What are Fish? The Ethics of Rules
summary
Connectionist systems like neural networks and large language models embed decision rules across weights so connectionist systems hide rules and explainability reaches a hard limit; an IBM demo surfaces probabilistic factors for loan denial, Google Photos’ gorilla failure shows biased training data and Google simply blocks the search as a mitigation. The claim is that ethics requires operational changes—continuous delivery / MLOps, broader data, and the provisional imperative of provisionality, irony, creativity and transgression to accept, test and undo models.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:33Defining fish
- 9:43Ethics in rule definition
- 14:01Rule-based vs connectionist AI
- 17:29Rules in connectionist AI
- 23:56AI as complex system
- 28:36AI limits contradiction
- 33:20Training data ethics
- 37:39Accuracy misses ethics
- 39:46AI obscures responsibility
- 40:28Building ethical AI
- 43:42Continuous delivery ethics
- 44:25Embracing irony
- 46:12Diversity in AI design
- 47:39Transgression in AI
- 50:31Ethical AI frameworks
