Ethical Machine Learning
summary
Covers concrete failures and fixes: predictive policing simulations show feedback loops that concentrate stops until under-policed areas vanish, and an improvement policy that counts surprising discoveries evens patrols; facial-gender APIs show a 30-point error gap between darker-skinned women and white men and prompted dataset fixes; adversarial training (GAN-style discriminator) drives disparate impact to 80% after ~150 epochs at ~10% accuracy cost; debiasing word vectors and adding fairness metrics into training round out the toolbox.
AI-written summary · about
chapters (10)
- 0:00The ethics of algorithms
- 2:11Policing and disparate impact
- 7:07Automating social inequality
- 12:20Identifying data prejudice
- 16:02Better data collection
- 18:49Addressing linguistic bias
- 25:15Fairness-driven evaluation
- 32:33Model interpretability
- 34:17Ethical design principles
- 37:50Call to action and conclusion
related talks
50mEthical Machine Learning
37mKeynote: Ethical Machine Learning: Creating Fair Models in an Unjust World
38mSocial Implications of Bias in Machine Learning
38mSocial Implications of Bias in Machine Learning
54mEthical Machine Learning
55mEthical Machine Learning
51mDigital Discrimination: Cognitive Bias in Machine Learning
35mBuilding Inclusive Machine Learning Models