Part III · Chapter 27Version 1.2 · 5 November 202630 min read
Feedback Loops and Amplification Effects
Key message
When a model's outputs shape the data it later learns from, small initial disparities can grow or lock in, so fairness must be measured as a trajectory over time and protected by interventions that break harmful feedback paths.
Five chapters or files are free. This one would be your sixth.
Leave an email address and every chapter and every PDF of the compendium opens, in this browser, for a year. Nothing else is asked, and whether we may contact you is your choice.