Bias in AI typically results in unfair outcomes due to which root cause?

• A. Random hardware faults
• B. Flawed data and algorithms
• C. User preferences
• D. Market volatility

Answer: (B)

Bias in AI that leads to unfair outcomes mainly comes from flawed data and the algorithms that learn from it. When training data reflect historical discrimination, are unrepresentative of certain groups, or contain mislabeled or noisy records, the model picks up those patterns and reproduces them in its decisions. The way the model is trained—what it optimizes for and what signals it uses—can also amplify unfairness if fairness isn’t built into the objective. For example, if a dataset underrepresents a group, the model may have lower accuracy for that group; if variables act as proxies for protected attributes, the model can make biased predictions even without any explicit use of sensitive data; and feedback loops can make bias self-reinforcing as predictions influence future data and outcomes.

By contrast, random hardware faults introduce random, unpredictable errors rather than consistent biases across cases. User preferences may shape individual interactions but don’t establish a systemic, structural cause of unfairness. Market volatility affects timing and costs rather than the underlying fairness of the model’s decisions.