Bias in AI outputs

AI models learn patterns from huge amounts of existing text, and existing text reflects the biases, stereotypes, and imbalances present in the real world and in whatever was included in training data. A model can reproduce or even amplify those patterns without any deliberate intent behind a specific output.

This matters most in contexts involving people: drafting job descriptions, summarising candidate applications, assessing performance, or generating content about demographic groups. Review AI-assisted outputs in these areas with the same, or greater, scrutiny you'd apply to a human colleague's first draft, and don't treat an AI-generated assessment of a person as inherently neutral just because it wasn't written by a human.