Privacy: the Last Stand for Fair Algorithms
summary
Argues that privacy protects fairness by design and shows concrete methods: differential privacy as a formal guarantee, learned fair representations that retain utility while reducing discrimination on the UCI Adult data set, and an adversarial training setup that after ~150 epochs meets disparate impact constraints. Cites a mathematical feedback‑loop proof for predictive policing, a Google internal pay‑gap spreadsheet example, and GDPR’s real‑world effect of sites blocking EU traffic to illustrate tradeoffs and failures.
AI-written summary · about
chapters (10)
- 0:00Intro
- 2:37Data Science: The Solution to Social Ills
- 6:42Predictive Policing
- 9:35Privacy as Privilege
- 12:03Privacy as Power
- 17:21Privacy and Fairness
- 22:27Learning Fair Representations
- 24:48Private and Fair Representations
- 27:47Private and Fair Machine Learning
- 31:45Privacy By Design
related talks
27mThe ethical algorithm
01h 20mUnlocking societal benefits of AI with privacy protective technologies.
11mAddressing Privacy, Bias, and Fairness in AI Systems
28mAdvertising, algorithms and Privacy: recent opensource developments
42mFairness in Machine Learning: From Theory to Practice
33mPractical Data Privacy
31mOperationalizing Responsible AI
35mOpen Source Fairness