Your AI Is Still Biased (Even After You Checked)
summary
Claim: bias re-emerges after deployment whenever models are retrained, personalized, or exposed to skewed users; an experimental avatar that reads calendar and auto-responds produces a fake angry breakup message, showing real privacy and behavior-change harms. Cites ISO 24368 and the AI Act four risk levels, and gives concrete cases: credit-card–based hospital queueing, MR appointment triage that deprioritizes older patients, predictive-policing patrol boosts, and individualized car-insurance pricing — all fail unless logs, human‑in‑loop checks, governance and cross‑team review are enforced.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 3:18Personal stories reveal bias
- 6:09EU AI Act explained
- 8:38Fairness privacy transparency
- 12:52What is bias and where it comes from
- 18:33Should we give up on AI
- 21:48AI ethics virtue utilitarian rights
- 24:17Checklist for fair AI
- 25:43Health industry case study
- 35:31Police predictive crime analysis
- 42:15Insurance dynamic pricing
- 46:16Summary
related talks
56mAn AI with an Agenda: How Our Cognitive Biases Leak Into Machine Learning
35mZero Bias AI
49mAn AI with an Agenda
41mAll About Biases in Artificial Intelligence
26m(Un)ethical Artificial Intelligence: How to Keep the AI Fair for Everyone
47mFight AI Bias With… Bias
22mTesting for Cognitive Bias in AI Systems
21mFairness and bias in AI