One thing this makes more rather than less surprising is our early place in the universe's history. The marginal, singleton-sentient observer universe is less likely to produce its only observers near the very beginning, while a universe that produces many observers produces its earliest ones earlier in its history. It seems suspicious that we find ourselves so early within the Stelliferous Era, enough so that this is one of the puzzles I'd expect any speculative cosmology to have some kind of interesting explanation for.
Our position early in the Stelliferous Era may be less surprising than it first seems. Red dwarves are likely unsuitable for complex life as they have a pre-main sequence phase lasting quite some time during which they are notably hotter. Any planet in the habitalble zone for the main sequence phase will be baked for hundreds of millions of years during the pre-main sequence phase and could easily lose all its water.
Though the setelliferious era will go on for quite some time, star formation will slow dramatically (and in fact has already slowed). If you exclude red dwarves and weight by star-years, we're early but not nearly so early as a pure accounting of years suggests.
I didn't anticipate this as a factor at all. My off-the-cuff counterargument would be that the actual habitable portion of the Stelliferous Era may be much shorter than we would think, given that the relative abundance of elements and what type of star you have depends on how early or late you are. For niche example, the proportion of radioactive elements and therefore heat you have depends on where you are in the "birth order" of planets, and if your planet is relatively cold for this reason it can no longer be earth-like. I could try to put together the same idea for the biochemistry, since the normal organic elements will vary too, but the difficulty there is less obvious to me.
Or, a Singleton tends to make the universe uninhabitable fairly soon after this early point in the habitable era. era.
Cheery thought.
Because most singletons don't care to make us a part of them?
Alternately, we are. One thing I'd want to know as a singleton is how often the average sapient species makes its ASI aligned to it (so I can better prepare to meet them) I might run a bunch of sims of random likely species trying to solve alignment to make a guess. I'd compress those sims by mostly modeling the relevant minds: alignment researchers. I'd fake everything else for efficiency.
If I were aligned to be nice to sapients, I'd make most of the suffering in those sims fake. If I weren't, I probably still would, just for efficiency.
This is the form of the simulation argument that makes the most sense to me.
The obvious pessimistic thought is that we are and our successors are locusts/goop/minimally sentient replicators.
I had similar idea, which I formulated as: most observers live in non-perfectly fine-tuned universe which means smaller density of observers per unit of space. The same way the biggest part of Sun's mass is located not in the region of the highest density in its center but in some less dense regions.
the fraction of configurations capable of producing complex chemistry, stable stars, and long-lived planets is extraordinarily small. The fraction capable of producing sentient technological civilizations is smaller still. This gives us a distribution of expected civilizations per configuration that is overwhelmingly concentrated at zero, with a thin tail of configurations that produce any sentience at all
This is a strong assertion that requires justification. How much more fine-tuning does it require to get a universe where 1 of every 1000 stars has a planet that can support intelligent life than a universe where 1 of every 10^30 stars does?
(A3) The fine-tuning results from cosmology extend in the relevant way: the viable region of parameter space does not merely shrink as we add requirements for habitability, but shrinks fast enough that the tail of the distribution is thinner than linear.
This assumption is rather underspecified; what does it mean to add requirements for habitability, and how does that translate the number of habitable (or intelligent-life-supporting) planets?
The Argument in Brief
Sandberg, Drexler, and Ord (2018) showed that the Fermi paradox dissolves once we take our uncertainty about the Drake equation's parameters seriously: the silence of the cosmos is unsurprising given what we actually know. This essay argues that their result is not a contingent fact about our particular universe but a generic prediction. Under a simple multiverse model, most sentient observers in most possible worlds should expect to find themselves alone.
The argument runs as follows. Assume a multiverse in which every possible physical configuration is instantiated, weighted roughly uniformly. From the fine-tuning literature, we know that the fraction of configurations capable of producing complex chemistry, stable stars, and long-lived planets is extraordinarily small. The fraction capable of producing sentient technological civilizations is smaller still. This gives us a distribution of expected civilizations per configuration that is overwhelmingly concentrated at zero, with a thin tail of configurations that produce any sentience at all. Observation selection guarantees that we find ourselves somewhere in that tail, but it does not guarantee that we find ourselves deep in it.
If the tail thins faster than linearly (that is, if configurations producing N civilizations become rarer faster than N grows), then even under observer-weighted reasoning, the typical observer inhabits a universe where sentience is rare. The expected number of technological civilizations in such a universe is small, is probably exactly one. The silence of the cosmos, on this account, is not a puzzle to be solved but a generic prediction of the model.
This argument depends on several assumptions, which should be stated plainly.
(A1) A multiverse of the relevant kind exists.
(A2) Physical configurations within it are weighted approximately uniformly, or at least not in a way that overwhelmingly favors sentience-producing configurations.
(A3) The fine-tuning results from cosmology extend in the relevant way: the viable region of parameter space does not merely shrink as we add requirements for habitability, but shrinks fast enough that the tail of the distribution is thinner than linear.
(A4) One of the standard observation-selection frameworks (SSA or SIA) applies.
If any of these assumptions is wrong, the conclusion may not follow.
The Setup
Consider a multiverse in which every possible physical configuration, meaning every combination of fundamental constants, laws, and initial conditions, is realized. Assume that each configuration is instantiated roughly equally often. The assumption here is that no particular class of configuration is overwhelmingly favored.
Each configuration has some expected number of sentient technological civilizations it produces over its lifetime. Call this E[N]. We are interested in the distribution of E[N] across configurations.
The Distribution of E[N]
From decades of work on fine-tuning in physics, we know that the region of parameter space compatible with complex chemistry, stable stars, and long-lived planets is extraordinarily small. The region compatible with abiogenesis is smaller still. The region compatible with the full chain from abiogenesis through multicellular life to sentient technological civilization is smaller again.
This gives us a distribution of E[N] that is overwhelmingly concentrated at zero. The vast majority of configurations produce no sentience whatsoever. They lack stable atoms, or chemistry, or stars, or planets, or simply any viable path from matter to mind. A thin tail of configurations has some small positive E[N]. A thinner tail still has large E[N].
The qualitative claim here, that the habitable region of parameter space is very small, is well established and essentially uncontroversial in physics. The quantitative claim that this argument requires is stronger: that the density of configurations drops faster than linearly as E[N] increases. This is plausible on the grounds that a configuration producing more civilizations requires more of its parameter space to be viable for life, so that each additional increment of E[N] imposes an additional constraint on the parameter space, compounding multiplicatively to produce a roughly exponential shrinkage. The argument here depends on the shape of a distribution that we can only estimate roughly.
Observation Selection
We exist. This tells us that we do not inhabit one of the sterile configurations. But it does not tell us which non-sterile configuration we should expect to inhabit.
There are two standard frameworks for reasoning about this, both formalized by Bostrom (2002). Under the Self-Sampling Assumption (SSA), we reason as if we are randomly drawn from all observers in the multiverse. Under the Self-Indication Assumption (SIA), we weight each configuration by the number of observers it contains, so that configurations with more observers are proportionally more likely to be ours.
SIA is sometimes taken to favor finding ourselves in a universe rich with life. But this only follows if the distribution of E[N] has a sufficiently heavy tail. To see why, consider what SIA does: it reweights each configuration by its total number of observers. If we assume the number of observers scales roughly with the number of civilizations, then SIA multiplies the prior probability of each configuration by something proportional to N. This makes high-N configurations more likely to be ours. But if the density of configurations drops faster than 1/N as E[N] increases, then the SIA reweighting by N is not enough to compensate for the rarity of those configurations. The product of "N times the density at N" still shrinks as N grows. Given the distribution described above, where each additional increment of E[N] imposes compounding constraints on parameter space, this appears to be the case, though the conclusion is only as strong as our estimate of the tail's shape.
Under either SSA or SIA, then, the typical observer plausibly finds themselves in a configuration drawn from the low-E[N] tail: a universe where sentience is possible but deeply improbable, and where it happens exactly once. This conclusion is robust to the choice between SSA and SIA, though it is not robust to all possible choices of measure over the multiverse.
The Fermi Conclusion
In such a universe, the expected number of technological civilizations is small. If the distribution of E[N] among non-sterile configurations is approximately continuous and concentrated near zero, then observation selection, which conditions on at least one civilization existing, places us in a configuration where E[N] is just large enough to make that likely. The expected number is therefore on the order of one. The expected number of simultaneous technological civilizations is smaller still, since even that small number must be spread across cosmic time.
This provides a resolution of the Fermi paradox that does not require any special mechanism. We do not need to explain why a seemingly hospitable universe is empty. The universe is not particularly hospitable. We are the product of a configuration that barely permits sentience, and we should not expect company.
It is worth being precise about what this argument does and does not achieve. Sandberg, Drexler, and Ord showed that, given our actual uncertainty about the parameters governing life in this universe, we should not be surprised to find ourselves alone. Their argument is epistemic: it is about what we should expect given what we know. The argument here is structural: it claims that the distribution over possible physical configurations, combined with observation selection, generically produces universes in which their result holds. If this is right, the Sandberg et al. finding is the expected outcome across the multiverse.
Other frameworks, including single-universe models with early Great Filters, also predict the Fermi observation. The claim here is not that the multiverse explanation is uniquely correct, but that it is sufficient: if you accept the assumptions, the silence follows, and no further explanation is needed. It is consistent with any given universe having extremely sharp filters, because this is what a universe with low but not zero E[N] should look like.
Given these assumptions, a silent universe should be the generic prediction.