I got that part. Here's my issue tho: the paperclip maximizer turns it's programming, it's vision, into a net negative for everyone else and is thus indistinguishable from the many people who are, while potentially sharing A - or THE greater goal - are turning the achievement of their sub-goals into a net negative for everyone, including themselves.
But an 'advanced' artificial intelligence wouldn't do that anyway, because 'advanced' means that you 'understand' - are aware of - the emergence and self-organization of 'higher-dimensional' structures that are build on a foundation.
Once a child understands Legos, it starts to build more and then more out of that ...
A lot can be build out of paperclips, but an 'advanced' AI would rather quickly find the dead end and thus decide - in advance - that maximizing the production of paperclips is nonsense.
But an 'advanced' artificial intelligence wouldn't do that anyway, because 'advanced' means that you 'understand' - are aware of - the emergence and self-organization of 'higher-dimensional' structures that are build on a foundation.
Once a child understands Legos, it starts to build more and then more out of that ...
A lot can be build out of paperclips, but an 'advanced' AI would rather quickly find the dead end and thus decide - in advance - that maximizing the production of paperclips is nonsense.