Avoiding Digital Bias
summary
Algorithms trained on messy personal data routinely produce unfair outcomes: COMPAS recommends incarceration for an 18‑year‑old woman but not for a comparable man, browser‑plugin studies show Google serves higher‑pay job ads to men, face recognition performs worse on darker faces, Staples’ pricing raises costs for lower‑income ZIP codes, and an event’s facial‑recognition scan yields about 94 false positives out of 100. Tools include proprietary ML models, browser plugins and production datasets; quadrupling data roughly doubles accuracy yet Internet data often amplifies bias.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 3:24Artificial Intelligence
- 7:20Machine Learning
- 9:18The Problem
- 10:25What is Bias
- 12:37Examples
- 13:29Counterargument
- 14:12Face Recognition
- 15:08Removing Bias
- 17:42Staples
- 18:22Amazon
- 19:10Facebook
- 19:31Compass
- 21:15Algorithms
- 22:47Regulation
- 23:27Recap
- 24:12How to determine if an algorithm is discriminatory
- 25:38Approach throughout the lifecycle
related talks
38mSocial Implications of Bias in Machine Learning
25mBias in Data
45mBias in Artificial Intelligence
38mSocial Implications of Bias in Machine Learning
05mHow to Erase Unconscious Bias From Your AI Datasets
51mDigital Discrimination: Cognitive Bias in Machine Learning
31mDoes This Algorithm Have a Soul?
21mFairness and bias in AI