The ethical algorithm
summary
Shows that common anonymization breaks via linkage attacks and cites the 2010 law‑review paper, then advocates differential privacy implemented by adding calibrated noise — gives randomized response as a 40‑year‑old instance and notes Google mobility reports and the 2020 US Census rollout. Quantifies the unavoidable fairness‑accuracy trade‑off: a college‑admissions toy where group shifts cause 100% false rejects for a minority unless constraints are added, and recommends plotting efficient frontiers to choose balances.
AI-written summary · about
chapters (9)
- 0:00Introduction to ethical algorithms
- 1:49Scientific agenda for ethics
- 6:05Anonymization and privacy limits
- 10:04Differential privacy explained
- 15:13Differential privacy in practice
- 17:24Fairness in machine learning
- 18:06Bias in predictive models
- 23:17Fairness trade-offs and constraints
- 26:57Summary and takeaways
related talks
38mPrivacy: the Last Stand for Fair Algorithms
33mPractical Data Privacy
01h 43mPerspective
11mAddressing Privacy, Bias, and Fairness in AI Systems
28mAdvertising, algorithms and Privacy: recent opensource developments
47mAI Ethics, Strategic Decisioning and Game Theory
29mPersonal Data Pseudonymization Versus Anonymization In The Age Of AI & Big Data
01h 13mThe risk-based approach to privacy: How far should we go?