Skip to content
BENIBIX Learn

AI Bias

AI⚖️ Ethics & RegulationAI Ethics Basics

Also known as Algorithmic Bias, Machine Learning Bias, Model Bias

When an AI is not fair to some groups of people, time after time.

Think of it like

It is like a scale that reads a bit heavy for some people. Each time, it is not fair.

Example

Jack's team builds a model to rank job seekers. It learns from ten years of past hires. Soon it ranks people from one school higher, because that is who was hired before.

Why it matters

One unfair model can make thousands of bad calls a day. Each one looks like a plain, fair number.

Definition

AI bias is when an AI system treats some groups of people unfairly, again and again. It most often comes from the data. If past data shows more men hired, a model can learn to prefer men. It can also come from what the model is asked to predict, or from who was left out of testing. Bias is hard to see from one case. It shows up when you compare results across groups, such as by gender, age or skin color.
Lopsided past data in, lopsided scores out.

Common mistake

Removing the gender or race column and calling it fixed. Other columns, such as postcode or hobbies, can stand in for it.

Key takeaways

  1. 1
    Unfair to a group. Again and again.
  2. 2
    Mostly from data. Past choices repeat.
  3. 3
    Seen across groups. Not in one case.