Oh, thank you! I thought that what doesn’t exist was a list of initial conditions that we could already work on. I didn’t expect that there is nothing at all, even far fetched. So, if I understand it correctly, for all the proposals so far developed, there have been someone suggesting a credible way an AGI could doge them. Do I understand it correctly?
To the best of my knowledge yes, though I am not an alignment researcher and there may be some proposals that could in fact work, and there are definitely some that will be obscure enough that the only people who know about them also believe they will work.
As far as I know, there aren’t any proposals that are generally believed to work. The field is far too young for that. We don’t yet even have a good handle on what the problems will be in practice, which sort of goes with the territory when we need to consider the behaviours of things far smarter than ourselves.
Oh, thank you! I thought that what doesn’t exist was a list of initial conditions that we could already work on. I didn’t expect that there is nothing at all, even far fetched. So, if I understand it correctly, for all the proposals so far developed, there have been someone suggesting a credible way an AGI could doge them. Do I understand it correctly?
To the best of my knowledge yes, though I am not an alignment researcher and there may be some proposals that could in fact work, and there are definitely some that will be obscure enough that the only people who know about them also believe they will work.
As far as I know, there aren’t any proposals that are generally believed to work. The field is far too young for that. We don’t yet even have a good handle on what the problems will be in practice, which sort of goes with the territory when we need to consider the behaviours of things far smarter than ourselves.