AI Bias
When an AI system consistently produces worse outcomes for certain groups — and the organization doesn't know it yet.
AI bias occurs when a model produces systematically unfair or inaccurate outcomes for specific groups, situations, or contexts. It can originate in training data that underrepresents certain populations, labels that reflect historical discrimination, objectives that optimize for the wrong thing, or feedback loops that compound early imbalances. Removing protected attributes like race or gender from the model doesn't eliminate bias — other features often serve as proxies, and the effect on outcomes remains.
Biased AI in hiring, lending, pricing, or service access can violate discrimination law, trigger regulatory scrutiny, and cause real harm to the people least able to push back. The exposure is compounded by scale: a biased model applied to thousands of decisions a day can produce discriminatory outcomes far faster than any manual process. Choosing a reputable vendor doesn't transfer the risk — the organization deploying the model is responsible for what it does in context.
Read next
Related concepts
Training Data
The data a model learned from — which means everything the model knows, gets right, gets wrong, and embeds as bias all traces back here.
Governance and RiskFairness
Whether an AI system produces outcomes that are equitable across different groups — and why accuracy alone doesn't guarantee it does.
Governance and RiskResponsible AI
Responsible AI is the difference between an organization that says it uses AI ethically and one that can actually prove it.
Optional map
Concept neighborhood
Focused neighborhood
AI Bias
When an AI system consistently produces worse outcomes for certain groups — and the organization doesn't know it yet.
In these paths
Selected concept
Directly related
One step further
via Training Data
via Fairness
via Responsible AI
via Model Evaluation