To the editors:

How common are civilizations as advanced as ours in the universe? To estimate this, one common approach compares the timing of key events in Earth’s history to a simple model wherein an initially lifeless Earth must progress through a sequence of steps to reach our current level. Each step has a constant chance of success per unit of time, and all steps must be finished before the window for habitability on Earth comes to an end. If the expected time to complete each step is much longer than the habitability window, then the chance of success within that window is low, and goes as time to the power of the number of steps.

Conditional on success, the durations between steps are all drawn from the same distribution, even when individual steps have very different difficulties. Much literature has pursued this approach and found Earth’s history to be roughly consistent with this model, with roughly two to ten steps. A recent paper in this literature by Andrew Snyder-Beattie et al., concludes that “the expected transition times [for each step] likely exceed the lifetime of Earth, perhaps by many orders of magnitude.”¹

In his review, Charles Lineweaver seems to accept the claims of Snyder-Beattie et al., and of this larger literature, regarding their estimates of a low chance for humans to have reached our current level on Earth by now. But he rejects this estimate as a basis for concluding that “intelligent life is rare in the universe.” Instead Lineweaver suggests that without good reasons to think “the major transitions that characterize our evolution happen elsewhere,” estimates regarding Earth do not allow us to make estimates regarding other planets.

On the contrary, I see two ways to compare planets so that Earth estimates become relevant for other planets, allowing us to infer a low overall rate at which advanced life appears elsewhere. First, if Earth is a random sample from planets that succeed in making life at our level, the success rate on Earth cannot be too different from the typical success rate on other such planets. Second, if there is a substantial chance that our descendants will soon become very visible in the universe, the fact that no other star in our galaxy has yet done so can set a low upper bound on the fraction of such stars that can have reached our level by now.

Allow me to elaborate. Let R be the chance of life at our current level—i.e., controlling nuclear power and practicing spaceflight—appearing on a particular planet within some fixed planet habitability duration. If we assume that R varies across planets according to some distribution, the fact that Earth succeeded suggests we can treat Earth as a random sample from a weighted version of that distribution, where each planet is weighted according to its R value. In this distribution, Earth’s R is unlikely to be greatly different from the median of R. This premise gives us a way to infer the value of R for other planets from that of Earth.

For example, let R vary across planets according to a lognormal distribution with parameters μ, σ. Yes, the distribution must be modified to maintain R < 1 always, but assume that almost none of this distribution is found above R = 0.5. In this case, the distribution of R as weighted by R is also lognormal with the same sigma σ. But instead of having median exp(μ), its new median is exp(μ + σ²). And for any μ, σ normal distribution, only one in 500 samples falls outside of the range [μ – 3.09σ, μ + 3.09σ].

From these elements, there is only a one-in-five-hundred chance that Earth’s R value fails to satisfy this equation:

exp(μ + σ² – 3.09σ) < R < exp(μ + σ² + 3.09σ).

Knowing Earth’s value of R would thus constrain our estimates of the μ, σ parameters describing how R is distributed across planets. That is, knowing the chance that advanced life would have appeared on Earth by some duration does indeed tell us about that chance for other planets.

The other way to use Earth data to constrain the rate at which advanced life appears on other planets is to use an estimate of the chance Q that, within the following ten million years, a planet at our level would give rise to a civilization that becomes permanently visible across its entire galaxy. We might estimate Q > 10^–6—i.e., a chance of at least one in a million. This seems to me a quite conservative estimate of humanity’s future chances to achieve such a large and visible impact.

We clearly observe now that no other star out of the ~2 × 10¹¹ in our galaxy has so far given rise to a permanently visible civilization. If P is the average chance of a star in our galaxy having a planet that reached our level by ten million years ago, then the expected number of such visible civilizations in our galaxy would be E = 2 × 10¹¹PQ. If the lack of visible alien civilizations allows us to conclude that E < 10³, we can infer that P < 5 × 10^–3. That is, life at our level is unlikely at a random star.

One might similarly estimate that no galaxy within the nearest hundred thousand galaxies—out of the roughly two trillion galaxies in the observable universe—has yet given rise to a permanently visible civilization, and that each one has at least a chance Q > 10^–6 of doing that. We can then infer that P < 5 × 10^–6. If we define a galaxy to have a mass of more than a million Suns, our galaxy has about one hundred times as many stars as the average galaxy.

Lineweaver suggests that the chance that advanced life would have appeared on Earth says nothing about the analogous chances for other planets. In this letter, I have shown two different relations between such chances. First, as Earth succeeded in creating life at our level, the chances for Earth to do so cannot be much different from the typical chance of planets that succeed. Second, as no other planet in the galaxy has surpassed our current level to become clearly and widely visible, a lower bound on the chance of that future transition implies an upper bound on the average chance that other planets have ever reached our current level.

Robin Hanson

Charles Lineweaver replies:

Robin Hanson and I disagree about a fundamental assumption. Hanson writes: “Lineweaver suggests that the chance that advanced life would have appeared on Earth says nothing about the analogous chances for other planets.” I don’t believe in the general group that he and many others call “advanced life.” Here is why. Our species, Homo sapiens, is the only species on Earth that is supposed to belong to this group. But if we Homo sapiens belong to any group general enough to include extraterrestrial members, our closest relatives on Earth should also belong to that group. Most biologists and I believe that our closest relatives in the universe are here on Earth. Our reading of the many-branched tree of life on Earth is that our evolutionary path has been so quirky and convoluted—so historically contingent, to use Stephen Gould’s phrase²—that no other life-forms in the universe will be genetically or phenotypically more similar to us than chimps, bonobos, gorillas, naked mole rats, or frogs.

Since Hanson and many others exclude our closest relatives from “advanced life,” they are—by their definition—not talking about a generic group with other members. They have defined “advanced life” as a single species. To me, this makes as much sense as talking about the group of all baseball-playing Japanese-speaking quadrupeds in the universe. Strong arguments can be made that these are non-existent groups.³

As an example of how important to the argument this generic versus specific assumption is, below I have modified one of Hanson’s paragraphs. In it, Hanson contrasts the general group of life at our current level on Earth with the level of life on other planets. I have modified the paragraph to show that the argument falls apart when the supposedly general group is made more specific.

Let R be the chance of life at our current level humans—i.e., controlling nuclear power and practicing spaceflight—appearing on a particular planet branch of the tree of life within some fixed planet habitability duration. If we assume that R varies across planets species according to some distribution, the fact that Earth humans succeeded suggests we can treat Earth humans as a random sample from a weighted version of that distribution, where each planet species is weighted according to its R value. In this distribution, Earth humans’ R is unlikely to be greatly different from the median of R. This premise gives us a way to infer the value of R for other planets species from that of Earth humans.

From this modified analysis, the R associated with other species is 0 since the experiment has already been run for tens of millions of years on the independent continents where there were no human ancestors. The evolutionary history of life on Earth strongly suggests that there is no evolutionary pressure to obtain human-like intelligence. Other species do not want to become humans any more than we want to evolve into oak trees.⁴ Among all the independently evolving species on Earth, humans are the only ones who have become humans at our level of technology. To then conclude that among all species, our species had an average chance of becoming humans at our level is meaningless. Our R = 1, while R = 0 for all other species. The average of 0, 0, 0, 0, 0, … is not 1.

We could also replace “humans” with “Asian elephants”: What is the probability of Asian elephants evolving on another planet? Well, based on the fact that they only evolved once on Earth, there does not seem to be any convergence of being an Asian elephant. I suggest that our Asian elephants are the only Asian elephants in the entire universe. And if there is a more general set that the Asian elephant belongs to, its closest relative here on Earth will also belong to that set.

The way around this logic has been championed by Simon Conway Morris, who argues that strong selection pressure leads to convergent evolution which then produces human-like intelligence.⁵ Hanson and most physicists subscribe to this view, but most biologists and I don’t. There is an important debate ongoing in astrobiology about whether such convergent evolution is best explained by a strong universal selection pressure to produce human-like intelligence or by shared features of common ancestors, also known as deep homology. If selection pressure is the explanation, we should expect humanoids or “advanced life” or “intelligent life” to evolve all over the universe. I call this the Planet of the Apes Hypothesis.⁶

At issue is whether it makes any sense to think that the life-forms on other planets, if there are any, have been and are being selected to be like humans and have human technology. Hanson summarizes my position thusly: “Lineweaver suggests that the chance that advanced life would have appeared on Earth says nothing about the analogous chances for other planets.” A better summary is this: Lineweaver suggests that the fact of a particular species having evolved on Earth does not suggest the species will evolve on another planet.

Hanson refers to whether “Earth is a random sample from planets that succeed in making life at our level [emphasis added].” Whenever faced with the language of “levels” in the context of biological evolution, I try to eliminate my natural human bias and ask: If we exclude our species from consideration, does this talk of levels make any sense when applied to the rest of life? Are dogs or red oak trees at a higher level? Are sparrows or frogs at a higher level? Are C. elegans at a higher level than E. coli? Such a projection of the rich variety of life on Earth onto a one-dimensional scale of levels is not scientifically useful. We are still working at removing racism from science. Removing speciesism will be harder.