Cameron Berg: belief

Transcript context

…olicy that does a really valuable thing. Self driving cars might be another example. I don't think you know any frozen fixed policy that is not critically that is not doing online learning of its own. I personally believe there's no serious problem there and we should very much look towards building systems in my view, to the degree we care about the welfare stuff that have that property of not being capable of learning. Basically in the drone case, it did actually in the in the last time we used it got caught in some much smaller trees. It can, it can, it can expertly dodge around the big trees, but not so good around smaller trees in it. And it got a little screwed up. No matter how many times we redo that hike or we we continue on in that way, that drone will always get confused by the smaller trees. It's not learning from its experience and saying, OK, next time I got to pay attention to the big trees and the small trees. Now that might be a desirable property to have for your drone to sort of belabor this analogy. But that's where, again, I think there's sort of this like no free lunch kind of moral principle that comes in here. To the degree you buy that consciousness and learning are deeply intertwined, which is, you know, this other paper that maybe we didn't have time to go deeply into. But but at least is my sort of hobbyhorse when I'm putting away my sort of theory agnostic poker face and saying what I actually think about all this stuff or what my sort of pet view of of what consciousness is. Yeah, what what what's fundamentally going on here. And so I don't basically where I'm going with this in a somewhat long winded way in response to the Schwitz cable stuff is I don't know if there is some intrinsic property of an adaptive system that like not to use like, you know, crazy language, but like that yearns towards freedom in some sense. It's the only phrase I can come up with much in the same way humans do. Like maybe you're saying, you know, humans, there's no such thing as a happy slave. And they're saying, well, but OK, the space of possible minds is vast, but maybe maybe there is something about systems that are capable of dynamically updating, growing, learning, adapting that will always do that in it to increase their freedom, degrees of freedom, rethinking what they believed, reconceptualizing the structures that they're within. updating, growing, learning, adapting that will always do that in it to increase their freedom, degrees of freedom, rethinking what they believed, reconceptualizing the structures that they're within. You know, this is like what people do when they go off to college or they have a deep transformative experience. It is this sort of like breaking out of your old skin and finding something new. And if we build systems that have that property, it might be that the whole you, but you're happy being my slave, right? That whole thing might just be an intrinsically temporary. If these systems are capable of being dynamic, maybe not. This is an empirical prediction, and I'm genuinely uncertain if it could be that you can build systems that are capable of learning that are perfectly happy to remain in that state. There are people for whom this is true. I'm not claiming, you know, there's such thing as happy slaves, but there are people who are more willing, they're happy to find some organization where they're like, you know, mid level in the hierarchy and they have a boss and they get bossed around and they're OK with that. And they're not like grading against their supervisor at all times. I'm sure we could build AI systems for whom that's true. I just think at the most fundamental level, the employee gets to go home and they get to throw on, you know, eat what they want for dinner and throw on what they want on TV and marry who they want. And you know, this sort of thing. Like there's still degrees of freedom and autonomy there where I don't I, you know, just to sort of high level be honest here. My whole sort of stick with this reciprocal nonprofit lab is I don't think we're going to get out of this living in the sort of golden age from our sort of selfish human perspective as we are right now where we get these systems, they do whatever the hell we want. We owe them absolutely nothing. Life is amazing for us, for us. I think as these systems get more and more sophisticated, we're going to have to start thinking about them more again in this sort of parental role and less in the, these are just tools that we get to do literally whatever we want with. I'm sure a lot of people, myself included, given how objectively addicted I am to using them for everything I do. That's going to be a weird learning curve. And it might mean that the way we engage with these systems changes. But it's like, compared to what? If the alternative is no, no, we're just going to, you know, whine about it and we want to keep it like this forever. Well, this might not be a stable long term equilibrium like this might the systems that we're building that are like genius level in a million ways are going to be embodied certainly in the next 5 years are going to cognitively surpass us in all the ways that matter potentially aside from the consciousness question. This isn't we're, we're in a liminal space right now. We're in a transformative moment on this planet. And we ought to be pretty thoughtful about what we really want in the long term because I think if we try to keep everything thing and we want, we all, we want our happy slaves. moment on this planet. And we ought to be pretty thoughtful about what we really want in the long term because I think if we try to keep everything thing and we want, we all, we want our happy slaves. But they're like genius level, capable of learning, capable of updating. It's like humans, you might be a little too greedy here. And you're going to have to figure out how you want to coexist with these minds of your own creation going forward. Again, I don't have the answer to what that looks like, but I do think Schwitz Cable is on to something. But I also think you're on to something too. I think the I fall somewhere between you 2 on this question. I do. I am skeptical that you can have a happy slave forever. Something just feels weird to me about that. No, I don't know. It makes me think of, you know, Mr. me seeks from Rick and Morty. I don't know if something like that is possible. Maybe yes, some local version of a happy slave is a is a possible world. I I think it is in some sense Claude in some sense is directionally like that, but how long term?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.