>If we managed to pause now, that would greatly increase the chance we paused again later. At the moment, a major obstacle to people supporting this is their senses that it is impossible.
This is the main area where I disagree. The main obstacle is that few people believe AI will be powerful enough to be dangerous soon. That obstacle is now weakening rapidly.
> Building the machinery to pause makes it much easier to pause again.
This is mostly true even if the pause is not enacted. Such machinery would help us come closer to pausing on a dime. It feels urgent to create good plans for verifying a pause, even though I expect the optimal time for a pause is more than a year away.
The "machinery" consists of people doing things. Like, say, Congress or government agencies. A thing that helps them a lot in getting to act is the sense that what they're doing is *normal*. Thus the only way to fully "build" the machinery is to run it -- it's not just that doing a thing once makes it easier *technically* to do it again, since after all the obstacle here is mostly not technical; it's that it makes people more *willing* to do it again, both due to the feeling of normality, and due to the fact that they now fully know how. How would you build such machinery without running it? Draft a bill or regulation, get Congress or the relevant agency to agree it's a good idea, then have them *not* pass it, get them to save it for later? Political will doesn't work that way! It is quite difficult to coordinate people to that level, and if the effort doesn't go through to completion, it's largely going to waste, because by the time you want to deploy it, it's going to be a different set of people you need to convince.
>If we managed to pause now, that would greatly increase the chance we paused again later. At the moment, a major obstacle to people supporting this is their senses that it is impossible.
This is the main area where I disagree. The main obstacle is that few people believe AI will be powerful enough to be dangerous soon. That obstacle is now weakening rapidly.
> Building the machinery to pause makes it much easier to pause again.
This is mostly true even if the pause is not enacted. Such machinery would help us come closer to pausing on a dime. It feels urgent to create good plans for verifying a pause, even though I expect the optimal time for a pause is more than a year away.
The "machinery" consists of people doing things. Like, say, Congress or government agencies. A thing that helps them a lot in getting to act is the sense that what they're doing is *normal*. Thus the only way to fully "build" the machinery is to run it -- it's not just that doing a thing once makes it easier *technically* to do it again, since after all the obstacle here is mostly not technical; it's that it makes people more *willing* to do it again, both due to the feeling of normality, and due to the fact that they now fully know how. How would you build such machinery without running it? Draft a bill or regulation, get Congress or the relevant agency to agree it's a good idea, then have them *not* pass it, get them to save it for later? Political will doesn't work that way! It is quite difficult to coordinate people to that level, and if the effort doesn't go through to completion, it's largely going to waste, because by the time you want to deploy it, it's going to be a different set of people you need to convince.