In defence of uncertainty
summary
Argues that chasing absolute certainty in social systems produces harm by embedding historical bias into models that are then deployed: examples include COMPAS recidivism algorithm producing racially skewed risk scores, an Amazon hiring model that downgrades women, face recognition that labels Black people as gorillas, proposals to predict shoplifters or sexual orientation using ML. Recommends treating these problems as ethical, not purely technical, asking whether tools do more harm than good, and prioritizing understanding and openness over prediction and finality.
AI-written summary · about
chapters (6)
- 0:00Start
- 21:39Amazon Hiring System
- 27:39Sentiment Classifiers
- 31:56Computational Models of Cognition
- 36:51Machine Learning To Predict Sexual Orientation
- 40:20Take-Home Message
related talks
16mExistential Engineering or How Ethics Precedes Error Handling
47mAI Ethics, Strategic Decisioning and Game Theory
31mEthics in technology: History, practice, and mitigation
54mEthical Machine Learning
29mResponsible AI Challenges and Recommendations
47mEthical Machine Learning
39mEthical ML: Who's Afraid of the Black Box Models?
01h 10mJournalist-in-Residence Talk: The Limits of Fairness