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 patients from a specific demographic group, leading to delayed or incorrect diagnoses for them, while performing well for the majority of other patients. Which ethical concern in AI/ML does this situation highlight?
- AData Leakage
- BAlgorithmic Bias
- CCatastrophic Forgetting
- DLack of Explainability
Show answer & explanationAnswer & explanation
Correct answer: B. Algorithmic Bias
Algorithmic bias occurs when an AI system produces unfair or inaccurate outcomes for certain groups of people. In this case, the diagnostic tool performing worse for a specific demographic group is a clear example of algorithmic bias, likely stemming from unrepresentative or biased training data.
Why the other options are wrong
- A. Data Leakage occurs when information from outside the training dataset is used to create the model, leading to an overly optimistic evaluation of the model's performance, but not directly to differential performance across demographics.
- C. Catastrophic Forgetting is a phenomenon where a neural network forgets previously learned information upon learning new information, which is a training stability issue, not an ethical concern about fairness.
- D. Lack of Explainability refers to the inability to understand how an AI model arrived at a particular decision, which is a separate issue from the outcome being unfair.
Algorithmic Bias
A systematic and repeatable error in a computer system's output that creates unfair outcomes, such as preferential or prejudiced results against certain groups of people.
- Often stems from biased or unrepresentative training data.
- Can lead to discriminatory outcomes.
- Identified through rigorous testing across different demographics.
- Mitigation involves fair data collection, bias detection, and debiasing techniques.
Memory trick: AI ethics: fairness, transparency, and harm.