Exam 312-41 Topic 4 Question 77 Discussion

The scenario clearly identifies that the model is functioning correctly from a mathematical and implementation standpoint, meaning the algorithm itself is not the source of bias. Instead, the bias originates from the choice of input variables used by the model.
The engineering team intentionally introduced new variables such as hardware brand and application timestamp. While these features are technically accurate, they act as proxy variables for socioeconomic status, indirectly encoding sensitive or protected characteristics. This leads to biased outcomes even though the model is technically correct.
This is a classic example of bias introduced during feature selection, which is the stage where decisions are made about which inputs the model will use. In CAIPM governance frameworks, feature selection is a critical control point because:
Features can unintentionally encode protected attributes or proxies
Bias can emerge even when data is accurate and algorithms are correct
Ethical risks often arise from what is included, not just how it is processed Other options are less appropriate:
Algorithm is functioning as intended and not introducing bias
Training data is not explicitly identified as biased in this scenario
User interaction is not relevant to model training or design
CAIPM emphasizes that responsible AI requires careful scrutiny of feature engineering decisions to prevent proxy discrimination and unintended bias.
Therefore, the correct answer is Feature Selection, as bias was introduced through the inclusion of problematic proxy variables.
=========