What are Fish? The Ethics of Rules
summary
Argument that modern AI models like ChatGPT DALL·E MidJourney and GitHub Copilot are connectionist systems whose rules live in distributed weights so explainability hits a fundamental limit, illustrated by the Google Photos mislabeling of gorillas and an IBM loan denial demo that returns probabilistic reasons not certainties. Claim that complexity is incompressible so accuracy alone is insufficient and engineering must include continuous delivery for ML / MLOps plus the provisional imperative of provisionality irony creativity and transgression.
AI-written summary · about
chapters (14)
- 0:00AI ethics introduction
- 1:29Defining fish and rules
- 9:43Rule-making exclusions
- 14:01AI ethical rules
- 21:00Explainable AI limits
- 23:56AI unpredictability
- 28:36Why AI cannot simplify complex reality
- 33:20Training data bias
- 37:39Accuracy vs ethics
- 39:46AI hides ethical choices
- 40:29Building ethical AI
- 45:49Ethical AI imperative
- 50:31Ethical frameworks
- 52:16Conclusion
related talks
53mWhat are Fish? The Ethics of Rules
43mAI Engineer Ethics: Shaping Responsible Generative AI
47mUnderstanding the Legal Context of AI
46mCan you trust your AI?
01h 00mThe Promise and Limitations of AI
01h 00mAI for Social Responsibility: Embedding (...) into AI
41mEthics Panel
43mAI Systems: Foundations for Artificial Minds