Bias in Data
summary
Data claims include concrete numbers and failures: 3.7 billion users generate 40 trillion gigabytes yet under 1% is analyzed, Facebook job ads target 85% of cashier roles to women and 75% of taxi ads to Black users, self‑driving detectors miss dark‑skinned pedestrians by about 5 percentage points, and COMPAS labels non‑reoffenders Black twice as often as White (42% vs 22%). A resume‑ranking tool at Amazon is alleged biased and later stopped using the resume tool while COMPAS defends its outputs. Remedies focus on pre/post‑processing, fairness metrics, and human‑centered feedback loops.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:10Speaker overview
- 3:17Data future and AI role
- 5:09Human vs machine bias
- 7:36AI bias from humans
- 8:20Biased AI in hiring
- 11:51Self-driving cars bias
- 14:45COMPAS algorithm bias
- 17:33Solving AI bias?
- 19:40Human-centered solutions
- 20:42Conclusion
- 21:03Q&A
related talks
31mAvoiding Digital Bias
27mCrouching Harm, Hidden Bias—Addressing issues of fairness....
45mBias in Artificial Intelligence
30mBias, Explainability, and Accountability: The Data Scientist's Burden
38mSocial Implications of Bias in Machine Learning
38mSocial Implications of Bias in Machine Learning
21mFairness and bias in AI
51mDigital Discrimination: Cognitive Bias in Machine Learning