Skip to main content

AI glossary Responsibility

What is bias in AI?

← All glossary terms
In short

Systematic distortion in the data, design or use of an AI system that produces unfair, discriminatory or inaccurate results for some people or cases.

Explained

Bias in AI is a systematic error, not a random one: the system performs worse, or decides differently, for certain groups or situations in a way that is consistent and often hard to see. It can come from training data that under-represents some people, from labels that reflect past human decisions, from design choices about what to optimise, or from how people use and trust the output. Bias cannot be removed entirely, but it can be measured, reduced and monitored. Responsible AI practice treats bias detection and mitigation as part of the life cycle, from data collection to operation.

Where bias comes from

  • Data: samples that leave groups out, historical records that encode past discrimination, or labels applied inconsistently.
  • Model and design: the target being optimised, the features chosen and the trade-offs accepted between accuracy and fairness.
  • Human and context: how people interpret and act on outputs, including over-trusting the system.
  • Systemic: practices and structures in the organisation or society that shape all of the above.

How it is detected and mitigated

Detection starts with measuring results separately for relevant groups and cases, not only on the overall average. Typical checks compare error rates, approval rates or quality scores across groups, and look for gaps.

Mitigation works at several points: improving data coverage and labelling, adjusting how the model is trained or thresholded, adding human review where decisions affect people, and monitoring in production, because data and behaviour drift over time. Documenting what was checked, for example in a model card, makes the work auditable.

Bias and regulation

The EU AI Act requires providers of high-risk AI systems to examine training, validation and testing data for possible biases that could affect health, safety or fundamental rights, and to take measures to detect, prevent and mitigate them. Bias management is therefore not only an ethical question but also a compliance one in regulated uses such as credit, employment or access to public services.

Frequently asked questions

What are the main types of bias in AI?

A common grouping distinguishes data bias (unrepresentative or historically skewed data), algorithmic or design bias (what the model optimises and how), and human bias (how people build, interpret and use the system). NIST also highlights systemic bias in institutions and society.

Can AI bias be eliminated completely?

No. Bias can be reduced and kept within acceptable limits, but not removed entirely, because data and context always carry some imbalance. The realistic goal is to measure it, document it and manage it over the system's life.

How do you detect bias in an AI model?

Evaluate the model separately for relevant groups and cases, compare error and outcome rates, and look for gaps. Combine those metrics with a review of the data sources and with feedback from the people affected.

What does the EU AI Act say about bias?

For high-risk AI systems, Article 10 requires data governance practices that include examining data for possible biases and taking appropriate measures to detect, prevent and mitigate them.

Do large language models have bias?

Yes. They learn from large amounts of human-written text, so they can reproduce stereotypes and uneven quality across languages or groups. Evaluation on your own use cases and guardrails help to control it.

Decision guides that use this term

How to put it to work

Where it shows up at Thinkia

Further reading

Sources

Updated: 2026-10-21

Want to see how this applies to your organisation?

Talk to an AI Expert