Unknown: there’s a difference between not knowing what the output of an algorithm will be, and not knowing anything about the output of an algorithm.
If I write a program that plays chess, I can probably formally prove (by looking at the source code) that the moves it outputs will be legal chess moves.
Even if you don’t know what your brand new Friendly AI will do—even if itself doesn’t know what it will do—that doesn’t mean you can’t prove it won’t turn the universe into paperclips.
Unknown: there’s a difference between not knowing what the output of an algorithm will be, and not knowing anything about the output of an algorithm.
If I write a program that plays chess, I can probably formally prove (by looking at the source code) that the moves it outputs will be legal chess moves.
Even if you don’t know what your brand new Friendly AI will do—even if itself doesn’t know what it will do—that doesn’t mean you can’t prove it won’t turn the universe into paperclips.