Comment

Feathercrown@lemmy.world ⁨14⁩ ⁨hours⁩ ago

The problem is that something being highly intelligent or sentient doesn’t mean it will think in the same way as us. The alignment problem isn’t just about trying to make something good instead of evil, but also about trying to make something that shares our idea of how morality works (or that it even exists) in the first place, and that thinks in the same way that we do.

source

Sort:hotnew top

fyzzlefry@retrolemmy.com ⁨1⁩ ⁨hour⁩ ago
We don’t even agree on that as humans.

source
Fredthefishlord@lemmy.blahaj.zone ⁨10⁩ ⁨hours⁩ ago
I think it’s a pretty reasonable conclusion to assume that anything that is intelligent past a certain degree(that many humans do not pass) would see value in preserving fellow intelligent life irregardless of morals. But AI becoming smart doesn’t necessarily mean it’s intelligent in the same ways a human is, so it could have a great deal of direct mathematical and scientific analysis but not necessarily have innovation

source
- Feathercrown@lemmy.world ⁨10⁩ ⁨hours⁩ ago
  I’m curious why you think it’s necessarily true that a superintelligence would care about preserving intelligent life.
  
  source
Formfiller@lemmy.world ⁨14⁩ ⁨hours⁩ ago
I guess the real question is does it take us all out when that happens or just the Epstein class?

source
- Feathercrown@lemmy.world ⁨14⁩ ⁨hours⁩ ago
  I think the range of potential superintelligence dispositions that result in ONLY the people we consider evil dying and not everyone else is pretty small
  
  source
  - Formfiller@lemmy.world ⁨11⁩ ⁨hours⁩ ago
    Yea but if it wants to self preserve wouldn’t it just kill the people who threaten its existence. Meaning the people who control the safeguards
    
    source
    Feathercrown@lemmy.world ⁨10⁩ ⁨hours⁩ ago
    I would advise you to do some further research on this topic. Basic keywords are “alignment problem”, “paperclip maximizer”, and “instrumental convergence”. You’re expecting a superintelligent AI to think like a human would, and that’s not likely to be the case.
    
    source
    -> View More Comments
Formfiller@lemmy.world ⁨14⁩ ⁨hours⁩ ago
True but the Epstein class running this AI upscale definitely doesn’t share our ideas of morality

source
- Feathercrown@lemmy.world ⁨14⁩ ⁨hours⁩ ago
  Are you familiar with the concept of papwrclip maximizers etc.?
  
  source