MEPX
Chapter 8 of 12All chapters
  1. 1What AI means
  2. 2How machine learning works
  3. 3Training data
  4. 4Kinds of learning
  5. 5Neural networks
  6. 6Large language models
  7. 7Limits and hallucination
  8. 8Bias and fairness
  9. 9Using AI well
  10. 10Privacy and safety
  11. 11Beyond text
  12. 12AI and work

Chapter 8 of 12

Bias and fairness

Patterns in the data become patterns in the output.

Where bias enters

A model trained on historical decisions learns the historical pattern, including the unfair parts. A hiring model trained on past hires will reproduce whoever was hired before.

  • Bias can enter through the data, the labels, the objective or the deployment.
  • Removing an attribute rarely helps, because other fields stand in for it.

Why it is hard

Fairness has several definitions that cannot all be satisfied at once, so a system is fair according to a choice someone made. That choice deserves to be stated rather than assumed.