The Myth of Neutrality: How AI is widening social divides
summary
Shows that commercial face recognition causes real harms — multiple wrongful arrests and Clearview scraping 10 billion images — and that core datasets like ImageNet (≈14 million images) and Flickr-scraped faces carry geographic, consent and labeling bias. Presents measured failures: Gender Shades finds up to ~34% accuracy gap on darker-skinned women, and Amazon’s resume model is abandoned for gender bias. Explains concrete toolchains (data scraping, Mechanical Turk labels, proxy features, deployment in hiring and policing) and why common metrics mask subgroup harms.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 0:39The AI landscape
- 3:53The birth of AI
- 4:59Who is funding AI
- 7:20Big Tech
- 11:31Where do images come from
- 12:21How are images labeled
- 15:40Gender Shades
- 18:00Diversity in Phases
- 19:02Centralizing Power
- 19:44Hiring and firing
- 23:02Amazons hiring tool
- 24:11Building fairness
- 28:30What is fair
- 33:34Questions
related talks
47mEthics & AI: Identifying & preventing bias in predictive models
38mSocial Implications of Bias in Machine Learning
27mThe Limitations and Danger of Facial Recognition
38mSocial Implications of Bias in Machine Learning
21mBe more responsible around AI - Less Bias, More Ethics
29mUnintended consequences of technology: Enabling NextGen discrimination
34m(Un)Ethical Machines? Why AI Bias Is a Problem and What to Do About It
28mAlgorithmic Bias: Data as a Weapon with Ayodele Odubela