When errors are discovered in training data, the primary risk is that the model has learned from incorrect or low-quality inputs , which can systematically distort outputs (e.g., misstatements of financial information, incorrect classifications, or erroneous flags). Data quality testing directly targets this issue by assessing completeness, accuracy, consistency, validity, and integrity of the data used to train, validate, and test AI models.
Functional testing (A) focuses on whether the application behaves according to requirements, not specifically on the underlying data quality. User interface testing (C) concerns usability and presentation, not training data. Model validation on benchmark data (D) helps assess performance but may not reveal or correct root # cause issues in the underlying training data if those benchmarks are also flawed or unrepresentative. AAIA content emphasizes that data quality controls are foundational to AI reliability and must be treated as a core operational control.
References:
ISACA, Advanced in AI Audit (AAIA) Exam Content Outline - Domain 2: Data Management Specific to AI (data quality, data balancing, data security).
ISACA AI guidance on data quality controls as the basis of trustworthy AI outcomes.