AI alignment 4
- A priori arguments on empirical matters, such as that AI will certainly kill everyone, are usually wrong however convincing they seem.
- Restricting general deep-learning AIs with rules will not work, for the same reason simple rules fail on sprawling ecosystems.