What is a major concern regarding AI bias?

• A. AI can perpetuate or amplify biases present in training data
• B. AI bias can be fully eliminated by more data
• C. Bias is irrelevant to AI outcomes
• D. Bias only affects performance metrics, not fairness

Answer: (A)

Bias in AI happens when a model learns from data that already encode prejudices or systemic inequalities. The major concern is that the model will reproduce these patterns in its outputs and can even magnify them as it tries to optimize for what it observed. Because the model maps inputs to outcomes based on historical data, biased patterns get reinforced, leading to unfair or discriminatory results for certain groups.

Adding more data does not automatically fix this. If the extra data reflect the same biases, or introduce new ones, the problem can persist or worsen. Bias also matters beyond just raw performance metrics; it affects fairness, trust, and real-world decisions that rely on the AI.

So the focus is on understanding and mitigating how data, representations, and models encode and propagate bias, not merely on collecting more information.