Algorithmic bias is when an automated system gives worse results for some groups of people, usually because of the data it learned from or how its goal was set.
Everyday examples
- Face recognition: the 2018 Gender Shades study found commercial systems misclassified darker-skinned women far more often than lighter-skinned men.
- Hiring: screening tools that learn from past hires can repeat past preferences.
- Image generators: prompts for “a CEO” or “a nurse” often return stereotyped results.
What helps
Diverse test data, measuring results by group, and a route for people to challenge a decision.
Comments