(Un)ethical Artificial Intelligence: How to Keep the AI Fair for Everyone
summary
Examples show concrete failures and fixes: an Apple Card applicant gets 20 times higher credit limit than a partner, Amazon deploys Amazon Rekognition racist face matches and a sexist recruiting model that gets scrapped, Microsoft Tay turns toxic in a day, Google removes “gorilla” labels, a self driving car killed a person, and Volkswagen engineers face legal consequences. The actionable points are data diversification, model inspection tools, embedding fairness in the validation pipeline, and treating “black box” excuses as engineering laziness.
AI-written summary · about
chapters (12)
- 0:00AI fairness challenges
- 1:06AI bias and discrimination
- 6:48Costly ethical mistakes
- 9:21Market pressures vs ethics
- 11:06Human biases in AI
- 13:16Engineers' responsibility
- 15:50Ethics and engineering in AI
- 17:15Data diversity for fairness
- 19:47Diverse teams reduce bias
- 22:38Training for fair AI
- 24:27Model inspection tools
- 25:10Fairness in model validation
related talks
47mEthics & AI: Identifying & preventing bias in predictive models
56mAn AI with an Agenda: How Our Cognitive Biases Leak Into Machine Learning
41mWhen AI Failures Reflect Our Own Social Biases
40mEthical AI: Addressing Bias & Algorithmic Fairness
21mFairness and bias in AI
52mSome pitfalls in AI
23mMachine Learning Ethics
49mAn AI with an Agenda