Plain explanation
You are here
Current learning moveOrient
Clarify a term, inspect evidence, or place an idea in context.
inAiLoading page content...
…
inAiClarify a term, inspect evidence, or place an idea in context.
inAiLoading page content...
Connected glossary
Plan for this page
Plain explanation
Example
A speech system trained mostly on one accent may have higher error rates for speakers with other accents.Why it matters
Average performance can conceal patterned harm. Examining data, error distribution, affected groups, decision process, and remedy is more useful than asking whether a system is simply “biased or not.”Common misunderstanding
Misconception: bias exists only when a developer intends discrimination. Correction: systematic effects can arise without intent through data, measurement, objectives, deployment, and institutions.Sources and scope
Frames harmful bias as a contextual lifecycle risk involving data, models, people and institutions; it does not prescribe one fairness metric.