Artificial Intelligence: Concepts, Systems and Lifecycle
Domain 1 - Artificial Intelligence. The exam tests whether you can use ISO/IEC 22989 vocabulary precisely and reason about the AI system lifecycle, not just recite model types.
Scope and vocabulary
An AI system, per ISO/IEC 22989, is an engineered system that generates outputs such as predictions, content, recommendations or decisions for a given set of objectives. It includes data, models, inference logic and the surrounding human and technical context. Distinguish AI from ML: ML is a subset in which models learn patterns from data; rule-based expert systems can be AI without learning. Know the lifecycle: problem framing, data acquisition and preparation, model development, verification and validation, deployment, operation and monitoring, and retirement.
Core rules
- Training, validation and test data must be separated; test data must never influence model selection.
- Verification asks whether the system was built correctly; validation asks whether it meets intended use and stakeholder needs.
- Metrics are task-specific: precision/recall for detection, RMSE for regression, calibration for probabilistic outputs.
- Accuracy alone is inadequate under class imbalance or unequal error costs.
- Document intended use, assumptions and limitations; monitor data drift and concept drift.
Worked example
A supervised model scores loan applications. Accuracy is 97%, but only 3% of applicants default. A model that approves everyone achieves 97% accuracy while missing every default. Reasoning: select recall for defaulters, precision of approvals and a cost-weighted metric; set the threshold from risk appetite; validate on a held-out, time-separated sample. Metric selection follows decision context, not the algorithm.
Traps
Equating AI with deep learning; assuming high accuracy means fitness for purpose; tuning on test data; treating explainability and accuracy as interchangeable; forgetting that deployed models degrade as populations change.