Bias in Artificial Intelligence
summary
Algorithms trained on biased records systematically misclassify and harm people of color: NIST tests show some systems falsely ID Black and Asian faces 10 to 100 times more than white faces; MIT/Stanford reports lighter‑skinned men under 0.8% error versus dark‑skinned women up to 20–34% error. Predictive tools like PredPol and vendor tools (Palantir received $2.5M from NYPD) drive policing decisions, and risk scores from CONPAS rated Brisha Borden 8 and Vernon Prater 3 despite Prater reoffending. Solutions: diversify data, target underrepresented cohorts, and add monitoring feedback loops.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:55Outline
- 3:31How does AI work
- 6:22Past decades
- 11:28Scenario
- 12:10Algorithmic Bias
- 19:52Technological shortcomings failures
- 30:51Thank you
- 31:52Hippographic oath
- 34:51Bias paradox
- 35:37Fixing humanity
- 37:10The problem of bias
- 39:24Bias in AI design
- 41:14Biases in Asia
- 41:57Facial Recognition
- 43:10Bias in different forms
- 43:43Outro
related talks
31mDoes This Algorithm Have a Soul?
21mFairness and bias in AI
47mDiversity and Inclusion for Algorithmic Bias Considerations
31mAvoiding Digital Bias
05mHow to Erase Unconscious Bias From Your AI Datasets
47mEthics & AI: Identifying & preventing bias in predictive models
38mSocial Implications of Bias in Machine Learning
56mAn AI with an Agenda: How Our Cognitive Biases Leak Into Machine Learning