations in real time, catching novel fraud patterns that no explicit rule anticipates.
Why B is wrong: Collaborative filtering identifies similarities between users to suggest products; it is designed for recommendation, not anomaly detection in transaction streams.
Why C is wrong: LLM summarisation addresses unstructured text workloads and has no mechanism for scoring numerical transaction features against a learned fraud baseline.
Why D is wrong: A graph database is a storage and query technology, not an AI model that learns patterns; it cannot adaptively score new transactions for fraud likelihood.
Answers that look right on this material and are not. Each one is a distractor from a different question in the NCA-AIIO bank for this domain.
Computer vision, which extracts structured information from product photographs and store shelf images
errorWhy it is wrong: Tempting because retailers widely use computer vision, but it processes images, whereas the assistant must understand and generate written language.
Classical ML models cannot process structured tabular data and require unstructured inputs such as images or text
errorWhy it is wrong: This is the opposite of reality. Classical ML algorithms such as gradient-boosted trees and logistic regression are well suited to structured tabular data and have historically dominated that format in competitions and production.
To continue training the weights so that they gradually adapt to the new integer representation
errorWhy it is wrong: Adapting weights to low precision is the idea behind quantisation-aware training, a different technique; the described step runs samples without updating any weights, so no training occurs.