How does one define the concept of fairness in machine learning? I've seen the term lots of times but never used it myself in research (1, 2). Is there a generally agreed-upon definition of fairness in machine learning? What are the different aspects of fairness? Or the intuition behind the concept of fairness?

  • 1
    $\begingroup$ It might be a good idea to provide 1-2 links/quotes that use this term. I've seen it in the context of explainable AI. $\endgroup$
    – nbro
    May 21 at 18:27

1 Answer 1


The Ethical Guidelines for Trustworthy AI list four principles:

  1. Respect for human autonomy
  2. Prevention of harm
  3. Fairness
  4. Explicability

In fairness, two dimensions are considered.

The substantive dimension implies a commitment to: ensuring equal and just distribution of both benefits and costs, and ensuring that individuals and groups are free from unfair bias, discrimination and stigmatisation. If unfair biases can be avoided, AI systems could even increase societal fairness. Equal opportunity in terms of access to education, goods, services and technology should also be fostered. Moreover, the use of AI systems should never lead to people being deceived or unjustifiably impaired in their freedom of choice. Additionally, fairness implies that AI practitioners should respect the principle of proportionality between means and ends, and consider carefully how to balance competing interests and objectives.

The procedural dimension of fairness entails the ability to contest and seek effective redress against decisions made by AI systems and by the humans operating them. In order to do so, the entity accountable for the decision must be identifiable, and the decision-making processes should be explicable.


  • $\begingroup$ Thanks for the answer! My question specifically mentioned Fairness in machine learning. How does this definition translate to definitions/applications in machine learning? $\endgroup$ May 22 at 16:47
  • $\begingroup$ @RobinvanHoorn consider a ML system (e.g. classifier) used in banking to rank possibly bad future/current clients. You don't want such a system to base its predictions solely or mostly on features like gender, age, race, etc. $\endgroup$ May 22 at 18:46
  • 1
    $\begingroup$ I understand the abstract definition and how i'd want my system to behave accordingly. But when can i say that my model is fair? Can i test it? Can i give my model incentives to converge to a 'fair' state? How is this incorporated into machine learning? $\endgroup$ May 22 at 19:25

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .