AWS Certified AI PractitionerAI/ML and Generative AI FundamentalsHard
A data scientist is investigating a trained AI model that predicts loan default risk. They discover that the model consistently assigns higher default probabilities to applicants from a specific zip code, even when other financial indicators are similar to applicants from other areas. This behavior leads to a disproportionately high rejection rate for individuals from that zip code. Which key AI/ML concept does this scenario most directly illustrate?
- AAlgorithmic Bias
- BOverfitting
- CFeature Scaling
- DUnderfitting
Show answer & explanationAnswer & explanation
Correct answer: A. Algorithmic Bias
This scenario describes a situation where the model's predictions are systematically unfair or discriminatory towards a specific demographic group (applicants from a specific zip code), leading to a disproportionate outcome. This is a direct manifestation of algorithmic bias, where the model has learned and amplified biases present in the training data or introduced during model development.
Why the other options are wrong
- B. Overfitting occurs when a model performs well on training data but poorly on unseen data, not necessarily due to systematic unfairness towards a specific group.
- C. Feature scaling is a data preprocessing technique to normalize the range of independent variables. While important, it doesn't directly address or describe the phenomenon of discriminatory outcomes based on group attributes.
- D. Underfitting occurs when a model is too simple to capture the underlying patterns in the data, resulting in poor performance on both training and test data, which is not the issue described here.
Algorithmic Bias
Algorithmic bias refers to systematic and repeatable errors or unfairness in a computer system's outcome, leading to discriminatory results against certain groups. It often stems from biased training data, flawed model design, or inappropriate use.
- Systematic unfairness in outcomes.
- Discriminates against specific groups.
- Caused by data, design, or usage.
Memory trick: Bad models lead to unfair results.