Systematic errors in AI systems that produce unfair or discriminatory outcomes for particular groups of people, often along lines of race, gender, age, disability, or other protected characteristics. Bias can originate from training data, algorithm design, or how models are deployed and used.

AI bias represents one of the most significant ethical and practical challenges in AI development. Biased AI systems can perpetuate or amplify existing social inequities when deployed in high-stakes domains like hiring, lending, healthcare, or criminal justice. Bias can enter AI systems in multiple ways: through historically biased training data, problematic labelling practices, selection of non-representative features, or misalignment between optimisation objectives and fairness considerations. Addressing bias requires holistic approaches spanning data collection, model development, testing, and deployment.

‍

A loan approval algorithm that consistently ranks applicants from certain zip codes lower than equally qualified applicants from other areas, effectively perpetuating historical redlining practices through algorithmic decisions.