Ethical AI: Addressing Bias & Algorithmic Fairness
summary
Demonstrates concrete tools and results: examples include a historical admissions dataset that admits twice as many men as women, face‑generation biases from PULSE/StyleGAN, and a McAfee deep‑fake dataset under review whose race breakdown shows two fairness metrics (statistical parity difference and disparate impact) outside the ±0.1 fair range. Uses AI360 Fairness metrics and the DeepFace attribute detector to measure bias, then applies adversarial debiasing at the classifier level and remeasures to confirm reduced racial bias.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 0:27Bias in AI
- 3:48Bias
- 6:07Ethical AI
- 10:03Explainability and Bias
- 14:57Where Bias Exists
- 18:47Measuring Bias
- 23:55Real World Example
- 27:59Malware Detection
- 31:00Summary
- 31:40Outro
related talks
11mAddressing Privacy, Bias, and Fairness in AI Systems
47mEthics & AI: Identifying & preventing bias in predictive models
23mRemoving Unfair Bias in the Machine Learning Pipeline
26m(Un)ethical Artificial Intelligence: How to Keep the AI Fair for Everyone
26mBuilding ethical AI: Transparency, bias mitigation, and trust
21mFairness and bias in AI
40mResponsible AI Practices: Fairness in ML
47mEthical Machine Learning