We are the guardians of our future
summary
A paper trains a model on the California gang database that feeds text and a biased roster and finds many false positives — about 25% of listed people lack proven gang ties and entries rarely get purged, yet downstream systems use it. Other claims include facial models that allegedly detect criminality at ~90% accuracy, gender at ~86%, and datasets like ImageNet and Labeled Faces in the Wild that are ~70–80% white/male, causing failures (IBM hits ~65% for Black women). Mitigations: ethics boards, data sheets, fairness tools (What‑If), LIME, and humans‑in‑the‑loop.
AI-written summary · about
chapters (4)
- 0:00Start
- 0:55Partially Generative Neural Networks For Gang Crime Classification With Partial Information
- 34:13CONFIRMATION BUBBLE
- 38:07RUNAWAY FEEDBACK LOOP
related talks
34mDespicable machines: how computers can be assholes
27mThe Limitations and Danger of Facial Recognition
31mDoes This Algorithm Have a Soul?
47mEthics & AI: Identifying & preventing bias in predictive models
47mEthical Machine Learning
28mAlgorithmic Bias: Data as a Weapon with Ayodele Odubela
45mThe Myth of Neutrality: How AI is widening social divides
38mSocial Implications of Bias in Machine Learning