An automated loan approval model consistently rejects applications from a particular demographic group, despite those individuals having high credit scores and meeting all financial criteria. What ethical issue in artificial intelligence does this scenario illustrate?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Pay close attention here, because ethics in AI is a massive topic on the exam and in the real world. If you build a model and it starts treating one group of people differently from another based on things like race, gender, or zip code, you've got a serious algorithmic bias problem on your hands. This usually happens because the training data itself is biased, or because the model picked up on proxy variables that mimic demographic data. It's not a security breach, and it's not a question of explainability—it's a fundamental fairness and bias issue that can land your company in hot water. Trust me, you need to actively audit your datasets to prevent this!
Full explanation below image
Full Explanation
Algorithmic bias occurs when a machine learning system reflects the implicit values, prejudices, or systemic inequities of the humans who created it or the historical data used to train it. In this scenario, the systematic denial of loans to a specific demographic group—despite their strong credit profiles—points to biased outcomes. This type of bias can stem from unrepresentative training datasets, historical prejudice embedded in past lending decisions, or the inclusion of variables that act as proxies for demographic identifiers (such as postal codes or educational institutions). Mitigating algorithmic bias requires rigorous fairness testing, data preprocessing, and bias mitigation algorithms during training.
Let's examine the incorrect options: - Explainability of the model (Option A) refers to the ability to understand and interpret how a model arrives at its decisions (e.g., using SHAP or LIME values). While an unexplainable model makes detecting bias difficult, the issue of unfair outcomes itself is classified as algorithmic bias. - Data security (Option B) involves protecting the model and the underlying databases from unauthorized access, cyber threats, or malicious modifications. - Loss of data confidentiality (Option D) focuses on ensuring that sensitive personal data is kept private and not disclosed to unauthorized parties, which is not the issue when the model produces biased predictions.