Open Problems In Friendly AI Research
From The Transhumanist Wiki
These problems do not yet have established solutions.
- Training of programmers to be capable of programming Friendly Seed AI. This is more complicated than it sounds, since programming Seed AI is probably the hardest thing any human will ever have to do.
- Ensuring that the Seed AI will not become unFriendly after it goes FOOM or reaches a point where it could escape if it were unFriendly. This is of particular concern because it either requires:
- FriendlinessArchiticture that is so good that the AI could not possibly end up UnFriendly, no matter how it started out.
- FriendlinessContent capable of being taught before the AI reaches a level of intelligence where it could escape or go FOOM.
