Skip to content
BENIBIX Learn

Algorithmic Fairness

AI⚖️ Ethics & RegulationAI Ethics Basics

Also known as Fairness in Machine Learning, AI Fairness, Fair ML

Checking that an AI treats each group the same, and fixing it if not.

Think of it like

It is like a game where each team plays by the same rules.

Example

Emma's team checks its credit model. Both groups are approved at similar rates, but one group is wrongly turned down twice as often, so they retune the model for that.

Why it matters

You cannot fix what you do not measure. Fairness tests turn a vague worry into a number a team can act on.

Definition

Algorithmic fairness is the work of measuring and fixing unfair results in an AI system. It starts by choosing what fair means for the task, and writing it as a number. One test asks if each group is approved at the same rate. Another asks if each group is wrongly turned down at the same rate. These tests can clash, so a team has to choose which one fits the harm it most wants to avoid. Fixes can change the data, the model, or the cut-off score.
Two groups compared: how often each is approved, and how often wrongly turned down.

Common mistake

Thinking there is one fairness test that covers everything. The tests can disagree, so pick the one that matches the harm.

Key takeaways

  1. 1
    Pick what fair means. For this task.
  2. 2
    Measure by group. Rates and errors.
  3. 3
    Tests can clash. So choose.