AI Bias
When an AI treats some people or groups unfairly because of patterns in its data or design.
In everyday terms
A hiring model trained on past hires may favour the kinds of people who were hired before, repeating old unfairness at scale.
For professionals
Systematic error from skewed data, labels, objectives or deployment context. Measured with fairness metrics; mitigated across the pipeline.
Think of it like…
A mirror: the AI reflects the world in its data, flaws included.
You've already seen it
Reports of image generators stereotyping professions, or facial recognition performing unevenly.
Myth vs reality
Myth: Computers are neutral, so AI can't be biased.
Reality: AI learns from human-made data and choices, which carry bias.
Quick check
The most common source of AI bias is…
- Electricity
- The training data
- Screen size
- The user's Wi-Fi
Show answer
The training data: Models absorb patterns from data.