Which statement best differentiates AI-specific requirements from traditional software requirements in CPMAI?

Which statement best differentiates AI-specific requirements from traditional software requirements in CPMAI?

• A. Focus on functional features and user interface rather than data governance.
• B. Include data requirements, model performance targets, explainability, safety constraints, data governance, and monitoring needs.
• C. Emphasize hardware requirements and deployment timelines only.
• D. Data requirements and performance targets while ignoring explainability and safety.

Answer: (B)

In CPMAI, AI-specific requirements must capture how the system uses data and how the model behaves, not just what it does for users. This means clearly defining data needs and quality, provenance and privacy considerations, labeling and data governance, and how data will be managed over time. It also means setting model performance targets—metrics like accuracy, calibration, robustness, and risk-based thresholds that reflect the domain—and specifying how the model should behave under different conditions. Explainability is essential so stakeholders can understand and trust the decisions the AI makes, whether through interpretable designs or documented explanations. Safety constraints outline risk controls, fail-safes, and escalation paths to prevent harm. Data governance covers lineage, access controls, retention, and regulatory compliance. Finally, monitoring needs ensure ongoing evaluation of data quality and model performance, drift detection, and automated triggers for retraining or intervention. Taken together, this mix of data requirements, model metrics, explainability, safety, governance, and monitoring differentiates AI-focused requirements from traditional software requirements, which typically center on features and user interface rather than data-centric and lifecycle considerations.

In CPMAI, AI-specific requirements must capture how the system uses data and how the model behaves, not just what it does for users. This means clearly defining data needs and quality, provenance and privacy considerations, labeling and data governance, and how data will be managed over time. It also means setting model performance targets—metrics like accuracy, calibration, robustness, and risk-based thresholds that reflect the domain—and specifying how the model should behave under different conditions. Explainability is essential so stakeholders can understand and trust the decisions the AI makes, whether through interpretable designs or documented explanations. Safety constraints outline risk controls, fail-safes, and escalation paths to prevent harm. Data governance covers lineage, access controls, retention, and regulatory compliance. Finally, monitoring needs ensure ongoing evaluation of data quality and model performance, drift detection, and automated triggers for retraining or intervention. Taken together, this mix of data requirements, model metrics, explainability, safety, governance, and monitoring differentiates AI-focused requirements from traditional software requirements, which typically center on features and user interface rather than data-centric and lifecycle considerations.