Ethics in the age of AI: Strategies for mitigation and their historical context
summary
Claims that technology is neutral get challenged with concrete examples: PredPol <1% accurate (0.4% overall in Wired’s analysis of ~24,000 predictions), a patented predictive‑policing tool that amplifies bias through a self‑reinforcing feedback loop driven by arrest and call‑for‑service data. Mitigations include model fine‑tuning, choosing smaller domain models, safety prompts and meta prompts with grounding, surfacing sources in the UX, and adversarial red‑team testing; Copilot examples show what is built and how these defenses actually reduce hallucinations and harmful outputs.
AI-written summary · about
chapters (15)
- 0:00Ethics in AI technology
- 1:43AI promise vs dystopian fears
- 3:07Technology not ethically neutral
- 3:27Predictive policing case study
- 6:14Predictive policing data problems
- 9:02Biased policing historical roots
- 10:27AI feedback loops
- 11:51Underrepresentation and AI bias
- 13:52Unintentional AI bias
- 14:35Technology’s harmful impacts
- 15:18Model mitigation strategies
- 16:42AI safety systems
- 18:28UX and grounded content
- 19:10AI safety red teaming
- 20:34Conclusion and responsible AI
related talks
19mEthical Bias
31mEthics of algorithms - from theoretical discussion to practical application
50mSecurity as Part of Responsible AI: At Home or At Odds?
35mRisks and Mitigations for a Safe and Responsible AI
28mEthical AI?!
23mThe AI Dilemma — Steering towards a More Humane Future
29mThe Political Roots of AI Policy
26mBuilding ethical AI: Transparency, bias mitigation, and trust