So you want to develop ethical AI. What does that mean in practice?
summary
Claims that ethical AI is a checklist fail; ethics is ongoing work that requires contextualized trade-offs, not more universal principles. Concrete failures include 1930s low bridges excluding buses, an Amsterdam smart city lights project that reinforces maintenance inequality, and algorithmic harms from COMPAS and biased image models. The talk offers no plug‑and‑play toolkit but insists teams must do the messy labor of specifying values, choosing which principles to prioritize, and building maintenance processes that actually respond when systems break.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:09Ethical AI is harder
- 4:25Data and AI infrastructure
- 9:47Limits of ethical principles
- 16:14Regulation and risk-based ethics
- 18:38Bias in AI systems
- 24:45Ethics beyond harm limitation
- 27:17No purely ethical AI
- 28:21Better ethical practices
- 31:12Final thoughts
related talks
03mDeveloping AI responsibly
20mImplementing Ethics: Developing Trustworthy AI
47mHow can we make the most of Ethics applied to AI?
34mResponsible AI: The Key to Ethical, Safe and Inclusive Software Development
19mEthical Bias
30mAI Ethics: The Trees that Conceal the Forest
25mResponsible AI/Ethics in AI engineering
05mToward Ethical AI