Part III · Chapter 25Version 1.2 · 5 November 202635 min read
Data Collection and Representation Biases
Key message
Bias enters many AI systems before any model is trained: through the world the data record, who is sampled, how constructs are measured and how features are encoded, so a fairness assessment starts with a documented audit of the data.
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.