Understanding the nature of Bias in Natural & AI Systems
summary
Defines bias as different treatment of groups, models it as a scaled-down perceived utility, and runs a stylized hiring model with 100 candidates (50/50 groups) showing that an affirmative action constraint — forcing at least one hire from the disadvantaged group — can increase overall utility and quickly update interviewer beliefs after a single exposure. Gives concrete examples (loan rules, SAT admissions), cites empirical surprises and points viewers to Weapons of Math Destruction and the documentary Coded Bias.
AI-written summary · about
chapters (15)
- 0:00Bias in AI systems
- 1:28Defining bias and origins
- 4:16What bias really means
- 8:13Loan approval bias
- 12:06Affirmative action limits
- 15:34Positive discrimination boundaries
- 19:54Social networks amplify bias
- 23:07Are humans discriminatory
- 24:51Affirmative action models
- 28:02Diverse teams reduce bias
- 31:57Multi-disciplinary teams in AI
- 36:50Why multi-disciplinary teams
- 39:35Collaboration challenges
- 40:16Resources on AI bias
- 41:41Closing remarks
related talks
47mFight AI Bias With… Bias
35mZero Bias AI
38mBiases are bugs: algorithm fairness and machine learning ethics
56mAn AI with an Agenda: How Our Cognitive Biases Leak Into Machine Learning
20mThere is no such thing as bias-free tech. How models shape our world.
34m(Un)Ethical Machines? Why AI Bias Is a Problem and What to Do About It
21mFairness and bias in AI
01h 31mHow To Measure And Mitigate Unfair Bias in Machine Learning Models