Ethics in the age of AI
summary
Claims that technology is neutral are challenged with specifics: PredPol’s predictions score about 0.4% accuracy overall and as low as 0.1% for burglary, training data shows overrepresentation plus a self‑reinforcing feedback loop that doubles targeting of communities of color, Amazon Rekognition drops to 68.6% accuracy for Black women, and mitigations that actually move metrics include fine‑tuning, grounding, safety and meta prompts (defect rate falls 67%→12%) and regular red‑team testing.
AI-written summary · about
chapters (12)
- 0:00Ethics in AI introduction
- 1:43AI’s dual narrative
- 3:27Predictive policing case
- 6:14Predictive policing data problems
- 10:07AI feedback loops
- 11:51AI bias and underrepresentation
- 14:35AI’s harmful effects
- 15:18Model-level risk mitigation
- 16:42Safety systems and prompts
- 18:28User experience and grounding
- 19:10AI safety red teaming
- 20:34Conclusion and call
related talks
26m(Un)ethical Artificial Intelligence: How to Keep the AI Fair for Everyone
11mAddressing Privacy, Bias, and Fairness in AI Systems
45mThe Myth of Neutrality: How AI is widening social divides
50mSecurity as Part of Responsible AI: At Home or At Odds?
54mEthical Machine Learning
19mEthical Bias
30mInside AI: What Every Professional Needs to Know to Stay Relevant, Protected, and Ahead
40mResponsible AI Practices: Fairness in ML