Term: Bias (in AI systems)
~6 min read
Estimated time: ~6 min read — for the in-app brief plus opening the primary source.
What this is
Bias means systematic error that disadvantages or misrepresents groups or outcomes — often inherited from historical data or design choices.
Everyday example
A hiring screen trained on ten years of 'who we hired before' keeps ranking candidates who look like the old majority — even when better applicants exist.
Bias in AI is systematic skew — often inherited from historical data or from how success was defined.
- Hiring, credit, and content tools can scale unfairness faster than a person.
- Accuracy on the past is not fairness on today’s population.
- The Workday/Mobley litigation is the live employment-law signal.
- Fixing it is design, data, and review — not a single “debias” button.
Next action: For any tool that ranks people, require an adverse-impact conversation before renewal.
What changes in how you lead
How decision rights, process, and ownership should change.
- People-decision AI is a selection procedure with Legal in the room.
- Vendors do not automatically absorb discrimination risk.
Compare related ideas
Bias (in AI systems) vs Hallucination
Fixing hallucinations is about truthfulness of content. Fixing bias is about fairness of outcomes across people or segments. Different tests, different owners.
Open HallucinationDeep dive
Ask: for whom does this fail more often? How was that measured?
Human review and appeal paths matter as much as model tweaks.
Document decisions — process evidence matters to regulators and employees.