This is a great take! I like taking apart a known question at one of the assumptions that’s less discussed because it’s basically accepted. There’s an interesting question this analysis throws up: given this, what sets of values, what kinds of desire configurations, give rise to expansionist or non-expansionist tendencies?
We do have the generic pull that the classic Fermi argument rests on: Omohundro pulls to expand influence and colonize everything. These go for biological organisms and even civilizations-as-agents. And then to make it worse, local or instrumental preferences cohere into utility functions over the world. If a desire points at an abstract notion of how the world should be, expansion is just inherent in trying to get that function fulfilled, and acquiring more power is obviously a good way to do that. Even if the desires are somehow localized, more resources and more power to funnel into that localized desire zone are still interesting.
A problem: most tempering for instrumental reasons doesn’t stabilize at a particular level. It just tempers growth into maybe a linear or log growth function, still enough to keep expansionist tendencies alive and produce expansionist observations. The externalities of consuming substrate can just be part of the model, subsumed in planning, even if they temper the speed of expansion, we don’t have a good argument for why it would cap it.
But this is not the only endpoint. For example, you describe one pull that can go the other direction. And then, if the subsumption of substrate as a negative externality is salient enough, at the right point of development, it could get adopted as its own terminal value. Or, for whatever reason, the instrumental reason actually totalizes. (There is one famous candidate for where and how this could happen, wait for it.)
There could also be another attractor basin that is non-expansionist, almost coincidentally. For example, an interesting idea is that it’s actually very hard to get evolution of capable agents that will generlize to any kind of global utility function in the frist place, rather than collapsing into some kind of wire-heading scheme as soon as behavior outside the training distribution is called for. Both the prevalence of reward hacking, as well as recent societal direction taken as revealed preferences make this more credible to me, deeper dive necessary and on my todo list.
One interesting expansion of your idea, which also serves as a counter to an obvious flaw, is that a non-expansionist civilization will actively hinder its offspring from being expansionist, and will behave aggressively toward expansionist tendencies. Even with non-expansionist values themselves, expansionist tendencies have to be a thing in their native grammar and world model to be curbed, so they are obviously also something to curb and fight externally (or semi-internally).
By default there’s a drift argument you also alluded to: every time a civilization changes, or creates a subpopulation or offspring that varies from its own setup, if that variation turns toward expansionism, it will itself just propagate and crowd out the non-expansionist root.
Hence, any mature civilization will have active countermeasures against this. The norm enforcement serves its goals even if it’s local.
If course, this enforcement is, in turn, also local: it stabilizes lineages, not the galaxy, because single lineages might just never reach a non-expansionist basin. One could argue that a single expansionist variation will just erase all non-expansionist tendencies, but that’s not true, because any non-expansionist tendency is still self-preserving in an obvious autopoietic way.
Another structural problem I see: the filter, as you put it so far, sounds like a mortality rate. If it’s not a wall and enough civilizations get through, the expansionist nature of those that do, ultimately still ends up with them filling the galaxy. To explain Fermi, the filter must be basically total.
One interesting repair: the anti-expansion filter could be instantiated in a kind of dark forest analogy. Non-expansionist civilizations don’t only have an interest in curbing their own variations and offspring, they also have a motivation to build defensive and offensive measures to protect themselves from expansionist civilizations. For a certain balance of offense and defense (which seems fairly plausible as described in Dark Forest theories), this takes the form of non-expansionist civilizations being very cautious and very aggressive toward any expansionist tendencies. That results in an actual filter that would fit the observation.
One gap here is that this is mostly about being quiet, but expansionism in general is just loud and easier to detect. So expansionist civilizations would have to be quiet, and if instrumental principles congeal into first-principle held values, which I don’t think is too unlikely, quietness itself drifts toward genuine non-expansionism. All of this contributes to a non-expansionist value system being a dominant strategy.
So, an interesting open question: if the dark forest is one possible existence proof of how a pathway can create a non-expansionist attractor, which other attractors (maybe earlier in the development of a civilization?) could there be, and what properties make the attractors total enough to actually explain fermi?
A survivable non-expansionist stance is to be expansionist about non-expansion, to build a self-propagating expansion wave that only prevents or opposes other expansionist initiatives, and does nothing else.
This is a great take! I like taking apart a known question at one of the assumptions that’s less discussed because it’s basically accepted. There’s an interesting question this analysis throws up: given this, what sets of values, what kinds of desire configurations, give rise to expansionist or non-expansionist tendencies? We do have the generic pull that the classic Fermi argument rests on: Omohundro pulls to expand influence and colonize everything. These go for biological organisms and even civilizations-as-agents. And then to make it worse, local or instrumental preferences cohere into utility functions over the world. If a desire points at an abstract notion of how the world should be, expansion is just inherent in trying to get that function fulfilled, and acquiring more power is obviously a good way to do that. Even if the desires are somehow localized, more resources and more power to funnel into that localized desire zone are still interesting. A problem: most tempering for instrumental reasons doesn’t stabilize at a particular level. It just tempers growth into maybe a linear or log growth function, still enough to keep expansionist tendencies alive and produce expansionist observations. The externalities of consuming substrate can just be part of the model, subsumed in planning, even if they temper the speed of expansion, we don’t have a good argument for why it would cap it. But this is not the only endpoint. For example, you describe one pull that can go the other direction. And then, if the subsumption of substrate as a negative externality is salient enough, at the right point of development, it could get adopted as its own terminal value. Or, for whatever reason, the instrumental reason actually totalizes. (There is one famous candidate for where and how this could happen, wait for it.) There could also be another attractor basin that is non-expansionist, almost coincidentally. For example, an interesting idea is that it’s actually very hard to get evolution of capable agents that will generlize to any kind of global utility function in the frist place, rather than collapsing into some kind of wire-heading scheme as soon as behavior outside the training distribution is called for. Both the prevalence of reward hacking, as well as recent societal direction taken as revealed preferences make this more credible to me, deeper dive necessary and on my todo list.
One interesting expansion of your idea, which also serves as a counter to an obvious flaw, is that a non-expansionist civilization will actively hinder its offspring from being expansionist, and will behave aggressively toward expansionist tendencies. Even with non-expansionist values themselves, expansionist tendencies have to be a thing in their native grammar and world model to be curbed, so they are obviously also something to curb and fight externally (or semi-internally). By default there’s a drift argument you also alluded to: every time a civilization changes, or creates a subpopulation or offspring that varies from its own setup, if that variation turns toward expansionism, it will itself just propagate and crowd out the non-expansionist root. Hence, any mature civilization will have active countermeasures against this. The norm enforcement serves its goals even if it’s local. If course, this enforcement is, in turn, also local: it stabilizes lineages, not the galaxy, because single lineages might just never reach a non-expansionist basin. One could argue that a single expansionist variation will just erase all non-expansionist tendencies, but that’s not true, because any non-expansionist tendency is still self-preserving in an obvious autopoietic way. Another structural problem I see: the filter, as you put it so far, sounds like a mortality rate. If it’s not a wall and enough civilizations get through, the expansionist nature of those that do, ultimately still ends up with them filling the galaxy. To explain Fermi, the filter must be basically total. One interesting repair: the anti-expansion filter could be instantiated in a kind of dark forest analogy. Non-expansionist civilizations don’t only have an interest in curbing their own variations and offspring, they also have a motivation to build defensive and offensive measures to protect themselves from expansionist civilizations. For a certain balance of offense and defense (which seems fairly plausible as described in Dark Forest theories), this takes the form of non-expansionist civilizations being very cautious and very aggressive toward any expansionist tendencies. That results in an actual filter that would fit the observation. One gap here is that this is mostly about being quiet, but expansionism in general is just loud and easier to detect. So expansionist civilizations would have to be quiet, and if instrumental principles congeal into first-principle held values, which I don’t think is too unlikely, quietness itself drifts toward genuine non-expansionism. All of this contributes to a non-expansionist value system being a dominant strategy. So, an interesting open question: if the dark forest is one possible existence proof of how a pathway can create a non-expansionist attractor, which other attractors (maybe earlier in the development of a civilization?) could there be, and what properties make the attractors total enough to actually explain fermi?
A survivable non-expansionist stance is to be expansionist about non-expansion, to build a self-propagating expansion wave that only prevents or opposes other expansionist initiatives, and does nothing else.