AWS Certified AI PractitionerAI/ML and Generative AI FundamentalsMedium
A healthcare provider is deploying an AI-powered diagnostic tool. During testing, it's discovered that the tool performs significantly worse for certain demographic groups compared to others, even though the overall accuracy seems acceptable. This discrepancy is likely due to which of the following issues?
- AUnderfitting the training data
- BAlgorithmic bias
- COverfitting to the training data
- DInsufficient model complexity
Show answer & explanationAnswer & explanation
Correct answer: B. Algorithmic bias
Algorithmic bias occurs when an AI system's performance is unfairly different across various demographic groups. This often stems from biases in the training data (e.g., underrepresentation of certain groups) or from the algorithm itself, leading to unequal or unfair outcomes for specific populations.
Why the other options are wrong
- A. Underfitting means the model is too simple to capture the underlying patterns, resulting in poor performance on both training and test data overall, not just specific groups.
- C. Overfitting means poor generalization to *all* unseen data, not specifically worse performance for *certain demographic groups*.
- D. Insufficient model complexity (underfitting) would lead to poor performance across the board, not specifically targeting certain groups.
Algorithmic Bias
Algorithmic bias refers to systematic and repeatable errors in a computer system that create unfair or unequal outcomes or predictions, often disproportionately affecting certain demographic groups.
- Can stem from biased training data (e.g., underrepresentation).
- Can arise from the algorithm's design or feature selection.
- Leads to disparities in performance or treatment across groups.
- Crucial ethical consideration in AI development.
Memory trick: Fairness in AI: Different groups, different results = Bias.