Gender Bias in Artificial Intelligence
summary
Claims: translation tools map doctor to male and nurse to female, credit-card automation gives identical-history women lower limits, facial recognition misclassifies dark-skinned women at 35% vs 0.8% misclassification, a review finds 44.2% of 133 systems show gender bias. Remedies: balance training samples, assess metrics by demographic cluster, add fairness penalties, center feminist practices and voices. Tools mentioned include speech recognition, skin cancer detection and Dali 2, with several deployed systems empirically failing for women and nonbinary people.
AI-written summary · about
chapters (8)
- 0:00Gender bias in AI intro
- 1:59AI inherits human biases
- 3:47Sources of gender bias
- 7:58Impact of biased AI
- 11:52Reducing bias in ML
- 14:43Social strategies for bias
- 16:12Inclusive AI practices
- 18:41Q&A on ethical AI
related talks
34mAI and Gender: Building Fair AI Systems for the Future
41mAll About Biases in Artificial Intelligence
28mEthical AI
47mHow gender imbalance is holding back the future of AI
38mSocial Implications of Bias in Machine Learning
45mBias in Artificial Intelligence
42mLet's Talk about AI, ML and Bias
35mIs artificial intelligence "fair"?