AI Bias
Systematic unfairness in AI outputs that advantages some groups and disadvantages others — often because the training data reflected historical biases, which the model absorbed and sometimes amplified. Examples: an AI that gives worse resume feedback to names that sound female, or recommends different loan amounts based on neighbourhood demographics. Bias is hard to eliminate entirely because training data reflects the world, which is not itself unbiased. Knowing your model's failure modes matters before deploying it in high-stakes decisions.
In practice
You build a hiring screening tool using Claude. You test it and notice it consistently ranks candidates from certain universities lower. That's AI bias — the model learned patterns from historical data that reflect past discrimination, not actual job performance. Catching this before deployment is why bias testing matters.
Related concepts