Algorithmic bias is a systematic error that advantages or disadvantages certain groups of people. It can come from non-representative data, labels marked by prejudice, or design choices, and is detected by comparing results between groups.
Example
A recruitment model trained on past hires reproduces past preferences.