Algorithmic bias
Also known as: AI bias · algorithmic discrimination · fairness
Algorithmic bias is a systematic difference in treatment produced by an automated system to the detriment of a group of people.
It rarely comes from intent. It comes from training data reflecting past decisions, and passes on silently: a model trained on ten years of hiring reproduces ten years of preferences.
Removing the sensitive variable is not enough. Other fields stand in for it through correlation — a neighbourhood, a school, a career gap — and the bias persists in a less visible form.
Detecting it means measuring outcomes by group, which paradoxically requires holding the sensitive data you are trying not to use. That is a genuine tension, not an implementation detail.
What it means for a small business
A small company buying a screening tool inherits its supplier's bias. The purchasing question is whether the supplier measured anything at all, and what they will show you.
So what actually applies to you?
A definition tells you what a term means, not what your organization must do. The assessment answers the second question — free, no credit card.
Updated September 1, 2026 · Educational definition; not legal advice.