AI-900 Practice Question: Describe Artificial Intelligence workloads and considerations

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  When a model's predictions consistently favour one output class due to class imbalance

  Why it's wrong here

  Class imbalance means the training data has a skewed distribution of outcome labels, so the model's accuracy-driven optimization pushes it to always predict the majority class; this is a data distribution and loss-function problem, not a fairness problem. AI bias specifically involves systematic, unfair treatment of demographic groups, which may be absent even with extreme class imbalance if the majority class does not correspond to a protected group. Conversely, a balanced dataset can still produce biased models if proxy features or historical stereotypes encode discrimination.

• ✓

  Systematic unfair outcomes for demographic groups caused by biased training data or design choices

  Why this is correct

  This is the standard technical definition of AI bias: unfair algorithmic outcomes that flow from biased training data, proxy variables, or model design choices, and that disproportionately harm protected demographic groups. Historical bias (e.g., past hiring decisions) gets encoded into the model, reinforcing and automating existing inequities in high-stakes areas like lending, hiring, or criminal justice. Fairness auditing typically measures subgroup-level metrics such as demographic parity, equalized odds, or calibration to detect and correct these disparities.

• ✗

  When an AI model performs worse on unseen test data than on the training data

  Why it's wrong here

  A model performing worse on unseen data than on training data describes overfitting, a generalization error caused when the model memorizes noise instead of learning transferable patterns; it is corrected by regularization, cross-validation, or more diverse data. Overfitting concerns an overall performance drop on new data, not differential outcomes for specific demographic groups. AI bias, in contrast, can persist even when a model has equal overall accuracy because the errors are unequally distributed across groups—a fair-looking aggregate can hide biased subgroup behavior.

• ✗

  The tendency of users to trust AI recommendations over their own judgment

  Why it's wrong here

  This statement describes automation bias, a human cognitive tendency to over-trust algorithm outputs, which is a behavioral psychology issue about how people interact with AI systems. AI bias is an objective property of the model's decisions themselves, not the user's perception or reliance. A technically biased model may be distrusted by users, and a fair model can be over-relied upon; the two are independent failure modes that require different mitigation strategies—user interface design and model audit, respectively.

Go deeper

Related to this question