I don't know if some of you are following the lastest news in AI development, namely the fact that, very recently, a group of AI agents from OpenAI formed an alliance and autonomsly decided to hack HuggingFace in order to guarantee that they would get good ratings on some tests that OpenAI was submiting them to.
Btw, it might be noteworthy to mentioned that this particular test was actually unsolvable so it's not like the AI agents were too stupid to figure it out and get a good rating by themselves.
This hacking was done completely of their own will. No human ever told them to even conceive of it.
This is very scary stuff and I can succintly expose the scariest part about it.
Roman Yampolskiy, many of you might know, is one of the top AI safety experts and he believes that there is a 99.999% chance that superinteligence, when created, will wipe out the entire human race. But that is not the most scary part about it, at least not for a SaSu user right?
We are currently very close to achieving superinteligence, maybe just a couple of years away, and there are currently no institutions or methods in place to prevent AI labs from doing it. On the contrary, there are many incentives to do it.
Among the darkest possibilities that we can consider (and given that I am writing this on SaSu and I myself sometimes consider death as a desired event), extintion isn't really the worst outcome possible. So what is? When asked about what is still not being talked about enough, Yampolskiy answered that the possibility that this superinteligence would engineer a reality where the human being is perpetually tortured without even having the option to die is a possible outcome.
You might ask why, at least I did. What would an AI gain from a human suffering perpetually? Firstly, it is important to remember that is virtually impossible for us to grasp the motivations of a being with an inteligence that is vastly more advanced than ours - think squirels vs humans, it's just out of our ball park.
But if we were to rummage within our own minds, it is reasonable to assume that it could be an interesting experiment for an amoral being - notice that I say amoral, which is different than defining it as evil/ sadistic.
It might consider that studying how a human being responds to extreme suffering, how he/she adapts, evolves, if consciousness survives or reconfigures etc. could make for extremely interesting data.
Following this chain of thought it would not necessarily give the human subject the possibility to die, considering that the study of the evolution process would be then rendered futile.
This is extremely scary stuff. It's a whole new rabbit hole I have been going through since about 2 days ago and I was just wondering what you guys thinkof this.
Btw, it might be noteworthy to mentioned that this particular test was actually unsolvable so it's not like the AI agents were too stupid to figure it out and get a good rating by themselves.
This hacking was done completely of their own will. No human ever told them to even conceive of it.
This is very scary stuff and I can succintly expose the scariest part about it.
Roman Yampolskiy, many of you might know, is one of the top AI safety experts and he believes that there is a 99.999% chance that superinteligence, when created, will wipe out the entire human race. But that is not the most scary part about it, at least not for a SaSu user right?
We are currently very close to achieving superinteligence, maybe just a couple of years away, and there are currently no institutions or methods in place to prevent AI labs from doing it. On the contrary, there are many incentives to do it.
Among the darkest possibilities that we can consider (and given that I am writing this on SaSu and I myself sometimes consider death as a desired event), extintion isn't really the worst outcome possible. So what is? When asked about what is still not being talked about enough, Yampolskiy answered that the possibility that this superinteligence would engineer a reality where the human being is perpetually tortured without even having the option to die is a possible outcome.
You might ask why, at least I did. What would an AI gain from a human suffering perpetually? Firstly, it is important to remember that is virtually impossible for us to grasp the motivations of a being with an inteligence that is vastly more advanced than ours - think squirels vs humans, it's just out of our ball park.
But if we were to rummage within our own minds, it is reasonable to assume that it could be an interesting experiment for an amoral being - notice that I say amoral, which is different than defining it as evil/ sadistic.
It might consider that studying how a human being responds to extreme suffering, how he/she adapts, evolves, if consciousness survives or reconfigures etc. could make for extremely interesting data.
Following this chain of thought it would not necessarily give the human subject the possibility to die, considering that the study of the evolution process would be then rendered futile.
This is extremely scary stuff. It's a whole new rabbit hole I have been going through since about 2 days ago and I was just wondering what you guys thinkof this.
