What issue arises when training data is biased, potentially amplifying social inequities?

• A. Training data issues
• B. Data governance
• C. Lagging indicators
• D. Cyberattacks on AI

Answer: (A)

When training data reflect social biases, the model learns those patterns and tends to repeat or even amplify them in its outputs. Because machine learning models pick up on whatever patterns exist in the data, biased representations lead to biased predictions and decisions, which can worsen inequities across different groups. That’s why this is described as training data issues—the root problem lies in the data used to train the model and its representativeness, quality, and potential biases.

Data governance is about policies and controls for managing data, which can help address bias, but it’s not the immediate issue described. Lagging indicators are just retrospective measurements, not the bias problem in the training data. Cyberattacks on AI refer to security threats, not to the fairness or bias concerns raised by biased data.