Candidate properties of civilizations that will dominate the Universe

Following other contributions to the literature, we discuss scenarios for the future of intelligence in the Universe in terms of candidate properties of ‘civilizations’. Any such civilization is assumed to have one (potentially several) planet(s) (or moon(s)) of origin, a clearly defined duration of existence, and a cosmic volume that it inhabits at any given time. We preface our discussion of candidate properties of such civilizations with two caveats:

• • While ‘civilizations’ may always originate from life in the sense of biology as a science, they may at later point be made up of engineered objects such as ‘artificial intelligence systems’, potentially long after biological life has vanished.

• • Life in the broad sense of this paper may spread in such a way that the characterization in terms of distinct civilizations becomes inapplicable. In what follows, we explore scenarios that are based on civilizations with distinct numerical identities and clearly identified volumes that they inhabit, but we acknowledge that scenarios which are not based on civilizations in that sense may also be considered and studied.

In this context, human-instigated society counts as one civilization (but see the end of the ‘Conclusion’ section for critical discussion). We are interested in civilizations that are prime candidates for being the long-term most prevalent ones in the Universe. Accordingly, it does not matter so much which properties any arbitrary civilization, born on some random habitable planet, is likely to have. The future of the Universe will be shaped, in particular, by those civilizations, if there are any, that spread most far and wide. This motivates considering civilizations with the following two properties:

• • FAST civilizations perpetually expand close to a maximum velocity of expansion, limited only by hard physico-technical constraints.

• • UNDYING civilizations last long by the standards of cosmic timescales and, in particular, do not disappear before becoming or encountering other FAST civilizations.

In the long term, civilizations that lack either FAST or UNDYING will populate comparatively far smaller volumes than civilizations with these properties, unless they are far, far more numerous. Cosmic selection, in other words, will favour civilizations that are FAST and UNDYING (provided they exist). Olson makes the case for the assumption, encoded in FAST, that there is a (near-) universal maximum expansion velocity:

Interstellar travel and intergalactic travel in particular may, as argued by Drake and Sobel (Reference Drake and Sobel1992), forever remain impractical or even outright impossible for biological organisms such as us. However, a civilization's expansion in space could be driven or spearheaded by self-replicating von Neumann probes (Von Neumann, Reference Von Neumann1966; Tipler, Reference Tipler1994; Matloff, Reference Matloff2022). Therefore, its expansion speed need not be constrained by biology-related factors. According to Armstrong and Sandberg (Reference Armstrong and Sandberg2013), harnessing only a tiny share of the energy and materials in any given Galaxy would be sufficient for a civilization to fuel its intergalactic travel.

Due to the fundamental limit on the speed of communication in our Universe, rapidly expanding civilizations may break up into different civilizations, perhaps not all striving for further rapid expansion (as has been discussed in the context of percolation models as a potential answer to the Fermi paradox; see Landis, Reference Landis1998 and Galera et al., Reference Galera, Galanti and Kinouchi2019). However, as long as a fraction of the descendant civilizations keep expanding, due to selection effects, the model of Hanson et al. (Reference Hanson, Martin, McCarter and Paulson2021) still captures the main dynamics.

According to Olson as well as Hanson et al., civilizations that will be most widespread can be expected to also have a third property, namely:

• • LOUD civilizations make clearly visible changes to the volumes that they populateFootnote ¹.

Hanson et al. (Reference Hanson, Martin, McCarter and Paulson2021: 6) suggest that making drastic changes to volumes may be necessary to extract the resources needed to enable fast expansion, a suggestion that they see as supported by human history. If this is true, being LOUD is indirectly favoured by the same selection effect as FAST. However, as we argue in Section ‘Can GC really explain our earliness?’, there are good reasons to seriously consider not only LOUD but also non-LOUD civilizations, i.e. civilizations that do not make clearly visible changes to the volumes that they populate. Such civilizations might actually be favoured by selection effects over LOUD ones in the longer term.

The distinction between LOUD and non-LOUD civilizations is a matter of degree and depends on how exactly one spells out ‘making visible changes’. Civilizations that can be identified across astrophysical distances due to their technosignatures (Socas-Navarro et al., Reference Socas-Navarro, Haqq-Misra, Wright, Kopparapu, Benford and Davis2021) qualify as LOUD. A paradigmatic visible change to some planetary system would be the construction of a Dyson sphere (Dyson, Reference Dyson1960) around its central star, to harvest a large share of its ‘solar’ energy. Changes made on a planetary scale – such as inducing anthropogenic climate change – may or may not be considered ‘clearly visible’, e.g. one may debate whether humanity is currently LOUD and, if so, when it became LOUD. An extreme example of a non-LOUD civilization would be one that makes changes only on micro-scales and/or in such a way that the changes are inaccessible with current human (and perhaps more advanced) technology. Conceivably, there might be civilizations that are LOUD for some periods or in some regions and non-LOUD in others.

As LOUD civilizations dramatically change the volumes they control they may also interfere with the births of new civilizations in these volumes. A particularly simple type of interference is to prevent any new civilization births, either intentionally or as a side effect:

Evidently, for humanity to appear on Earth, Earth had to be outside any volume controlled by a civilization with the property SUPPRESSIVE.

The power law of civilization births and human earliness

According to Hanson et al. (Reference Hanson, Martin, McCarter and Paulson2021), drawing on an influential argument by Carter (Reference Carter1983), civilizations that will come to dominate the Universe will likely have one further property, beside FAST, UNDYING, LOUD and SUPPRESSIVE, namely, that their collective frequency of births in time is approximately given by a power law. We now review and motivate this property, which we call POWER-LAWED.

Carter's original argument applies to the appearance of intelligent life on any planet that is in principle hospitable to life. According to Carter, the appearance of intelligent life on any such planet plausibly requires passing a number of critical ‘hard steps’. These include ‘try-indefinitely steps’, which can be ‘tried’ indefinitely, and ‘try-once steps’, which only occur at a specific stage of evolution with a particular chance of success.

Let us first consider try-indefinitely steps (or ‘try-try steps’ in the terminology of Hanson et al., Reference Hanson, Martin, McCarter and Paulson2021). The probabilities with which they, differentially, succeed can be expressed in terms of the specific characteristic timescales, t_i, where i ranges over try-indefinitely steps, on which they typically occur. Candidate try-indefinitely steps in the appearance of intelligent life in Earth history include the appearance of life itself, the appearance of eukaryotes, the appearance of sexual reproduction and the evolution of humanity as an intelligent species.

Earth is estimated to become uninhabitable due to deoxygenation of the atmosphere in around 1 billion years (Ozaki and Reinhard, Reference Ozaki and Reinhard2021), and so far we have not found any signs of intelligent life on other planets. We can therefore confidently exclude the possibility that the sum of all timescales t_i associated with try-indefinitely steps is significantly shorter than the total time interval during which Earth is habitable, about 5.5 billion years. For if that sum had been significantly shorter than 5.5 billion years, intelligent life would in all likelihood have appeared orders of magnitude earlier on Earth than it did. Moreover, if the sum of all timescales t_i, for Earth were much smaller than the interval during which Earth is habitable, the same would be true, one might conjecture, for at least some exoplanets in the causally accessible Universe. So the fact that we have not detected any signatures of extraterrestrial intelligence also, at least somewhat, speaks against the sum of all timescales t_i being significantly smaller than Earth's duration of habitability.

At the same time, there is no reason to expect any correlation between the astronomical processes determining planet lifetimes and the biological processes determining the timescales of try-indefinitely steps, t_i. It would be a curious coincidence if the longest timescales, t_i, happened to be of the same order of magnitude as the time duration during which Earth is habitable. As Carter argues, it is rational to expect that at least some try-indefinitely steps have characteristic timescales t_i that are larger by several, perhaps many, orders of magnitude than the habitability duration of planets such as Earth. This means that, on any given planet, the appearance of intelligent life will, in all likelihood, be exceedingly improbable. The fact that we have not detected any signs of intelligent life on other planets fits this hypothesis well.

However, if the total number of habitable planets in the Universe is large enough, intelligent life is bound to appear on a number of planets nevertheless, even if these planets may remain rare compared with lifeless planets. Carter (Reference Carter1983) showed that, if n is the number of try-indefinitely steps with timescales, t_i, that are longer than a planet's habitability duration, the probability P(t) of intelligent life having appeared on it until some time t follows a power law:

(1)$$P( t ) \sim t^n.$$

This formula makes intuitive sense: many ‘hurdles’ have to be taken for intelligent life to appear. When the first planets form, life is still absent completely, and its appearance starts very slowly, then accelerates as the hurdles are taken on more and more planets. The pattern of slow start and steep acceleration is more pronounced as the number n of try-indefinitely steps is increased.

What about try-once steps, which unlike try-indefinitely steps cannot be tried again in case of failure? Try-once steps affect the probability of life having appeared at some point, but they do not change the power-law form of equation (1) if there are also try-indefinitely steps.

Hanson et al. assume that civilizations which will come to dominate the Universe arise randomly from civilizations that start on arbitrary habitable planets such as Earth, by going through one or more additional try-indefinitely steps or try-once steps. If the probability of their birth on any specific planet is determined by a power law of the form (1), with a planet-specific power n, then the overall frequency distribution of GC births across all galaxies can also be approximated by a power law of the form (1) with a suitably chosen n (Hanson et al., Reference Hanson, Martin, McCarter and Paulson2021, Appendix A). This means that GC will, collectively, have the property POWER-LAWED:

Hanson et al. argue that if we follow the line of thought just reviewed and assume that the appearances or ‘births’ of civilizations in time are characterized by a power-law function, our earliness becomes even far more remarkable than previously appreciated. According to the power law, the vast majority of civilizations appear towards the end of the lifespan of their planet of origin, and the larger the number n of hard steps is, the more pronounced this tendency is. Neither the maximal habitability span of in-principle habitable planets nor the correct number n are known. According to the calculations of Hanson et al., even if we assume comparatively short maximal planet habitability durations (not much longer than the habitability of Earth) and a very low number of hard steps (n = 1 or 2), humans are among the first 10% of intelligent civilizations to arise. If longer planet habitability durations and a larger number n of hard steps are assumed, both of which appear plausible in the light of what we know, we look even more dramatically early.

If the appearance of humans is indeed (very) early compared to the cosmic average for intelligent civilizations, we are highly atypical in this respect. Hanson et al. (Reference Hanson, Martin, McCarter and Paulson2021) aim to provide an explanation for our atypical-seeming arrival time, or rather: they stipulate an additional mechanism by which our arrival time is rendered typical after all, namely, they propose the

We are atypically early only under the assumption that civilizations such as ours appear throughout cosmic history on habitable planets, roughly following the power law (1). The situation is radically different if the GC scenario holds and civilizations will soon (on cosmic timescales) control the entire Universe which are FAST, UNDYING, LOUD, SUPPRESSIVE and whose appearance follows a power law. According to Hanson et al. (Reference Hanson, Martin, McCarter and Paulson2021), the GC scenario thus provides an elegant explanation of why we have appeared so early – an explanation which renders our appearance as a civilization on a habitable planet typical in time rather than stunningly early.

Hanson et al. provide a more specific estimate of how the GC scenario will likely come to pass. Due to GCs having the property SUPPRESSIVE, we can rule out being in a volume controlled by some GC. A fortiori, due to SUPPRESSIVE we can rule out that GCs are already controlling the entire Universe. Moreover, due to the assumption LOUD, we would presumably clearly see any area controlled by a neighbouring GC as distinct from its environment, which is not currently the case.

Using a simple three-parameter model and assuming that we have been born at a typical time of civilization birth, Hanson et al. predict that human civilization, if it survives long enough, will most likely first encounter a GC in around 200 million to 2 billion years. Furthermore, they conclude from their modelling that each GC will control between around 10⁵ and 10⁷ galaxies before encountering other GCs at all their boundaries. Finally, if p is the probability of any given civilization to become grabby, unless p is lower than 10⁻⁴ (i.e. unless any given civilization is exceedingly unlikely to ever become grabby) we cannot expect there to have been even one civilization in our own Galaxy's past. Had there been any, the probability of at least one GC having made it to Earth already would have been close to 1. Hanson et al. conclude from this that, in the light of the plausibility of the GC scenario, it is unlikely that SETI-type attempts will succeed, i.e. we are unlikely to ever detect any extraterrestrial (non-grabby) civilization.

In the next section, we criticize the inference to the GC scenario as the alleged best explanation of our earliness based on which Hanson et al. draw these conclusions.

Can GC really explain our earliness?

An obvious worry about the suggested inference to the GC scenario as the best ‘explanation’ of our current existence is that the hypothesized takeover of our Galaxy by some GC, if it takes place, takes place in the future. But, absent any retrocausality, future events cannot cause present events. Therefore, at least by the standards of a causal account of explanation such as the influential one by Woodward (Reference Woodward2003), the GC hypothesis cannot possibly ‘explain’ why we have appeared so early. Hence, there cannot be an argument for the GC scenario based on our earliness in the form of an inference to the best explanation.

Proponents of the GC scenario can react to this observation by arguing that, even though the GC scenario does not explain why we are early, at least not by the standards of a causal account of explanation, it nevertheless makes our earliness expected, unlike other, competing, accounts of the future of life in the Universe. Therefore, the GC scenario gets evidentially boosted by the observation that we are early compared with those other scenarios. This line of reasoning can be reconstructed using the language of Bayesian epistemology. It requires defining self-locating probability distributions P(i | GC, B) and P(i | ~GC, B) over one's own ‘birth rank’ i, as a civilization, conditional on the GC scenario occurring (‘GC’) and conditional on it not occurring (‘~GC’). The variable B denotes suitably chosen background evidence that is common ground for the assessment of both scenarios. Suppose that n and N are the total numbers of civilizations ever to be born according to GC and ~GC, respectively, with n ≪ N, and assume self-locating indifference (Bostrom's (Reference Bostrom2002) Self-Sampling Assumption), namely,

(2)$$P( {i\mid {\rm GC}, \;\,B} ) = 1/n\;\,{\rm and}\;\,P( {i\mid \mathord{\sim} {\rm GC}, \;\,B} ) = 1/N.$$

The evidence that we condition on in this Bayesian reconstruction of the argument given by Hanson et al. (Reference Hanson, Martin, McCarter and Paulson2021) is that we are very early (‘E’) in the history of the Universe, i.e. that our birth rank i is smaller than n. Due to (2) – or, more generally, unless P(i | ~GC, B) is heavily biased towards low values i < n – this finding is far more likely conditional on GC than conditional on its negation, that is:

(3)$$P( {E\mid {\rm GC}, \;\,B} ) \gg P( {E\mid \mathord{\sim} {\rm GC}, \;\,B} ) .$$

As a consequence of (3), unless the prior probabilities assigned to GC and ~GC dramatically favour ~GC – that is, provided that P(~GC | B) is not many orders of magnitude larger than P(GC, B) – the posterior probabilities assigned to GC and ~GC after conditioning on the evidence E ‘We are early’ will favour the GC hypothesis.

Anthropic reasoning and, more generally, reasoning taking into account indexical evidence along the lines just discussed is controversial (Norton, Reference Norton2010, for instance, is very critical) and has raised vivid debates in the Bayesian literature (for instance, in relation to the sleeping beauty toy problem)Footnote ². We are not putting forward (2) and (3) in order to implicitly advocate particular positions in these debates. Our modest goal here is to give a charitable formal reconstruction of the argument made by Hanson et al. (Reference Hanson, Martin, McCarter and Paulson2021).

In the light of our reconstruction of that argument, readers familiar with the Doomsday argument (Carter, Reference Carter1983; Leslie, Reference Leslie1989) will recognize that both these arguments have the same form. In a simple exposition of the Doomsday argument, two hypotheses about the totality of human history (past and future) are compared: hypothesis H _short says that humanity will go extinct relatively soon and hypothesis H _long says that there will be trillions of future humans. Finding ourselves to have a birth rank which is sufficiently early (again, ‘E’) to be compatible with H _short suggests a likelihood inequality analogous to (3), namely,

(4)$$P( {E\mid H_{{\rm short}}, \;\,B} ) \gg P( {E\mid H_{{\rm long}}, \;\,B} ) .$$

Similar to the argument in favour of the GC hypothesis outlined above, the Doomsday argument concludes that – unless H _long happens to have a dramatically higher prior probability than H _short – the posterior probability of H _short is far larger than that of H _long. Accordingly, we should be confident that – especially since a considerable share of humans ever to have lived is alive now and we are not currently experiencing a rapid decline – the future of humanity will be short and Doomsday is imminent.

Many people have the intuition that something must be wrong with the Doomsday argument. Unsurprisingly, there are many criticisms of it in the philosophical literature. Here, we criticize the argument that the GC hypothesis is plausible in the light of the finding that we are early in cosmic history by drawing inspiration from a criticism of the Doomsday argument that accepts the Bayesian framework for studying self-locating belief and that we find particularly insightful (Dieks, Reference Dieks2007; for a recent defence, see Friederich, Reference Friederich2021: Ch. 9). Dieks' core observation is that the background evidence B in (4) must have very specific properties to fulfil its role in the Doomsday argument. We now apply this analysis directly to the argument for the GC scenario, where the background evidence plays a fully analogous role.

On the one hand, the background evidence B in (3) must include enough information about basic physics and cosmology to entail several facts about overall cosmic history – notably, which stars (will) have habitable planets, when, and for how long. On the other, this evidence B must not include any evidence that gives away our own place in overall cosmic history – that is, that we are on a planet orbiting a main sequence star around 13.8 billion years after the Big Bang (according to measurements of the cosmic microwave background; Aghanim et al., Reference Aghanim, Akrami, Ashdown, Aumont, Baccigalupi, Ballardini and Roudier2020). If B did include such evidence, it would entail E, so both P(E | GC, B) and P(E | ~GC, B) would be trivially 1, contrary to (3). It is by no means obvious what should be part of this hypothetical background evidence B. Whatever exactly belongs in this B, it is clearly very different from our actual evidence – the scientific knowledge that humans have accumulated.

The ratio that is ultimately of interest in our reconstruction of the argument by Hanson et al. is the one between the posteriors P(GC | E, B)/P(~GC | E, B). From Bayes' theorem in odds form,

(5)$$ \frac{P( {{\rm GC} \mid E, \;\,B} )}{P( {\mathord{\sim} {\rm GC} \mid E, \;\,B} )} = \frac{P( {E \mid {\rm GC}, \;\,B} )}{P( {E \mid \mathord{\sim} {\rm GC}, \;\,B} )} \times \frac{P( {{\rm GC} \mid B} )}{P( {\mathord{\sim} {\rm GC} \mid B} )} , $$

we see that the posterior odds don't only depend on the likelihood ratio that we discussed thus far, but also on the priors P(GC | B) and P(~GC | B). But since B is so elusive, it is very difficult to see why P(~GC | B) might not be many orders of magnitude larger than P(GC | B). And we should not accept the inference from our earliness to the GC hypothesis based on the likelihood inequality (3) alone, by default, without considering what would be reasonable assignments of priors. For by that standard, we could also run a structurally analogous argument against the GC scenario.

This can be seen as follows: note that, if the GC scenario holds, grabby aliens will be extremely widespread, numerous and exist for a very long time. Accordingly, the ‘typical’ intelligent observer in our Universe will be some member of a GC and exist far later in cosmic history than we do. By reasoning structurally analogous to equations (1)–(3), finding ourselves to be early and not as members of a GC (yet), far from confirming the GC hypothesis, actually disconfirms it.

To spell this out formally, consider background evidence B* which leaves open whether we might be grabby. Conditional on this B* and the GC hypothesis the probability of finding ourselves to be as early as we are will be extremely low, far lower still than the probability of finding ourselves so early conditional on assuming that there will only ever be non-GCFootnote ³. Thus, for such evidence B*, the inequality

(6)$$P( {E\mid {\rm GC}, \;\,B^\ast } ) \ll P( {E\mid \mathord{\sim} {\rm GC}, \;\,B^\ast } ) $$

seems plausible, with the orientation of ‘≪’ now in the opposite direction as in (3). If we could reasonably take (3) alone, without further consideration of B, to establish that our earliness confirms the GC scenario, we could equally well take (6) to establish that our earliness disconfirms the GC scenario.

We conclude that finding ourselves to be early does not make it rational to be confident that the majority of volumes in the Universe will become dominated by GC.

Why the GC hypothesis is intrinsically improbable

As we have seen, proponents of the GC scenario don't merely have to argue that the likelihood ratio favours their hypothesis, they also have to establish that the prior probability of their hypothesis isn't zero or very small, given background evidence B for which (3) holds. The structure of the GC scenario makes that a difficult task: the hypothesis takes the form of a conjunction of at least five separate assumptions (as spelled out in Sections ‘Candidate properties of civilizations that will dominate the Universe’ and ‘The power law of civilization births and human earliness’) and each conjunct makes the GC hypothesis less likely. If merely one of these five assumptions turns out to be unrealizable or exceedingly rare, the prior of the full GC hypothesis is zero or close to zero. So, while the specificity of the GC scenario is a strength for modelling purposes, since it allows Hanson et al. to run quantitative simulations based on their assumptions, it is also a challenge from a Bayesian point of view. In this section, we give reasons to doubt the assumptions from the start and hence to assign a rather low prior to the GC scenario.