A retail company is evaluating vendors for an AI system that will approve or deny store credit applications. During due diligence, the company learns the training data historically contained far fewer approved applications from one zip code than its population would suggest. Which responsible AI principle is most directly at risk if this is not addressed before deployment?
Select an answer to reveal the explanation.
Short Explanation
The tell here is the phrase 'far fewer approved applications from one zip code than its population would suggest,' which is a classic sign of imbalanced training data that can bake historical bias straight into a model's future decisions. That is squarely about whether the system treats different groups of applicants equitably, which is a distinct concern from whether anyone can explain how the model reached a decision, whether personal data is being kept safe, or whether the system behaves consistently day to day. Explainability matters, but it does not fix a dataset that already reflects skewed history. Data protection matters too, but it is about keeping information secure, not about who the data favors. Consistent performance is about stability, not equity. The concern that matches an underrepresented group in the training data is the one about equitable treatment.
Full Explanation
The correct answer is A. Fairness concerns whether an AI system treats different groups of people equitably, and training a credit-approval model on data that under-represents approvals from a particular zip code risks the model learning and perpetuating that historical imbalance, producing biased outcomes for applicants from that area. Option B is incorrect because transparency is about whether the system's decisions and workings can be explained and understood by users and stakeholders, which is a separate concern from whether the underlying data itself is skewed. Option C is incorrect because privacy and security concern protecting personal data from misuse or unauthorized access, not the statistical composition of a training dataset. Option D is incorrect because reliability and safety concern whether a system performs consistently and predictably under expected conditions, which is a different issue from whether it produces systematically unequal outcomes across demographic groups.