I have had more time to think about this since I posted this shortform. I also posted a shortform after that which asked pretty much the same question, but with words, rather than just a link to what I was talking about (the one about why is it assumed an AGI would just use us for our atoms and not something else).
I think that there is a decent chance that an unaligned AGI will do some amount of human experimentation/ study, but it may well be on a small amount of people, and hopefully for not very long. To me, one of the most concerning ways this could be a lot worse is if there is some valuable information we contain, which takes a long time for an AGI to gain through studying us. The worst case scenario would then probably be if the AGI thinks there is a chance that we contain very helpful information, when in fact we don’t, and so endlessly continues studying/ experimenting on us, in order to potentially extract that information.
I have only been properly aware of the alignment problem for a few months, so my opinions and understanding of things is still forming. I am particularly concerned by s-risks and I have OCD, so I may well overestimate the likelihood of s-risks. I would not be surprised if a lot of the s-risks I worry about, especially when they are things which decrease the probability of AGI killing everyone, are just really unlikely. From my understanding Eliezer and others think that literally everyone dying makes up the vast majority of the bad scenarios, although I’m not sure how much suffering is expected before that point. I know Eliezer said recently that he expects our deaths to be quick, assuming an unaligned AGI.
Why do you think it more likely than extinction?
I have had more time to think about this since I posted this shortform. I also posted a shortform after that which asked pretty much the same question, but with words, rather than just a link to what I was talking about (the one about why is it assumed an AGI would just use us for our atoms and not something else).
I think that there is a decent chance that an unaligned AGI will do some amount of human experimentation/ study, but it may well be on a small amount of people, and hopefully for not very long.
To me, one of the most concerning ways this could be a lot worse is if there is some valuable information we contain, which takes a long time for an AGI to gain through studying us. The worst case scenario would then probably be if the AGI thinks there is a chance that we contain very helpful information, when in fact we don’t, and so endlessly continues studying/ experimenting on us, in order to potentially extract that information.
I have only been properly aware of the alignment problem for a few months, so my opinions and understanding of things is still forming. I am particularly concerned by s-risks and I have OCD, so I may well overestimate the likelihood of s-risks. I would not be surprised if a lot of the s-risks I worry about, especially when they are things which decrease the probability of AGI killing everyone, are just really unlikely. From my understanding Eliezer and others think that literally everyone dying makes up the vast majority of the bad scenarios, although I’m not sure how much suffering is expected before that point. I know Eliezer said recently that he expects our deaths to be quick, assuming an unaligned AGI.