Responsible AI
AI fairness: what it means, and what a team can check
Fairness is easy to name and hard to define. A practical review starts with the harm that could reach a person, not with a single number.
AI fairness is the effort to make sure an AI system does not cause unjust harm to people or groups. NIST describes it as including concerns for equality and equity by addressing harmful bias and discrimination, and notes that standards of fairness are complex and differ between cultures and applications. A system with balanced results across groups is not automatically fair. A practical review names who could be affected, looks for bias in the data, the method and the people using the output, and records what was found and decided.
plain answer
What is AI fairness?
The NIST AI Risk Management Framework lists "fair with harmful bias managed" as one of the characteristics of trustworthy AI. NIST writes that "fairness in AI includes concerns for equality and equity by addressing issues such as harmful bias and discrimination," and that "standards of fairness can be complex and difficult to define because perceptions of fairness differ among cultures and may shift depending on application."
That is a useful starting point because it does not pretend there is one formula. Fairness is decided in context, by people, and it has to be written down.
framework
Three kinds of bias to look for
NIST has identified three major categories of AI bias to be considered and managed. Each "can occur in the absence of prejudice, partiality, or discriminatory intent."
Systemic bias
NIST says it "can be present in AI datasets, the organizational norms, practices, and processes across the AI lifecycle, and the broader society that uses AI systems."
Computational and statistical bias
NIST says it "can be present in AI datasets and algorithmic processes, and often stem from systematic errors due to non-representative samples."
Human-cognitive bias
NIST says it relates to "how an individual or group perceives AI system information to make a decision or fill in missing information, or how humans think about purposes and functions of an AI system."
ai gap
Why balanced numbers are not enough
NIST warns that "systems in which harmful biases are mitigated are not necessarily fair." Its example is a system whose predictions are somewhat balanced across demographic groups but which may still be inaccessible to individuals with disabilities, affected by the digital divide, or may exacerbate existing disparities or systemic biases.
NIST also says bias "is broader than demographic balance and data representativeness." A review that only compares one metric across groups can pass a system that still harms someone.
how to
What a fairness review can include
Name who could be affected
List the people and groups a system's output could reach, including people who never use it directly.
Name the decision the output influences
A tool that drafts internal notes needs a light check. A tool that influences a decision about a person needs more.
Look in all three places
Check the data, the method, and how people read and act on the output. A reviewer who looks only at the model misses the third.
Include more than one view
NIST GOVERN 3.1 says decisions about AI risks should be "informed by a diverse team (e.g., diversity of demographics, disciplines, experience, expertise, and backgrounds)."
Record the result
Keep a dated record of what was checked, what was found, what changed and who accepted any remaining risk.
Check again
Review on a schedule and after changes to the data, the model or the way it is used.
differentiation
How this differs from our related pages
This page is the general guide to AI fairness. Our AI underwriting bias page covers bias in insurance underwriting models, and our population health model bias page covers it in healthcare. Our AI governance framework page covers the wider policy framework, and our NIST AI Risk Management Framework page covers the framework used here.
kriv role
What Kriv AI does
For each AI system or use case we record its purpose, the data it touches and the people it affects, then rate the risk and set the review that rating requires. We do not certify a system as fair. Our AI ethics page sets out the principles we apply. Every engagement is scoped and priced after a short discovery call.
Sources
Cited sources
- NIST, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1, January 2023, sections 3.7 and GOVERN 3.1 (read 10 November 2026)
- NIST, Towards a Standard for Identifying and Managing Bias in Artificial Intelligence, SP 1270, March 15, 2022 (further reading; no claim on this page relies on it; read 10 November 2026)
Straight answers
Frequently asked questions about AI fairness: what it means, and what a team can check
What is AI fairness?
NIST describes it as including concerns for equality and equity by addressing issues such as harmful bias and discrimination. What counts as fair differs between cultures and applications.
What are the types of AI bias?
NIST names three major categories: systemic, computational and statistical, and human-cognitive. Each can occur without any intent to discriminate.
Is an AI system fair if its results are balanced across groups?
Not necessarily. NIST notes that a system with somewhat balanced predictions may still be inaccessible to people with disabilities or may worsen existing disparities.
Who should review an AI system for fairness?
A team with different disciplines and backgrounds. NIST says decisions about AI risks should be informed by a diverse team.
Talk to the team that would do the work
Bring your requirements to a working session with the person who'll actually deliver.
Book a Discovery Call