Where is everybody?
A civilization that spread from star to star, even slowly, could have crossed the Milky Way many times by now. None seems to have. I built a model of the possible reasons and tested it against public data. The evidence firmly rules out only one of them: that someone has visibly settled our galaxy. Beyond that, the answer depends mostly on what you assume to begin with. On my assumptions, there is about a 73% chance that we are alone in the Milky Way, with no one on the way here from another galaxy. Plug in the assumptions behind other well-known views and that number runs from 13% to 94%.
We’re alone
No other civilization has ever arisen in the Milky Way, and none is heading our way from another galaxy.
That number comes from my own assumptions. I also reran the analysis 400 times with the assumptions varied at random. Those runs mostly came out lower: the typical result was 56%, and 90% of them fell between 26% and 78%.
They never left home
Other civilizations have arisen in the Milky Way, but none has kept spreading from star to star. I assumed the likeliest reason is a barrier that stops every civilization from spreading, including us. (Travel between the stars might simply be impractical.) The evidence can’t confirm that a barrier is the reason.
They spread out
Some civilization has spread between the stars. Either it is still on its way here from another galaxy (about 4%), or it is already here, unnoticed or hiding (about 7%).
On the “our existence proves nothing” view, the three answers come out as: we’re alone at 73%, they never left home at 16%, they spread out at 11%.
This view draws no conclusion from the fact that we exist. It takes Earth as given and asks only what else is out there. The rest of this page uses this view unless it says otherwise.This view treats our civilization as if it had been picked at random from all the civilizations that ever arise. That gives a little more weight to possible universes in which most civilizations look like ours. On my assumptions the result barely moves.This view treats each of us as if picked at random from every individual in every civilization. If spreading between the stars were possible, most individuals would live in civilizations that had spread. We don’t, and that counts against spreading being possible at all.This view says that the more civilizations like ours a universe would contain, the more likely it is to be the real one. A galaxy with nobody else in it is then almost ruled out before looking at any evidence.
No other assumption of mine changes the result as much as this one. Reason as if you were one person picked at random from everyone in every civilization (the third button), and the chance that we’re alone falls to 54%, while the chance that they never left home rises to 45%. Reason that a universe with more civilizations like ours is more likely to be the real one (the fourth button), and the chance that we’re alone drops below 1%. The section What does our own existence tell us? explains why.
Figure 1 How the evidence changed the odds
Three things to know about these numbers
The assumptions are my choices. Taken together, they sit toward the empty-galaxy end of what I think can be defended. (One exception is evolution, where I assume fewer unlikely steps than most recent published estimates.) The range under the first answer comes from picking every assumption at random from a wide range and running the whole analysis again, 400 times. The well-known views in the section It comes down to what you assume reach further in both directions. For someone sure that life starts easily, the leading answer is “they never left home”.
A slightly different question is whether any other civilization has ever arisen in the Milky Way. On the “our existence proves nothing” view, the chance that none has is 78% (77% on a stricter count, explained under “The numbers behind this chart” in the section What the evidence rules out). That is a little higher than the 73% for “we’re alone” because it also counts cases where no one else arose here but settlers from another galaxy are on their way or already here. Those cases come to about 5%. (Bar labels in the chart are rounded one by one, so they can differ from the three answers by a point. The three answers above add up to 100% on each of the four buttons.)
The reason I give in the second answer is one I built in. I gave a 15% chance to a barrier that stops every civilization from spreading. Without that barrier, I did not let lasting expansion be rarer than about one civilization in a hundred million. The evidence can’t tell such a barrier apart from expansion that is just very rare (see how I did this).
What the evidence rules out
“Where is everybody?” started as a remark Enrico Fermi made over lunch in 1950 (Jones 1985). The argument now attached to it goes like this. The Milky Way is more than ten billion years old. It can be crossed in about ten million years at 1% of the speed of light, or in under two billion at the speed of the Voyager probes. So if technological civilizations arise at all often, one of them should have been everywhere long ago.
I built a model with a list of unknowns, from the number of habitable planets to the chance that a civilization keeps spreading. It picks a value for each at random from a wide range, and each set of picks is one simulated universe. Every simulated universe falls under one of seven explanations, and the seven are grouped into the three answers at the top of this page.
Then I asked which universes still fit four observations. The first is the timing of life on Earth: life began early, but we arrived late in Earth’s history. The other three are things we have looked for and not found: no one is visibly here, no galaxy has been found whose starlight someone is collecting for energy, and 60 years of radio searches have heard nothing.
The evidence rules out one explanation, which I call obvious settlement: someone has taken over the Milky Way, planets included, in a way we could not miss. (The charts call it “Here, and obvious”.) My starting assumptions gave it a 36% chance. The evidence leaves it under 1%.
Everything else survives, in almost the same proportions as before. The probability taken from obvious settlement was shared among the other six explanations, roughly in proportion to what each already had. That is why the chance that we’re alone in the Milky Way went from 47% before looking at the evidence to 73% after.
So the numbers at the top of this page are mostly my starting assumptions with one explanation crossed off. Here is a way to see it. Before I looked at any evidence, “we’re alone” had 47% of the total. Leave obvious settlement out of the total, and that 47% becomes 73% of what is left. “They never left home” has 15% of what is left, and “they spread out” has 12%. After the evidence, the same three shares are 73%, 16% and 11%.
Figure 2 Six of the seven explanations fit what we see about equally well. The seventh, obvious settlement, doesn’t fit at all.
Which of the four observations moved the odds most? The next chart answers that for a slightly broader question than “we’re alone”: has any other civilization ever arisen in the Milky Way? If none has, we are the first civilization here. The evidence raised the chance of that from 58% to 78%. The 78% is a little higher than the 73% for “we’re alone” because it also counts cases where no one else arose here but settlers from another galaxy are on their way or already here.
Figure 3 Almost all of the shift in the odds comes from one simple fact: we see no one here
The numbers behind this chart
The label on each bar is in percentage points. Each move can also be written as a multiplier of the odds. (The odds are the chance that something is true divided by the chance that it is not.) The chart splits the timing of life into two bars, life’s early start and our late arrival, and it draws the galaxy surveys and the radio searches as one bar. All four bars are worked out on the “our existence proves nothing” view.
Because the size of a move depends on which observations were counted before it, each bar is the average over every order in which the evidence could be added. (An average of this kind is called a Shapley value.) The thin line under the bar is the range across those orders.
Going by the earliest proposed dates, life appeared on Earth almost as soon as it could. That favors an easy origin of life, so it lowers the chance that we are the first civilization in the Milky Way (−7 points, or odds multiplied by 0.7). We arrived late, in the last quarter or so of the time Earth can support life, and that raises the chance a little (+3 points, odds multiplied by 1.2). Seeing no one here, in a galaxy old enough to have been settled many times over, raises it much more (+22 points, odds multiplied by 2.8). The surveys of other galaxies and all the radio searches together come to +2 points, which multiplies the odds by 1.1.
With my starting assumptions, all the evidence together multiplies the odds that we are the first civilization in the Milky Way by about 2.6. That is modest evidence. The multiplier is different for the well-known views described in the section “It comes down to what you assume”. Among the ones that also use the “our existence proves nothing” view, it runs from about 1.2 to about 9, depending mostly on how likely you think it is that civilizations spread and how visible you think they would be.
The chance that we are the first civilization in the Milky Way comes up again in some later charts and tables, so here are three more things about it. The cases with settlers from another galaxy come to about 5%, and they are the whole gap between this number and the chance that we’re alone. Strictly, the number in the chart is the chance that the Milky Way would produce, on average, fewer than one other civilization. The chance that none actually arose is 77%. And in charts that stack the seven explanations, such as “How the evidence changed the odds” at the top of this page, the block “No one else in the Milky Way” is only one piece of this number. The other pieces are the block “No one else in the visible universe” and the 5% of cases with settlers from another galaxy.
Seeing no one here invites an obvious objection. We could only have evolved on a planet nobody had settled, so maybe finding Earth unsettled tells us nothing. Taken in its strictest form, that objection would wipe out the main result, because the evidence would then rule out nothing. I don’t take it that strictly, and the reason is timing. Suppose the galaxy does end up settled. Then nearly every civilization that arises on an unsettled planet arises early in the galaxy’s history, before settlers have reached most of it. We arose late in the galaxy’s history, so finding Earth unsettled still counts as evidence. (The last part of the section “What does our own existence tell us?” goes through this.)
Earlier work, and whether settlers could have missed Earth
Others have said before me that the evidence rules out only settlement that is both obvious and lasting (Freitas 1985; Carroll-Nellenback and colleagues 2019). Carroll-Nellenback and colleagues also argue that a settled galaxy can keep unsettled gaps for a long time, and that Earth could be in one. My starting assumptions make such a gap around Earth unlikely. When a check dropped that assumption, the odds changed little (see “Where the well-known theories fit”).
Sandberg, Drexler and Ord (2018) also found that seeing no settlement moved their result far more than searches of nearby stars did (see “How many civilizations? The range is enormous”).
The argument that now goes with Fermi’s remark was set out by Michael Hart (1975) and extended by Frank Tipler (1980). The idea of hard steps (very unlikely evolutionary leaps) comes from Carter (1983). Some people object to calling the question Fermi’s, or to calling it a paradox (Freitas 1985; Gray 2015).
The five assumptions behind my numbers
My numbers rest on five judgment calls that the data can’t settle. Two are about biology: does life start easily, and how many hard steps lie between a microbe and a species with technology? (Hard steps are evolutionary leaps so unlikely that they might happen only once, if at all, while a planet is habitable.) Two are about spreading from star to star: can a civilization keep spreading, and would we see it if it did? The fifth is the one behind the buttons near the top of this page: what does our own existence tell us?
I gave only a 20% chance that life starts easily. By “easily” I mean that, on average, it would start at least once while a planet is habitable. That choice leans toward fewer civilizations. On hard steps I leaned the other way: I assumed fewer of them than most recent estimates do. Before looking at any evidence, I put the chance that a habitable rocky planet ever goes on to produce a civilization at about 1 in a trillion. That figure is uncertain by an enormous margin in both directions.
My starting assumptions in numbers
The 1 in a trillion is for a rocky planet in the habitable zone of a Sun-like star, where liquid water could last. In powers of ten it is 10−12 (the raised number counts the zeros). Before looking at any evidence, I was 90% sure that the chance for a single planet lies between 10−54 and 10−4.
By life “starting easily” I mean that it starts at least once, on average, while a planet is habitable. Life can be slower to start than that and still end up widespread. My starting assumptions put the chance that life starts, at some time, on at least 1,000 planets in the Milky Way at 57%. After the evidence it is 44%.
The evidence on whether life starts easily cuts both ways. In this model, life’s early start on Earth favors an easy origin by about 2.4 to 1. The fact that we see no sign of anyone counts against an easy origin by about as much. So I started at 20% for an easy start and I finish at 20%. It is a coincidence of my starting assumptions that the two cancel. Change the assumptions about civilizations spreading, or the dates for the first life, and the chance finishes anywhere from 12% to 31%. Spiegel and Turner (2012) showed that life’s early start is weak evidence, and that most of the result comes from the starting assumptions.
What if you are more confident than I am that life starts easily? Someone who gives an easy start a 60% chance gets 53% for “we’re alone”. Someone who is 90% sure gets 38%. That puts “we’re alone” just behind the second answer, “they never left home” (others arose in the Milky Way, but none kept spreading from star to star), which gets 41%. The section “It comes down to what you assume” shows the odds for each of the well-known views.
How I checked the work
Alongside the model, I ran 22 studies on public data. (Eight are in the next three sections, and 14 more follow.) Some measure a number that was missing or doubtful, and some reproduce a published one. A few are step-by-step judgments of whether we would by now have noticed a second, separate origin of life, an alien probe, or the traces of a past visit.
Each study then got a second look: a separate run of the same AI whose job was to break the study. The model itself went through three rounds of fault-finding, also by separate runs of the same AI, each briefed to argue an opposing case. There were no other reviewers.
Those rounds led me to rebuild the model twice and change the headline twice. The first round showed that an early version got its result almost entirely from one unsupported assumption. The third round showed that I had presented a change in my assumptions as if the evidence had grown stronger.
The rest of the page backs up this result, with ranges, caveats and earlier work in notes you can open. To skip ahead, try your own assumptions, read what this means for us, or see where well-known theories such as the zoo hypothesis and the dark forest fit.
Are the stars too far apart, or has there been too little time?
Two popular replies to “Where is everybody?” are that the stars are too far apart and that there hasn’t been enough time. Distance matters less than you’d think, because other stars keep drifting close to the Sun. Lack of time doesn’t work at all: most Sun-like stars in our part of the galaxy are older than the Sun.
Both findings come from studies that repeat published measurements with newer data. They serve as a check, and my model doesn’t use either number. Each study got a second look, linked under its chart.
Take distance first. Using the Gaia satellite’s measurements of where nearby stars are and how fast they move, I counted how often another star system passes within 3.3 light years of the Sun. (Astronomers call that distance a parsec. It is about where the Sun’s comet cloud ends.) I get about 11 passes every million years (11.1, give or take 1.0), or about one every 90,000 years. That agrees with García-Sánchez and colleagues (2001). It is about half of a widely quoted later figure, and “Earlier counts of passing stars, and caveats” below explains why.
Those passes add up, as Hansen and Zuckerman (2021) and Zuckerman (2026) argued. Oxygen in Earth’s atmosphere is a sign of life that a passerby might detect. Since it first built up, about 21,000 star systems have come that close (between 11,000 and 59,000).
Suppose just 1 passing system in 20,000 carried a civilization able to cross 3.3 light years and leave something lasting behind. It could have come here without ever crossing the Milky Way, and nothing shows that one did. Keep two things in mind, though. Even 3.3 light years is a journey between stars. And if whatever a visitor left behind lasted only a million years, finding no trace today tells us almost nothing about past visits.
Earlier counts of passing stars, and caveats
A parsec is 3.3 light years, the distance light travels in 3.3 years. García-Sánchez and colleagues (2001) found 11.7 passes per million years, give or take 1.3. Fernandez-Puig and colleagues (2026) found 10.6, give or take 4.5. The widely quoted later figure is 19.7, give or take 2.2, from Bailer-Jones and colleagues (2018), and my rate is about half of it. Fernandez-Puig and colleagues trace that higher figure mainly to faulty distance measurements. Bailer-Jones and colleagues had flagged that risk themselves.
The 21,000 is less than you get by multiplying today’s rate by the time since oxygen built up. That is because the Sun’s surroundings change as it orbits the Milky Way, and over that span the rate averaged about four fifths of what it is now.
The count also depends on when life here could first be detected from afar. If that was not until oxygen neared modern levels, the count is about 6,000. If it was not until plants covered the land, the count is about 4,000. With the same uncertainties as for the 21,000, those two counts run from 2,600 to 13,000 between them.
Most passing stars go by at about 50 kilometers a second. Only about 1 in 8 goes by slower than 30 kilometers a second, and if only those can be reached, each count is eight times smaller. Later in this section I use two ship speeds, and the slower one is that same 30 kilometers a second (one ten-thousandth of the speed of light). A ship that slow would itself be slower than about 7 in 8 of the stars that pass the Sun.
About half the nearby systems lack a well-measured velocity, so they can’t be drawn on the chart below. My rate of 11.1 still includes them, because the way I counted works without a well-measured velocity for each star. The chart shows Gaia-catalog stars that are now within about 330 light years and that pass within about 33 light years of the Sun.
Figure 4 A star passes within 3.3 light years of the Sun about every 90,000 years
Each dot is one star’s closest pass to the Sun, from one million years ago to one million years from now.
The closest passes, as a list
- 1.9 light years, in 83,000 years, at 83 kilometers a second relative to the Sun
- 2.0 light years, 557,000 years ago, at 70 kilometers a second relative to the Sun
- 2.1 light years, 896,000 years ago, at 71 kilometers a second relative to the Sun
- 2.8 light years, in 930,000 years, at 33 kilometers a second relative to the Sun
- 2.8 light years, 284,000 years ago, at 76 kilometers a second relative to the Sun
- 2.9 light years, 308,000 years ago, at 79 kilometers a second relative to the Sun
- 3.0 light years, 522,000 years ago, at 31 kilometers a second relative to the Sun
- 3.0 light years, in 36,000 years, at 81 kilometers a second relative to the Sun
- 3.1 light years, in 27,000 years, at 32 kilometers a second relative to the Sun
- 3.3 light years (just outside the red band), 719,000 years ago, at 89 kilometers a second relative to the Sun
Then there is time. How much of a head start did other planets have? I used about 8,500 stars whose ages can be measured well. They show that about 70% of the Sun-like stars in our part of the galaxy formed before the Sun, by about 4 billion years on average.
The 70% counts stars through the full thickness of the galaxy’s flat disk. Near the middle of that thickness, where the Sun is now, 54% to 61% are older. Earlier published estimates run from 50% to 77%.
At 30 kilometers a second (one ten-thousandth of light speed), 4 billion years is enough to cross the Milky Way four times. At 1% of light speed, it is enough for 400 crossings.
The main possibility these two measurements leave open is that, until recently, something kept setting life back to the start all across the Milky Way. The usual candidate is gamma-ray bursts, the most violent explosions known (Annis 1999b; Ćirković and Vukotić 2008). My model leaves this possibility out.
Figure 5 Most Sun-like stars in our part of the galaxy are older than the Sun
Which stars I counted, and earlier estimates
“Our part of the galaxy” means stars at the Sun’s distance from the center of the Milky Way. The stars I used are subgiants, which are stars just leaving the Sun-like phase of their lives. The subgiant phase is brief and can be dated well, and I worked out the times on the chart from those dates. I checked the result against a second catalog of 17,000 stars. The bars show the main sample, in which 72% of stars are older than the Sun. Across the samples and the different ways of dating the stars, the share runs from 64% to 78%. The 70% in the text is the middle of that range.
The head start of about 4 billion years is an average over the older stars only. Count the younger stars too, and the average Sun-like star formed about 2.4 billion years before the Sun.
The earlier published estimates of 50% to 77% are worked out from histories of star formation (Lineweaver 2001; Madau 2023; Gallart and colleagues 2024). Lineweaver (2001) worked from the history of star formation and the build-up of heavy elements. He estimated that 74% of Earth-like planets in the universe are older than Earth (give or take 9), and that the average one is 1.8 billion years older (give or take 0.9). My figure for the average Sun-like star is the 2.4 billion years above. Madau (2023) got 77% for temperate rocky planets of stars a little smaller than the Sun, within 330 light years. Other published histories of star formation give between 50% and 74% (the 50% is Gallart and colleagues 2024, for the same volume of space). My figures fall within the spread of these estimates.
The model doesn’t use this measurement or the count of passing stars. It takes the ages of stars from its own tables. I made the two measurements as a check that there has been enough time and that other stars do come close.
Hart (1975) and Tipler (1980) argued that anyone able to settle the Milky Way would be here by now. So how rare must such settlers be? About 4 billion rocky planets have formed in the habitable zones of the galaxy’s Sun-like stars, where liquid water could last. Most formed early enough for settlers to have crossed the galaxy since, and we see none. If settlers would have filled the galaxy and been obvious, at most about 1 in a billion such planets can have produced them. Allowing for the uncertain number of such planets, the ceiling rises to between 3 and 9 in a billion.
Other galaxies show no sign of anyone
A civilization that spread through a whole galaxy and captured much of its starlight would leave it dimmer than a galaxy of its mass should be. The galaxy would also be too bright in the infrared, because captured starlight comes back out as waste heat.
Most searches look for the heat, and that works only if it comes out at a temperature they can detect. Among about 100,000 galaxies, Griffith and colleagues (2015) found none turning more than 85% of its starlight into heat.
A second test depends only a little on temperature: it compares each galaxy’s brightness with the brightness its mass predicts. Annis (1999a) used it, and so did Zackrisson and colleagues (2015). I applied it to 114,000 elliptical and lenticular galaxies (smooth and rounded, or lens-shaped) and to 9,400 gas-rich spirals. Each chart below links to its working paper and to its second look.
Earlier searches of other galaxies
The idea of looking for waste heat comes from Freeman Dyson (1960), who proposed it for single stars. Griffith and colleagues found only 50 galaxies that might turn more than half of their starlight into heat, and Garrett (2015) showed that nearly all of that survey’s candidates look like ordinary dusty galaxies.
For the brightness test, Annis had 137 galaxies in all, spirals and ellipticals together, and Zackrisson and colleagues had 1,359 spirals. Two months before I ran my test, Curtis and colleagues (2026) proposed a similar one in a preprint (a paper posted online that has not been through a journal’s review). Theirs uses the gap between a galaxy’s mass in stars and its mass worked out from the motions inside it.
Figure 6 Six galaxies in 114,000 look as if three quarters of their light is missing. Five of those are measurement errors.
The galaxies I could not explain
The brightness that a galaxy’s mass predicts is worked out from its size and the motions inside it. Of the six galaxies in the red zone, five are measurement failures: four have images that overlap a neighbor and one has a bad fit. The sixth is isolated and seen face-on, with a faint outer disk. A smaller sample from the southern sky holds four more unexplained cases: two similar disk galaxies, and two whose internal motions are poorly measured.
Across both samples, 11 galaxies look as if at least three quarters of their light is missing, and only one looks too bright by the same margin. I found a fault in six of the 11. Faulty measurements are the likely cause for the other five as well, but I have not shown that.
With a looser cutoff, at least two thirds of the light missing, 23 elliptical and lenticular galaxies remain unexplained in all. Among the spirals, 34 that seem to be missing at least three quarters remain unexplained. None of the 23 shows excess infrared heat. For the spirals, infrared measurements rule out warm waste heat in all of the 34 except 2 to 4. Nobody has followed up any of these galaxies with new observations.
A test like this always leaves some galaxies it can’t explain. The chart’s red zone leaves one, and a looser cutoff (at least two thirds of the light missing) leaves 23 ellipticals. The spirals are not in the chart, and they add 34 that seem to be missing at least three quarters of their light. Nobody has observed any of these galaxies again. “Unexplained” means only that I did not find a fault. Until someone observes these galaxies again, none of them should be treated as a possible sign of anyone.
So few dim galaxies turned up that at most 0.29% of large elliptical galaxies, and at most 2.6% of large spirals, can be missing three quarters of their light. (Those are 95% confidence limits: if the true share were higher, the chance of seeing so few would be under 1 in 20.)
The 2.6% for spirals comes with two cautions. First, it beats the limit Zackrisson and colleagues published in 2015 only because I included a catalog whose galaxies were picked by hand. If I use only the one sample that was picked by a fixed rule, my limit is 4.5%, which is no better than theirs. Second, the 2.6% weakens if the waste heat is warmer than about 250 kelvin (minus 23 Celsius).
Figure 7 At most about 3 in 1,000 large elliptical galaxies (0.29%) could have someone collecting three quarters of their starlight
The galaxy numbers in detail
The elliptical limit covers red elliptical and lenticular galaxies about as luminous as the Milky Way or more, within about 1.5 billion light years. It could be as low as 0.24% or as high as 0.45%, and it reaches 1.4% under the most pessimistic assumption about noise. The spiral limit covers massive, gas-rich disk galaxies and runs from 1.9 to 3.5% depending on the analysis choices. The test cannot detect missing light in disks that rotate slower than about 120 kilometers a second. Both limits are based on the galaxies in which missing light would have been detectable, about 15,000 of the ellipticals and 1,800 of the spirals.
Zackrisson and colleagues gave about 3% as their conservative limit, and 0.3% as a tentative one after inspecting their candidates. Counted the same way as mine, their conservative limit is 4.6% and mine is 2.6%. Mine is lower because I explained more of the outlying galaxies, not because my sample is larger. That gain depends on a catalog whose galaxies were picked by hand. I modeled how likely that catalog was to include a galaxy with missing light, and did not measure it. The ALFALFA radio survey is the only spiral sample picked by one fixed rule. On ALFALFA alone my limit is 4.5%, which is no improvement on theirs and is still weaker than their tentative 0.3%.
A smaller southern-sky sample of ellipticals (the 6dF Galaxy Survey) gives weaker limits, and I left it off the chart. The elliptical limits weaken at the right of the chart, where the most light is missing, because few galaxies are bright enough to test. The spiral limit for galaxies missing 90% of their light (1.5%) is based on about 700 galaxies.
A civilization might enclose only some of its stars to catch their light. Enclosing every star up to the Sun’s brightness would remove at most about 20% of an old galaxy’s light, and enclosing everything up to ten times the Sun’s brightness at most about 36% (both figures from Lacki 2019). My limits say very little about either case, and nothing about a civilization that enclosed only the dim red stars.
Radio searches have covered very little
Suppose every star in the Milky Way hosted a civilization like ours today. Even then, you would expect the radio searches so far to have picked up less than one of them (0.7 on average). That figure is generous, because it assumes all the radio power a civilization leaks into space comes out as one narrow tone, at a single frequency. Without that assumption the figure is zero. So 60 years of SETI, the search for extraterrestrial intelligence, have set no limit on how many transmitters like ours the Milky Way holds.
Near the Sun, though, there is a published limit for signals as strong as the Arecibo planetary radar, which was for decades the most powerful radar on Earth. At most about 0.07% of star systems within 160 light years are beaming a continuous signal that strong at us (Wlodarczyk-Sroka and colleagues 2020).
The searches do put a ceiling on very powerful beacons. Take a beacon of 100 million billion watts (1017), which is a bit over half the power of all the sunlight striking Earth. At most about 400,000 of those, roughly 1 star in 200,000, could be broadcasting now without the searches having caught one.
My figures combine the major published radio searches. As far as I could find, nobody had combined them before. The conclusion matches Wright and colleagues (2018), who compared the searching done so far to a hot tub’s worth of the oceans.
Figure 8 Radio searches rule out crowds of very powerful transmitters and say almost nothing about weak ones like ours
The radio numbers in detail
My figures combine the major published radio searches, counting each patch of sky and frequency only once. The published limits most like mine are Breakthrough Listen’s, which count every star in the patches of sky that search listened to (Wlodarczyk-Sroka and colleagues 2020; Mason and colleagues 2025), and Lacki’s (2024, 2025).
The 400,000 is read off the amber curve at 100 million billion watts. It could be as low as 170,000 or as high as 2.6 million. Four fifths of that limit comes from the stars in Breakthrough Listen’s patches of sky, about half of them in ten crowded fields some 20,000 light years away. For those patches, Mason and colleagues (2025) found that at most about 1 star system in 100,000 within 8,000 light years has a continuous, steady-frequency transmitter above 50 million billion watts (5 × 1016).
The 5,600 on the blue curve is for a beacon that sits exactly on the natural frequency of hydrogen and whose frequency holds steady. Almost all of that number comes from two sky surveys begun in 1985 and 1990, taken at face value. I could defend anything from about 1,000 to 370,000. For a transmitter on a spinning planet, whose frequency drifts, the figure is about 100,000.
For beacons with the full power of a star, the Milky Way could hold up to about 100 that the searches have missed. Here too, almost everything depends on one survey from the 1990s, taken at face value. Going by modern searches alone, the number is 2,000 to 9,000, and the full range is 55 to 18,000. All these figures for star-powered beacons are probably too high. That is because a beacon that strong would also be among the brightest sources in ordinary radio maps of the sky whenever its frequency fell inside a mapped band. Lacki (2024, 2025) works out limits from ordinary radio observations. I did not put a number on how much lower my figures should be.
For transmitters like our own, the published limit is a share of nearby stars, the 0.07% in the main text. Enriquez and colleagues (2017) had put it below 0.1%. If every star hosted a civilization like ours, and its signal were a radar beam switched on only some of the time and pointed at random, the expected number of detections would be under one in a million.
What Earth’s history can and can’t tell us
Two facts about Earth keep coming up in arguments about life elsewhere. Life showed up early, and intelligence showed up late, in roughly the last quarter of the time the planet can support life. Both are weaker evidence than they look.
Brandon Carter (1983) took the late arrival as a sign that intelligence needed one or two “hard steps”: evolutionary leaps so unlikely that they might happen only once, if at all, while a planet is habitable. Later estimates put the number at about four (Watson 2008) or five or six (Carter 2008).
The early start is weak evidence that life starts easily, as Spiegel and Turner (2012) showed. The conclusion depends mostly on what you assume beforehand.
Figure 9 Life showed up early. Complex life took billions of years more.
The events on this chart, as a list
- A habitable Earth: The impact that formed the Moon, the earliest point at which a habitable Earth could begin (published ages run from about 4.42 to 4.52 billion years ago), 4.51 billion years ago (model estimate)
- A habitable Earth: First evidence of liquid water: a single mineral grain from Jack Hills, Australia, that formed where water was present, 4.4 billion years ago (accepted)
- A habitable Earth: Oceans present, read from the oxygen in ancient mineral grains, 4.25 billion years ago (accepted)
- A habitable Earth: A proposed burst of giant impacts that could have sterilized Earth (the Late Heavy Bombardment), 3.9 billion years ago (disputed)
- Evidence of life: Last common ancestor of all life today (dated from molecular clocks, which read time from the rate at which genes change), 4.2 billion years ago (model estimate)
- Evidence of life: Carbon that may have come from living things, sealed inside a mineral grain from Jack Hills, Australia, 4.1 billion years ago (disputed)
- Evidence of life: Carbon that may have come from living things, in rocks from Saglek, Labrador (the age of the rocks is disputed too), 3.95 billion years ago (disputed)
- Evidence of life: Microscopic tubes of iron oxide that may be fossil microbes, in rocks from Nuvvuagittuq, Quebec, 3.77 billion years ago (disputed)
- Evidence of life: Layered structures and carbon that may be traces of life, in rocks from Isua, Greenland, 3.7 billion years ago (disputed)
- Evidence of life: Earliest undisputed fossils: layered mounds built by microbes in hot-spring deposits of the Dresser Formation, Australia, 3.48 billion years ago (accepted)
- Evidence of life: Microscopic structures in the Apex chert, Australia, whose reading as fossils has been disputed since Brasier and colleagues (2002), 3.46 billion years ago (disputed)
- Milestones: Photosynthesis that releases oxygen (dated from the rate at which genes change), 2.99 billion years ago (model estimate)
- Milestones: The ancestors of mitochondria, which supply the energy of complex cells, split from free-living bacteria (dated from the rate at which genes change), 2.25 billion years ago (model estimate)
- Milestones: Complex cells: the last common ancestor of all cells with a nucleus (dated from the rate at which genes change), 1.74 billion years ago (model estimate)
- Milestones: Origin of animals, dated from the rate at which genes change, with revised limits from fossils, 750 million years ago (model estimate)
- Milestones: Animals: the last common ancestor of all living animals, 600 million years ago (model estimate)
- Milestones: Land plants: the last common ancestor of all living land plants, 470 million years ago (model estimate)
- Milestones: Humans: the oldest known fossils of our species, from Jebel Irhoud, Morocco, 300,000 years ago (accepted)
- Oxygen in the air: The Great Oxidation, when oxygen first built up in the air and stayed, 2.33 billion years ago (accepted)
- Oxygen in the air: A second rise in the oxygen level, with swings up and down, 640 million years ago (disputed)
- Oxygen in the air: The oxygen level settles near today’s, 410 million years ago (accepted)
- Biosphere ends: End of the biosphere: plants run out of carbon dioxide (the range shown runs on to the point where Earth is too hot), in 900 million years (model estimate)
- Biosphere ends: End of the biosphere, in a model where living things speed up the weathering of rock, in 1 billion years (model estimate)
- Biosphere ends: Oxygen in the air falls below 1% of today’s level, in 1.08 billion years (model estimate)
- Biosphere ends: Land plants die out, in a model where the weathering of rock depends only weakly on temperature, in 1.73 billion years (model estimate)
- Biosphere ends: Plant life ends, in a three-dimensional climate model, in 1.84 billion years (model estimate)
- Biosphere ends: The last single-celled life dies out in its final refuges (the limit for microbes, not for complex life), in 2.8 billion years (model estimate)
Dates for complex life, and the sources
Today’s complex cells share a last common ancestor that lived roughly two and a half billion years after the first life. The study that dates this ancestor (Kay and colleagues 2026) says complex cells started coming together about a billion years before it lived, and that the nucleus was one of the early stages. Animals appear about a billion years after that ancestor.
The biosphere (the living world as a whole) has roughly one to two billion years left. The arrowhead at the right marks an estimate that falls off the end of the axis: the last microbes dying out in their final refuges, about 2.8 billion years from now.
The first evidence of liquid water comes from Wilde and colleagues (2001). The earliest undisputed fossils are from the Dresser Formation in Australia (Djokic and colleagues 2017). The gene-based date for the first life is from Moody and colleagues (2024), and the date for the origin of animals follows Lole Durbin and colleagues (2026).
Kipping (2025a) took the earliest date proposed for the first life. He found that an early start is at least 13 times more likely if life starts easily than if it doesn’t (odds of 13 to 1). I get the same figure. But his model has only one slow process after the origin of life, in effect one hard step, and the odds depend on what follows. More hard steps would lower them, and having none would raise them.
My model puts the odds lower, at about 2.4 to 1, for two reasons. It covers the whole range of dates proposed for the first life, and it counts an origin as easy even if life starts only once, on average, while a planet is habitable.
If my model used 13 to 1, the chance that no other civilization has ever arisen in the Milky Way would be 62%, down from 78%.
Figure 10 Life’s early start on Earth is only weak evidence that life starts easily
The odds in detail
Odds like these say how many times more likely an early start is if life starts easily than if it starts with difficulty. Kipping’s 13 to 1 is a lower limit for his model, which uses the earliest date for the first life and has a single slow process after the origin. With two, three or four hard steps after the origin of life, the figure drops below that limit, to 9, 7 and 6 to 1. It falls because, as Hanson (1998b) and Carter (2008) had shown, even a hard origin tends to come early when hard steps follow it. If nothing hard comes after the origin, the figure rises to between about 20 and 26 to 1.
It also matters when Earth is first counted as habitable. If you start the clock at the impact that formed the Moon, instead of at the first evidence of liquid water, the 13 to 1 drops to about 9 to 1. Kipping gives more than 9 for a similar case, and notes that the gene-based date is recent and may not hold. Going by undisputed fossils alone, the figure is about 3 to 1 or less for any number of later hard steps (the lowest line on the chart), and at most about 6 to 1 if nothing hard comes after the origin.
All these figures assume that life was equally likely to start at any time while Earth was habitable. If it could start only on the young Earth, say up to 200 million years after a giant impact, the odds fall to between 1 and 2.3 to 1 (see the study “Did giant impacts delay life’s start?”).
My model takes none of these figures as an input. Its own figure, about 2.4 to 1, is well below Kipping’s 13. Part of the reason is that it covers the full range of dates proposed for the first life. But with the earliest date alone it still gives only between about 4 and 5 to 1. The rest comes from how it defines an easy origin. Any rate counts as easy if life starts at least once, on average, in the time a planet stays habitable. At the low end of that range of rates, an early start would be a surprise too.
The model could be set to favor an easy origin by 6, 9 or 13 to 1 instead. That would lower the chance that no other civilization has ever arisen in the Milky Way from 78% to 71%, 66% or 62%. (The 78% is a little higher than the 73% for “we’re alone” because it also counts cases where no one else arose here but settlers from another galaxy are on their way or already here.)
I redid the published calculations and got the same main results, with two exceptions. The bigger one is in Snyder-Beattie and colleagues (2021), who argue from Earth’s timeline that intelligence needs several hard steps. I could match their Table 2 only by assuming that some numbers printed as powers of ten were actually worked out as powers of a smaller number. (The note below, “How I rechecked the numbers in two published papers”, explains.) If I’m right, the paper’s main conclusion still stands: these steps probably take longer, on average, than Earth stays habitable. But its suggestion that intelligence might arise less than once per visible universe gets weaker.
The smaller exception is in Kipping (2020), where two of my numbers come out 5 to 6% lower than his. That may be my own error, and it doesn’t change his conclusion.
So both exceptions are open questions about my reading, not confirmed errors.
How I rechecked the numbers in two published papers
Table 2 in the paper by Snyder-Beattie and colleagues uses thresholds for how rare intelligence is, printed as powers of ten. I could reproduce the table only by assuming they had been calculated as powers of e (a mathematical constant, about 2.7). The paper has a second table, which roughly matches my calculation if it is read the same way and doesn’t match at all as printed. The second look at my study found the same thing.
Anyone can check this from the bottom rows of Table 2, which give the paper’s starting assumptions. Under those assumptions, each step has a chance of at least about 1 in 2 billion of happening while Earth is habitable. So in a two-step model, the chance that intelligence evolves can’t be lower than about 1 in 10 billion billion (10−19). But the two-step row says the starting assumptions put a 6% probability on values below 1 in a trillion trillion (10−24). Read as powers of e, all nine entries in those rows fit closely.
If the reading is right, at least three of the paper’s results change, two of them by a lot. In its three-step model, the paper puts over 90% of the probability on intelligence having less than a one-in-a-trillion chance of evolving. On my reading that falls to about half. The share below 10−24, a threshold a trillion times lower still, falls from over half to about 1%. In the paper those two thresholds stand for the number of stars in the Milky Way and in the visible universe. For two steps, the paper says that over 90% of the probability lies below a 1% chance. On my reading that becomes 86 to 96%. The paper’s “less than 1% chance” statements still hold for three and four steps.
In Kipping (2020) I get exactly his odds for life’s early start. I also get his two odds that intelligence is rare (1.47 and 1.42) when the starting assumption is flat, meaning that every value between the two extremes, zero and one, is equally likely. With the starting assumption he prefers, my two odds come out 5 to 6% lower than his: 1.55 and 1.49 against the published 1.64 and 1.57. His preferred assumption piles up at the two extremes. So the gap probably comes from how one of the two calculations handles the spike at zero, and I haven’t worked out which. His conclusion, odds of roughly 3 to 2 that intelligence is rare, doesn’t change.
A hard step would look like a new trait that evolved only once, but the fossil record can’t settle how many hard steps there were. Vermeij (2006) found that traits that evolved only once are older than traits that evolved repeatedly. He suggested that the old ones may only look unique because the evidence of repeats has been lost.
I re-ran his analysis and found the same pattern. But some traits were inherited together, as a package. I grouped them in three ways, and in two of the three the pattern is too weak to tell from chance. The dates of life’s big changes, such as complex cells and animals, can’t show whether luck or the environment set the pace either (see the study “Did complex life need luck, or just oxygen?” below).
Things evolution did only once, and when life’s big changes came
Vermeij’s point was that an old trait with one known origin may only look unique. Mills and colleagues (2025) add that the first lineage (line of descent) to arrive may shut later ones out.
In my re-analysis, each tenfold increase in a trait’s age multiplies the odds that it evolved only once by 3.2. Some traits were inherited as a package from the first cell, the first complex cell or the first animals. When each package is counted once, the multiplier is only 1.4 to 2.9, depending on which of three ways of grouping I use. With two of the three, the result can’t be told apart from no effect.
A plain count doesn’t settle the question either. Of the 13 proposed hard steps, 8 have one known origin, and 3 of those 8 lie within the human lineage alone. Of the other 5, 3 have happened more than once and 2 are disputed.
My timing study (the working paper behind the first two charts in this section) also compared explanations for the dates of life’s big changes (evolution’s major transitions). An explanation in which the environment sets the pace fits those dates about four to five times better than hard steps arriving at random. It fits only one to two times better than hard steps that had to wait for oxygen. Both figures are upper limits, because the environmental thresholds were chosen after the dates were known.
A later, fuller study of the same question (“Did complex life need luck, or just oxygen?”) found that, on balance, the dates give the environmental explanation no support. The result swings from strongly for to strongly against, depending on which dates are used and on how much oxygen animals needed.
One candidate hard step, a large brain, can be checked against living animals. Relative to body size, three or four separate groups of mammals have brains at least half as big as ours: dolphins, and two or three groups of monkeys. So large brains evolved more than once. The step from a large brain to technology has happened once, and nothing in the record shows how hard it is.
Figure 11 Brains that are big for the body evolved more than once
Details, neuron counts and sources
By 16 million years ago, relatives of dolphins had reached about half of our relative brain size, and the best-measured living dolphins are at about two thirds to three quarters of it. One fossil outside the dolphins and monkeys also qualifies: a dwarf island elephant, whose ratio is high because its body shrank. If the brain itself must also be at least half the size of ours, only dolphins and porpoises qualify.
Counting nerve cells is another way to compare brains. By the counting method most used now, we have 16 billion cortical neurons (nerve cells in the brain’s outer layer), and no other species has a measured count that is even half of that. An older method gives the pilot whale 37 billion, almost twice the human count by that method (Mortensen and colleagues 2014). Only one whale species has been counted both ways, and the two counts differ fourfold (Avelino-de-Souza and colleagues 2025), so neuron counts can’t settle the question either.
Marino and colleagues (2004) documented the dolphins’ rise. Lineweaver (2009) argued that it does not show that human-like intelligence is repeatable. The brain and body masses of living mammals are from Burger and colleagues (2019). The fossil points are from Bertrand and colleagues (2022) for early placental mammals, Waugh and Thewissen (2021) for whales, Benoit and colleagues (2019) for elephants and their kin, and Du and colleagues (2018) for the human lineage. For the human lineage I assumed an average body weight for each species, so the height of those points is approximate.
Did we arrive early in the universe’s history?
We came late in Earth’s history, but we may be early in the universe’s. Red dwarfs (small, dim stars) will shine for trillions of years. If civilizations can keep arising around them, humanity is surprisingly early among all the civilizations that could ever arise, as Loeb and colleagues (2016) pointed out. (Measured only against the stars already formed, we are late.)
Hanson and colleagues (2021) showed that the more hard steps there are, the earlier we turn out to be. Their explanation for our being so early is that expanding civilizations will soon fill the universe and leave no room for latecomers.
I checked what telescopes have seen so far. The JWST and Spitzer space telescopes have clear results for 14 rocky planets orbiting red dwarfs, and none has a thick atmosphere. But all 14 are hotter than the usual habitable zone (where liquid water could last), though the coolest only slightly. So this is indirect evidence that red-dwarf planets are poor homes for life. By itself, that evidence is enough to explain why we seem so early only if there is a single hard step.
Figure 12 Whether we’re early depends on how long planets stay habitable
The telescope evidence, and earlier work
If ambiguous results are counted, at most 3 of 17 red-dwarf planets have a thick atmosphere. None of the 17 is inside the habitable zone as I drew it, which stops at 0.9 times the starlight Earth gets. The coolest, TRAPPIST-1 d, gets about 1.1 times Earth’s starlight, and the team that observed it places it at the zone’s edge.
For us to be no longer early, each red-dwarf planet would have to count about 20 times less as a home for life (a 20-fold discount) if there is one hard step, and about 20,000 times less if there are two. Leaving red dwarfs out altogether goes further. We are then among the first 1% only if there are five or more hard steps, and only if orange dwarf stars as small as 0.6 times the Sun’s mass also stay habitable for tens of billions of years.
Hanson and colleagues found that we are no longer early only if there are fewer than three hard steps and no planet stays habitable for more than about 50 billion years. With six steps, their paper still placed us within the earliest 11% even if no planet lasts beyond 10 billion years. Their paper has the formation of habitable stars peaking much later than in the standard history of star formation, which I used for this chart. With that standard history, we are not early at the 10-billion-year limit for any number of hard steps from one to ten.
Kipping (2025b) argued that a shorter planetary lifetime is not enough by itself, since red dwarfs must also fail to produce observers. With a short planetary lifetime we are no longer early, but our yellow host star still looks unusual. That is the leftover puzzle Kipping found.
What 14 more studies found
After the eight studies above, I ran 14 more. They were aimed at the explanations my model left standing, and at the numbers in it that I knew least about. None found anyone, and none produced a number that the model uses. They show how much room is left to hide. Warm, bright structures around nearby stars are rare, but we would have missed a small probe, a short visit long ago, or much that is cold, faint or far away.
Each study then got a second look from a separate run of the same AI. In all 14, at least one of my first claims did not survive. In one study, a one-line error left the search blind to the thing it was looking for. In another, I had run no search, and my first figures only repeated what I had assumed. For 13 studies, the revised figures have not been checked again. Under each study, “What this study can’t show, and the mistakes I fixed” also gives the ranges and the closest published work.
Do civilizations keep spreading?
What can the settlement of the Pacific tell us?
When Pacific islanders could reach new islands, they settled them fast: 29 of 50 islands or island groups in East Polynesia within roughly 250 to 360 years. But settling a galaxy means colonies founding colonies founding colonies, thousands of times over. In the eight human migrations I studied, that chain never ran past six rounds. So human history can’t tell us whether a wave of settlement keeps going.
Checked againWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. The burst of settlement in East Polynesia came after a pause of about 1,700 to 1,900 years. Of the 29 islands and island groups that were settled, 16 were later abandoned, most of them small islands far from a settled neighbor. Every wave I studied ended after four to six rounds of colonies founding colonies. By my own count, six of the eight waves ended because the settlers had run out of new places to reach. So the fact that these waves ended tells us nothing either way about whether settlers have a built-in tendency to stop.
Earlier work. Comparing the settling of the Pacific with settling the stars is an old idea (Finney and Jones 1985). I tested a model by Landis (1998), called the percolation model, in which each new colony either goes on to found others or stops for good. Every historical finding here points the same way as work already in print, and none of my figures is more precise than the published ones.
What the second look corrected. At first I reported that 97% of habitable islands were filled, and that the study gave me a usable starting assumption about spreading between the stars. The second look showed that both results came from the way I had defined my terms, so I withdrew them.
How hard is the road from a bare planet to intelligence?
Did giant impacts delay life’s start?
Life started early on Earth, and Kipping (2025a) found that this favors an easy origin of life by at least 13 to 1. Giant impacts may have shortened the time life had to start, so I took their timing from 30 dated basins (giant craters) on the Moon. Kipping’s odds stay the same as long as only an impact that melts a planet’s whole surface can sterilize it. That threshold is my reading, and it isn’t settled.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. Melting the whole surface takes an impactor about 2,000 km across, and that threshold is my reading of the published work. The Moon’s basins tell us when the big impacts came, but whether an impact of a given size would sterilize a planet is untested. Any impactor big enough to sterilize a planet would also be larger than the biggest body that hit the Moon, so I had to assume how many such giants there were.
My middle estimate is that the last impactor big enough to boil the oceans (at least 440 km across) struck between 4.12 and 4.25 billion years ago. One big enough to melt the surface had only a 0.4 to 0.9% chance of arriving after the oceans formed. So Kipping’s figure of at least 13 to 1 stays where it was. That figure assumes that the common ancestor of all living things lived 4.2 billion years ago. Dating life from accepted fossils instead gives 2.9 to 1, and that doesn’t change either. If impactors of 440 km do sterilize, life had less time to start, and the 13 to 1 rises to between 17 and 24 to 1.
Life’s early start stops counting as evidence in just one case: when life can start only within a set window of time after a large impact. For windows of up to 200 million years, the figure falls to between 1.0 and 2.3 to 1, where 1.0 means no evidence either way. For a window of 500 million years it falls to between 1.8 and 5.6 to 1.
Earlier work. The last impact by a body larger than 500 km has been dated to about 4.3 billion years ago (Marchi and colleagues 2014), and my date for that size is the same or somewhat later. The new part is using those dates to ask about the origin of life. Abramov and colleagues (2026) treat ocean-boiling impacts as sterilizing, so not everyone shares my 2,000 km reading.
What the second look corrected. At first I used the measured mass of late-arriving material only as an upper limit. Using it as a measurement, with a lower limit as well, turned my single date of 4.25 billion years into the range given above. I also withdrew a claim about Earth-like planets. All the figures given here are the revised ones.
If life had started twice on Earth, would we know?
If life starts easily, the searches of Earth and the Solar System so far had a 3 to 25% chance of finding a second, independent origin of life. Adding their result to the model would move the chance that no other civilization has ever arisen in the Milky Way by at most about 3 percentage points, up or down. I scored how complete the searches are using public databases and mission records, and those scores are judgments.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. None of my scores is a measurement. Life might have started more than once in the same place, and three ways to count such repeats put the chance of finding a second origin at 3, 9 and 25%. If each of my scores can be wrong independently of the others, the range widens to 1 to 42%. The model also treats a “thorough search” as something we might carry out in the future. A search that thorough would have had an 87% chance of finding a second origin.
The model puts the chance that no other civilization has ever arisen in the Milky Way at 78%. (The 78% is a little higher than the 73% for “we’re alone” because it also counts cases where no one else arose here but settlers from another galaxy are on their way or already here.) Counting what the searches found would move the 78% by somewhere between about 2 points down and 3 points up. Read as having found nothing, they would raise it by 0.3 to 2.6 points, depending on which of the three ways to count is used.
The 78% would go down only if a rock on Mars called Cheyava Falls counts as evidence of life. The rock has features that might be signs of life, and the question is unresolved. If those features are 1.1 to 1.4 times likelier with life than without, that cancels the two smaller of the three rises. (Across the full range of my scores, the figure needed can be as high as 2.7.) If the features are twice as likely with life, the shift is up to 2 points down. I can’t say which of these holds.
Earlier work. Spiegel and Turner (2012) showed that one independent origin of life would be far stronger evidence than Earth’s early start. Davies and Lineweaver (2005) proposed looking for a second sample of life on Earth. Neither my first analysis nor the second look found a published estimate of how complete the search has been, so there is no earlier figure to compare mine with.
What the second look corrected. At first I reported a single figure, a rise of 0.3 points. That used one way to count and left out the Mars rock, so I now give a range. I also withdrew my statement that the searches so far amount to “about 1% of a thorough search”. Three of the figures above, the 42%, the 1.4 and the 2.7, were worked out after the second look and have not been checked.
Did complex life need luck, or just oxygen?
Complex cells, animals and humans may each have needed a hard step (a very unlikely evolutionary leap), or may only have had to wait for oxygen. I made 48 different comparisons between their published dates and dated rises in oxygen. Depending on the comparison, the result runs anywhere from ruling out the oxygen explanation to favoring it over luck by 28 to 1. So the dates can’t tell us which it was.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. Across the 48 comparisons the figure runs from 0 to 28. It is how likely the dates are if life was waiting for oxygen, divided by how likely they are if it needed luck. A figure of 0 means that waiting is ruled out, a figure above 1 favors waiting, and one below 1 favors luck. The main unknown is when animals first had enough oxygen. The date used for complex cells matters too: with three of the published dates the figure is 1.2, 1.9 or 2.3, and with a fourth it is 28.
A third possibility is lucky steps that also had to wait for oxygen. Against that possibility, the figure for waiting alone is 0.4 to 0.8 with the three dates and 11 with the fourth. I also compared a model with no lucky steps against a model with at least one. The dates are 0.87 times as likely with no lucky steps (the range is 0.37 to 1.3), and a figure that close to 1 gives neither model real support. If animals had enough oxygen by 2.2 billion years ago, the 0.87 becomes 0.03 to 0.34, which favors luck.
The dates for complex cells are the dates of their last common ancestor (the most recent cell that all of them descend from). Complex cells had begun to take shape about a billion years earlier, before oxygen first rose (Kay and colleagues 2026). We have only one planet to study, and life itself made most of the oxygen, so whether oxygen was the cause is untested. I also checked 31 later evolutionary innovations and found no link to rises in oxygen, but the check could not have detected delays much longer than 10 million years.
Earlier work. Mills and colleagues (2025) proposed that the environment set the pace of evolution. They computed no odds, so there is no published figure to compare mine with.
What the second look corrected. At first my main figure was 8.3. It was an average dominated by one date, and I, not the supporters of the oxygen explanation, had chosen which oxygen event to use for animals. I dropped both the figure and that choice. I also dropped a claim that choosing the oxygen events after the fact gave a 200-fold gain, which I had miscalculated. After the second look I added the figure for the case where animals had enough oxygen by 2.2 billion years ago.
How long do rocky planets stay habitable?
For rocky planets that formed when the Sun did, I estimated how long they stay geologically active enough to be habitable. By one rule for when habitability ends, my middle estimates are 4 to 11 billion years, and by other rules they reach about 21 billion. That spread is too wide to say whether we are early: humanity could be among the first 2% of civilizations to arise, or could come after 70% of them.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. The estimate starts from thorium and europium measured in Sun-like stars, two elements that trace how much radioactive heating a star’s rocky planets get. It also uses four models of how planets cool. The 4 to 11 billion years holds if habitability ends when the gas released by volcanoes falls to a tenth of Earth’s, and the choice of that rule matters as much as the choice of cooling model.
Among stars very like the Sun (solar twins), the measured amounts of thorium and europium differ from star to star by only 7 to 10%, which is too little to matter. In broader samples of stars, all that can be said is that the difference is no more than about 19 to 27%. Of the other sources of radioactive heat, uranium has never been measured in these stars, and potassium-40 comes entirely from models.
Humanity’s range (among the first 2% of civilizations, or after 70% of them) is for four hard steps. It comes from a set of reasonable assumptions about the minority of planets that stay habitable much longer than the rest. With six hard steps, humanity could be among the first 0.3% or could come after 63% of civilizations.
Earlier work. Unterborn and colleagues (2022) estimated such lifetimes from europium in stars, for planets whose crust is one rigid shell with no moving plates. I did not compare my lifetimes with theirs on the same terms, so I can’t say whether mine are more precise or less. What is new here is using the lifetimes to work out where humanity falls among civilizations.
What the second look corrected. My first claim was that geology explains away the puzzle of why we seem to be early among civilizations. That claim did not survive the second look. I worked out the ranges for humanity’s place afterward, and they have not been checked.
Could they be out there, or even here, without our noticing?
Would Earth’s rocks record a visit?
We would almost never know of a small visit to Earth. Take a site of 100 square kilometers that was occupied for 10,000 years at some time in the last 500 million years. We would know of it from buried rock about 1 time in 40,000. I measured only one part of that estimate, how much ancient land has been inspected, using 47,525 collections of land fossils in the Paleobiology Database. The rest is assumed.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. For that site the range is 1 in 500,000 to 1 in 3,000. The part I measured is how much ancient land has been inspected, and each fossil collection marks a place where someone has looked. Pick a patch of 100 square kilometers and a span of 20,000 years at random, and the chance that a fossil collection comes from it is 1 in a million. Everything else is an assumed range: how much has been inspected in other ways, and whether anyone would recognize an artifact for what it is.
A larger site, of 10,000 square kilometers, would be known about 1 time in 430 (the range is 1 in 5,500 to 1 in 30). An industrial period like ours, at some time in the same 500 million years, would be known in about 10 to 20% of cases (the full range is 1 to 70%). The study does not cover sites underground, at sea or off Earth.
Earlier work. Schmidt and Frank (2019) asked a similar question, about an earlier industrial civilization native to Earth, and answered it in words, without giving probabilities. Davies (2012) wrote about the footprints that alien technology might leave. I found no published probabilities to compare with mine.
What the second look corrected. My first figure for small sites was higher. Almost all of it came from an assumption about recent remains still lying at the surface, so it is no longer the figure I lead with. I had also claimed that the survival of traces in rock is not what limits detection, and I withdrew that claim. The site ranges and the upper percentages were worked out after the second look and have not been checked.
Would we notice a probe parked nearby?
Picture an inactive alien object 1 to 10 meters across. It has a 20 to 36% chance of already being cataloged if such objects mostly sit near Earth, and about 2% if they sit in the asteroid belt or beyond. Its chance of also being recognized as alien is 0.01 to 0.6%, which rests on values I assumed. The second look found a fault in my simulation of asteroid surveys that I have not fixed.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. I used images of the Moon and Mars, the asteroid catalogs and a simulation of the sky surveys that find asteroids. Depending on what I assume, the chance of being cataloged is 4 to 44% if such objects mostly sit near Earth, and 0.3 to 5% if they sit in the asteroid belt or beyond. The chance of being both cataloged and recognized is at most 11%, and that is with every assumption set in its favor. The values I assumed include a 1% chance that the object’s origin is correctly identified. Heat, radio signals and deliberate concealment are outside the study.
Earlier work. I followed the approach of Haqq-Misra and Kopparapu (2012). The first searches of this kind looked at the stable points of the Earth-Moon system, places where a parked object stays put (Freitas and Valdes 1980; Valdes and Freitas 1983). Pan and colleagues (2026) posted a similar estimate before mine, in a preprint (a paper posted online that has not been through a journal’s review). Under their standard assumptions, my figures for the chance that an object is cataloged agree with theirs for objects of 10 meters and of 1 kilometer. In their other cases, mine and theirs differ by up to six times.
What the second look corrected. My simulation expects surveys to find fainter objects than they actually do. The second look found that fault, and I have not fixed it. I also withdrew two figures. One was a size threshold of 0.84 meters for objects near Earth, which was too precise because it rested on two captured asteroids. The other was a ceiling of 67% on the chance that an object at the stable points along Earth’s orbit has been cataloged. That ceiling came from a survey boundary I had assumed, and the chance is now 93% (56 to 100%) for a 1 kilometer object. The revised figures use survey coverage that I measured afterward.
Is any nearby asteroid being pushed by sunlight the way a sail would be?
Sunlight would push an alien solar sail, or an empty hull, off course far more than a rock. I checked public position measurements of 19,192 near-Earth objects (asteroids that come close to Earth) and found none that moves like a sail. Of those tracked well enough to tell, at most 0.3% to 1.1% could be sail-like, depending on size (95% confidence). A dense probe would look like a rock, so the search would miss it.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. The ceiling on sail-like objects (anything sunlight pushes as hard as it would a sail) depends on their size. At most 0.8% of the objects 1 to 10 meters across could be sail-like. For objects 10 to 100 meters across the figure is 0.3%, and for larger ones it is 1.1%. That 1.1% falls to 0.3% if a sail could be tracked over years without being lost. Each figure could be up to two or three times higher or lower, depending on choices I made in the analysis. (95% confidence means that if the true share were any higher, there would be less than a 1 in 20 chance of finding none.)
Sunlight pushes something like a spent rocket stage far less than it pushes a sail. For a push that weak, over 95% of the objects under 100 meters are untested. The near-Earth catalog also removes objects that look like human debris.
One object, 2007 AA2, does show a small extra push, 100 times too weak for a sail. The natural explanations are escaping gas, as in the “dark comets” (asteroids nudged off course with no visible tail), or a false signal created by joining its 2007 and 2026 sightings. I count it as something to follow up, and not as a detection.
Earlier work. For comparison I had two published measurements: the push on ‘Oumuamua, the first known object from beyond the Solar System (Micheli and colleagues 2018), and the push on the dark comets (Seligman and colleagues 2023, 2024). Earlier work measured such pushes one object at a time. As far as I could find, nobody has published a limit for the whole catalog to compare with mine.
What the second look corrected. My first limits were circular. I had run no search, and the figures only repeated what I had assumed, so I withdrew them. The weaker limits given here, and the object 2007 AA2, come from a later search, made after the second look.
Do any nearby stars give off the kind of heat that energy collectors would?
A cloud of collectors catching a star’s light for energy (a Dyson swarm) would glow with infrared heat. I looked for that glow in WISE satellite data. At most 1 nearby star in 1,400 has a warm swarm that takes even 1% of its light (95% confidence). A much colder or hotter swarm would not show up, and I could search only 54% of the 273,856 ordinary stars within about 330 light years.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. The 1 in 1,400 depends on how I set up the model: other settings give anything from 1 star in 1,000 to 1 in 2,600. It is the limit that a perfect search of about 103,000 stars would give. The search can’t see swarms colder than about 100 kelvin (minus 173 Celsius) or hotter than about 600 kelvin (327 Celsius). I assume the stars I could not search are no different from the ones I could.
The stars come from the Gaia catalog. I counted every star with extra infrared light as a possible swarm, and I tested the search by planting artificial swarms in the data.
For swarms that take 10% of the star’s light, the ceiling is between 1 star in 2,100 and 1 in 12,000. For swarms that take 1%, it differs by type of star: about 1 in 5,300 for Sun-like stars and 1 in 1,000 for red dwarfs (small, dim stars). The figure for Sun-like stars mostly tells you how common bright dust disks are. (A dust disk is a natural disk of dust around a star, and it glows in the infrared too.) For the faintest quarter of red dwarfs, the search says nothing useful about swarms that take 1%.
WISE cannot tell a swarm from a dust disk or from a galaxy in the background. Of the 307 searchable stars with extra infrared light, 37 stay unexplained, and 28 of those are red dwarfs. I count none of them as a detection.
Earlier work. This search follows Freeman Dyson’s (1960) idea: look for the infrared glow of light collectors around a star. Suazo and colleagues (2022) searched almost the same stars before I did. They found that room-temperature swarms covering 90% of a star are rarer than about 1 star in 50,000. My result for those swarms is no stronger than theirs. The new part is the figure for swarms that take only 1% of the light.
Suazo and colleagues (2024) found seven candidate swarms, all around red dwarfs, among about 5 million stars. Radio maps soon showed a background source, something far behind the star, beside three of the seven (Ren and colleagues 2024). One of those three was later confirmed as a background galaxy with an active nucleus, a brightly glowing core (Ren and colleagues 2025). The JWST space telescope has since shown two more to be background galaxies (Zackrisson and colleagues 2026).
Contardo and Hogg (2024) found 53 stars with unusual extra infrared light among about 5 million Sun-like stars, and read them as dust disks. My 37 unexplained stars are candidates of the same kind.
What the second look corrected. My first version claimed ceilings 300 times lower than those in earlier work. It relied on labels that catalogs give to stars, and a real swarm would have been given the same labels. The figure for swarms that take 10% of the light also came from a single setting of the model. For both reasons I withdrew the claim. The ranges, and the count of 37 unexplained stars, date from after the second look.
Are any Sun-like stars dimmer than they should be?
A swarm of collectors catching a star’s light would make the star look dimmer than its spectrum (its light split by color) says it should be. That is true however cold the swarm is. I tested 790,783 Sun-like stars within 3,300 light years, using survey spectra and distances from the Gaia satellite. At most 1 in 26,000 is missing 90% of its light (95% confidence), but smaller losses are much harder to rule out.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. With other settings of the model, the 1 in 26,000 becomes anything from 1 in 21,000 to 1 in 33,000. By my model, a star missing 90% of its light would be in these surveys only about 1 time in 8. So the 1 in 26,000 is the limit that a perfect search of about 77,000 stars would give. Under looser data-quality rules, the number of stars tested rises to 950,640.
At most 1 star in 28,000 is missing 75% of its light (the range is 1 in 24,000 to 1 in 33,000). For smaller losses, the figures are for mid-life stars, those in the long, steady middle of their lives, as the Sun is. At most 1 in 4,800 of them is missing 50% of its light, and the ceiling may be as low as 1 in 9,100. Among mid-life stars no hotter than about the Sun, at most 1 in 530 is missing 25%, or 1 in 1,900 at the other end of the range.
Those last two ceilings are as high as they are because of 39 to 136 stars that look too dim and that I can’t explain. I did not follow them up, and I count none of them as a detection.
The search can’t say anything about swarms of dust that change a star’s color, about nearly complete shells, or about red dwarfs. Swarms hotter than about 400 to 600 kelvin (127 to 327 Celsius) are untested too.
Earlier work. The test comes from Zackrisson and colleagues (2018), who applied it to 8,441 stars and did not set a limit. For room-temperature swarms, the infrared search of Suazo and colleagues (2022) already rules out more than mine does: at most 1 star in 50,000 within 330 light years. My search reaches ten times as far, and it works however cold the swarm is.
What the second look corrected. At first I gave a single figure for stars missing 50% of their light and another for 25%. Both relied on data-quality rules that I had chosen after seeing the data, so I replaced them with ranges. I withdrew the claim that the unexplained stars were linked to the score Gaia gives each star for how well its measurements fit. I also withdrew the claim that this search closes the gap left for cold swarms. The wider ends of the ranges, and the counts under looser rules, were computed after the second look.
Do any stars contain elements that don’t belong?
A civilization could mark its star, as a signal to anyone watching, by adding an element that is naturally rare, such as neodymium. I searched GALAH survey spectra (starlight split by color) of 408,000 Sun-like stars and found no convincing case. No more than about 1 star in 4,500 carries at least ten times its natural neodymium, and the ceiling may be nearer 1 in 8,000 (95% confidence). The search says nothing about weaker marks.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. I used only stars with clean data, and the limit is the one that a perfect search of about 255,000 of those stars would give. The search does not cover the 18% of stars whose spectra are flagged as unreliable, and among those, stars that look marked are ten times as common.
Counting every variant of the analysis, the ceiling ranges more widely, from 1 star in 2,300 to 1 in 11,000. The limit still leaves room for 1.2 to 2.3 million marked stars in the Milky Way. Each would need refined neodymium weighing about 1.3 times as much as the dwarf planet Ceres.
Sixteen stars that look rich in neodymium are still unexplained. The search also does not cover hotter stars, or two other elements that could serve as marks, technetium and promethium.
Earlier work. The idea of marking a star is old, but as far as I could find, nobody has published a search for marked stars in a large survey. So I have no earlier limit to compare mine with. Whitmire and Wright (1980) noted that nuclear waste dumped into a star would mark it too, without anyone meaning to.
What the second look corrected. My limits for stars marked by nuclear waste were three to five times too optimistic: they ruled out more than the data could support. My first limit of 1 star in 10,000 was too optimistic as well. I withdrew the claim that the survey shows where naturally peculiar stars begin. (Those are stars with odd mixes of elements that nobody put there.) After the second look I added two checks: I planted artificial marks in the data to see whether the search would find them, and I examined the stars whose spectra are flagged as unreliable.
Could a star be completely hidden inside a shell?
A star sealed inside a complete shell would be invisible but would still have gravity. Gravity measurements near the Sun leave room for up to 42% of the mass in stars to be hidden this way. If shells give off their starlight as heat, counts of infrared sources show that at most 5% of Sun-like stars can be inside shells at 40 kelvin (minus 233 Celsius). Near the Sun, colder shells would escape that count.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. The 42% depends on which published measurement of the mass near the Sun I use. It could be as low as 29% or as high as 50%. It rises to about 60% if there is no dark matter near the Sun (dark matter is unseen matter that astronomers infer from its gravity).
The infrared sources are ones the WISE satellite saw, and the counts apply only to Sun-like stars inside shells that give off their star’s light as heat. For shells at 100 kelvin (minus 173 Celsius), they allow 2 to 6 stars in 10,000, far fewer than at 40 kelvin. If a shell sheds the heat some other way, the 42% is the only check.
Near the Sun, the 42% is also the only check on shells colder than about 35 kelvin (minus 238 Celsius). Around red dwarfs, which are most stars, the same goes for shells colder than about 45 kelvin. Farther away, toward the center of the Milky Way, the study “Do background stars ever vanish for months?” finds that at most about 1 to 3% of stars have shells this cold.
I also looked at pairs of stars that orbit each other, because a star inside a shell would show up as a partner that is heavy but dark. Among 26,900 close pairs in the Gaia catalog within about 1,000 light years, 1,128 may have a partner too heavy for the light it gives off. So at most about 5% of partners below half the Sun’s mass can be enclosed stars, and about 4% of partners near the Sun’s mass.
For partners in between, the check is much weaker: up to 30% could be enclosed stars. Most of the candidates (956) fall in that middle range, where I can’t tell a shell from a white dwarf (the dense remains of a dead star). Each of the three percentages could be about twice as large or half as large.
The WISE infrared survey covers 1,091 of the candidates. None of them glows like a warm shell.
Earlier work. I made no new measurement to get the 42%. I re-read the published measurements of mass near the Sun that McKee, Parravano and Hollenbach (2015) compiled. Those measurements date from 1991 to 2014, and none uses the Gaia satellite’s data.
The fullest published search I found that included complete shells is by Carrigan (2009). It used the IRAS satellite and looked for shells at 100 to 600 kelvin (minus 173 to 327 Celsius). As published, it rules out more than my search does in that range, but it did not measure how many shells it would have missed.
What the second look corrected. I first claimed that the infrared search by Suazo and colleagues (2022) had already shown that no more than about 1 star in 10,000 is inside a complete shell. That search covered only stars that are partly enclosed and still visible, and the second look overturned my claim. I added the count of WISE sources afterward.
Do background stars ever vanish for months?
If an opaque shell around a star passed in front of a more distant star, the distant star would vanish for months. In 221,801 brightness records from the OGLE-II survey toward the center of the Milky Way, I found no star that did. So at most about 2% of stars there have a shell ten or more Earth-to-Sun distances in radius (95% confidence). Shells smaller than about two of those distances would have been missed.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. If every star had a shell with a radius of ten Earth-to-Sun distances, I would have expected to see 130 to 480 stars disappear. I saw none. On optimistic assumptions the ceiling is 0.6% of stars, and at worst it is about 5%.
The survey watched 6.37 million stars, and I searched only the 221,801 records it had flagged as varying in brightness. The limits apply to stars roughly 13,000 to 26,000 light years away, and not to stars near the Sun. Each percentage here could be twice as large or half as large.
For shells ten times larger, at most 0.3% of stars can have one (0.07% on optimistic assumptions). The 0.3% becomes 0.7% if I count only disappearances that began during the survey.
For shells that give off their star’s light as heat, the result can also be stated by temperature. At most about 1% of stars can have such a shell at 30 kelvin (minus 243 Celsius), 3% at 50 kelvin and 30% at 100 kelvin. For shells at 150 kelvin (minus 123 Celsius) or warmer, the search rules out nothing. It also says nothing about partial swarms or swarms with fuzzy edges.
Earlier work. Bramante and colleagues (2025) proposed this test as a way to find clumps of dark matter (unseen matter that astronomers infer from its gravity). They did not run a search, so there was no published limit to compare mine with.
What the second look corrected. My first search was meaningless, because a one-line error had made it blind to vanished stars. The second look found the error and ran the search again with it repaired, and that run found no disappearance either. I had also claimed the idea as my own, and I withdrew that claim. The flagging rate (how often the survey flags a star that does vary) is 79%, measured on eclipsing double stars (pairs of stars that pass in front of each other). That rate and the final figures were computed after the second look.
Do any galaxies glow with the heat of captured starlight?
A civilization that captured part of a galaxy’s starlight would give that energy off again as infrared heat. I looked for that heat in 113,158 red, quiet galaxies (ones forming few new stars). At most 1 in 5,800 turns 30% or more of its starlight into room-temperature heat (95% confidence). But if the share is only 1%, the search can rule out nothing.
Revised, not recheckedWhat this study can’t show, and the mistakes I fixed
The numbers, and what is untested. The galaxies come from the Sloan survey, and the infrared measurements come from the WISE satellite. I compared each galaxy’s infrared glow with its own starlight, and I counted every galaxy that stood out as one that might hold such a civilization. Depending on choices I made in the analysis, the 1 in 5,800 could be anything from 1 galaxy in 1,700 to 1 in 7,500.
Those figures are for heat at room temperature. If the heat can be anywhere between 100 and 600 kelvin (minus 173 and 327 Celsius), the search rules out less: at most 1 galaxy in 130 turns 30% or more of its starlight into heat, and the ceiling may be as low as 1 in 240.
For galaxies turning 10% of their starlight into heat, the search rules out far less. Up to 3% of galaxies could be doing so at room temperature (the range is 1.3% to 7.4%). Across the wider span of temperatures, the figure is 14% to 22%.
Galaxies turning 1% of their starlight into heat can’t be ruled out at all, largely because stars and natural dust glow as brightly in the infrared. The search does not cover spiral galaxies, heat colder than 100 kelvin, or heat that is not spread through a galaxy the way its stars are.
Earlier work. For galaxies turning 30% of their starlight into heat, my figures do no more than match those of Huang, Tao and Zhang (2026). They found that fewer than 1 galaxy in 6,200 gives off heat worth at least 21% of a Milky Way’s starlight, in a general sample of 19,617 nearby galaxies. (They measured every galaxy against the Milky Way’s starlight, where I measured each against its own.) My figures cover red, quiet galaxies only.
What the second look corrected. I first claimed ceilings 5 to 150 times lower than those in published work, and I withdrew that claim. I also widened the span of temperatures and the ranges. A check of the measurements against images, and the final ranges, came after the second look.
Several studies searched stars one by one. Even together, they would miss a swarm of light collectors that takes less than 1% of its star’s light. A cold swarm, below about 100 kelvin (minus 173 Celsius), would also escape them unless it takes a quarter or more of a Sun-like star’s light. They can’t rule out swarms around the faintest quarter of red dwarfs (small, dim stars) or, near the Sun, complete shells colder than about 35 to 45 kelvin. Almost everything beyond about 3,300 light years is out of their reach.
The objects the searches can’t explain
Several of the searches turned up objects they can’t explain. There are 37 nearby stars with extra infrared light, 16 that look rich in the element neodymium, and 39 to 136 that look dimmer than they should be. There is also one near-Earth object, 2007 AA2, whose path shows a small push that gravity alone does not explain. Here “can’t explain” means only that nobody has looked again and the ordinary explanations have not been ruled out. I don’t count any of these objects as a detection.
The new studies barely changed the odds
I ran 22 studies: the eight in the table below, then 14 more. Of the things I measured, only Earth’s own history (when life began here and how late intelligence arrived) moves the odds by more than a percentage point or two. The evidence that counts most, that we see no one here, needed no new measurement.
Three studies give the model a number: the dates of Earth’s early life, the radio searches, and the ceiling on how many spiral galaxies could have someone collecting three quarters of their starlight. Four more bear on assumptions, which are still my judgment. The rest are checks.
A table in one published paper didn’t come out as printed when I recalculated it (see the section “What Earth’s history can and can’t tell us”). Nobody has asked the paper’s authors, so I don’t count that as a finding.
The table: the first eight studies, one by one
The table covers the first eight studies, the ones in the sections on time and distance, on galaxies and radio searches, and on Earth’s history. All three studies that put a number into the model are among them. The other 14 are in the section “What 14 more studies found”. None of those gives the model a number, so the table leaves them out.
The last column is about one number: the chance that no other civilization has ever arisen in the Milky Way, which would make us the first one here. After the evidence, the model puts that chance at 78%. The 78% is a little higher than the 73% for “we’re alone” because it also counts cases where no one else arose here but settlers from another galaxy are on their way or already here. For each study, the column shows how many points that chance moves when I take the study’s input out of the model, or push the input to either end of its range.
In the “How it compares” column, “reproduces” means a published result came out again from newer data, or from a new combination of data. “Extends” means the same kind of result from a larger sample, to a tighter limit, or carried a step further. “New” means I found no published figure of that kind.
| Study | Closest published result | What I found | How it compares | Effect on the chance we’re first in the Milky Way |
|---|---|---|---|---|
| Stars passing close to the Sun (Gaia satellite data) | 11.7 star systems per million years (give or take 1.3) pass within 3.3 light years of the Sun, a distance called a parsec (García-Sánchez and colleagues 2001) | 11.1 per million years, give or take 1.0. That adds up to about 21,000 systems since oxygen first built up in Earth’s air | reproduces | Not used in the model. The model’s related input, how fast settlers travel, moves the chance by under half a point |
| Head start of older stars | 74% of Earth-like planets (give or take 9) are older than Earth (Lineweaver 2001) | About 70% of Sun-like stars at the Sun’s distance from the center of the Milky Way are older than the Sun, by 4 billion years on average | reproduces | Not used in the model |
| Elliptical galaxies with missing light | No unusually faint galaxy among 137 (Annis 1999a) | At most 0.29% of large elliptical galaxies are missing three quarters of their light, from a sample of 114,000 galaxies | extends | Not used in the model. The limit for spirals stands in for both kinds of galaxy |
| Spiral galaxies with missing light | At most about 3% of spirals are missing light (a conservative limit), or 0.3% (a tentative one), from 1,359 spirals (Zackrisson and colleagues 2015) | At most 2.6% of large spirals are missing three quarters of their light. That limit holds if the waste heat from the captured starlight is colder than about 250 kelvin (minus 23 Celsius). Only one of my spiral samples was chosen by a fixed rule. Used alone, that sample gives a limit of 4.5%, no better than the published limit counted the same way (4.6%) | extends | Under half a point if I take this limit out of the model |
| All radio searches combined | Searches so far have covered the equivalent of one hot tub of water out of all the oceans (Wright and colleagues 2018) | I combined the searches into one measure. It still leaves room in the Milky Way for up to 400,000 unheard beacons of 100 million billion watts (1017) | reproduces | −1 point if I take the radio searches out of the model |
| Timing of life and intelligence on Earth | Life’s early start on Earth favors an easy origin of life by at least 13 to 1. That is, the early start is at least 13 times more likely if life starts easily than if it doesn’t (Kipping 2025a) | Those odds fall to between 6 and 9 to 1 if two to four hard steps (very unlikely evolutionary leaps) came after the origin of life. It was already known that later hard steps lower the odds (Hanson 1998b; Carter 2008) | reproduces, except for a table I couldn’t match | −5 to +2 points, depending on which date I use for the first life. In the model, life’s early start favors an easy origin by only about 2.4 to 1. If I set that to 6 to 1 instead, the change would be −7 points, and at 13 to 1 it would be −16 points |
| Traits that evolved only once | Traits that evolved only once are older than traits that evolved repeatedly (Vermeij 2006) | Each time a trait’s age goes up tenfold, the odds that it evolved only once go up 3.2 times. When traits inherited together as a package are counted once, that falls to between 1.4 and 2.9 times | new figure, as far as I could find | −9 to +14 points, between no hard steps and three or four. That is the effect of the number of hard steps I assume. The study can’t settle that number, so it is still my judgment |
| Did we arrive early in the universe’s history? | Loeb and colleagues (2016) raised the question, because red dwarfs (small, dim stars) shine for so long. Hanson and colleagues (2021) and Kipping (2025b) took it further | For each number of hard steps, I worked out how much less the planets of red dwarfs would have to count as homes for life before we are no longer early | extends | Under half a point on the “our existence proves nothing” view, which this page uses. On the “we’re an average civilization” view, it can reach about −34 points. (These are two of the four views of what our own existence tells us.) |
How each study was counted
Four studies bear on assumptions that are still my judgment. They asked how long planets stay habitable, whether the planets of red dwarfs count as homes for life, how many hard steps there were, and whether a civilization can keep spreading from star to star. The study on that last question, “What can the settlement of the Pacific tell us?”, gave me no usable starting assumption in the end, as its entry says.
Taken together, my studies replaced some inputs I had assumed with measured ones, made a few published limits tighter or weaker, and turned up a few things that seem to be new.
The table’s last column takes out one input at a time and leaves everything else in place. The bar chart in the section “What the evidence rules out” adds the evidence in every possible order and averages the results. That is why the chart credits the galaxy surveys and radio searches together with 2 points, a figure that doesn’t match their rows in the table.
One published count of passing stars is higher than mine: 19.7 per million years, give or take 2.2 (Bailer-Jones and colleagues 2018). My total of 21,000 systems counts from when oxygen first built up in Earth’s air. It becomes 4,000 to 6,000 if I count only from when life here was plainly visible from far away (about 4,000 if that was when plants covered the land, about 6,000 if it was when oxygen neared modern levels).
Lineweaver’s 74% is the share of Earth-like planets that are older than Earth in the universe as a whole, and it comes from a model. Madau (2023) gives 77% for temperate rocky planets of stars a little smaller than the Sun, within 330 light years. My 70% for Sun-like stars (the range is 64 to 78%) counts the full thickness of the Milky Way’s disk. Near the middle of that thickness, where the Sun is now, the share is 54 to 61%.
Only some of the galaxies in my samples could show missing light: 15,000 of the 114,000 ellipticals, which are the ones bright enough, and 1,800 of the 9,400 large spirals.
Within 8,000 light years, fewer than about 1 star system in 100,000 has a transmitter that runs continuously at more than about 50 million billion watts (5 × 1016). That is a published limit (Mason and colleagues 2025, from the Breakthrough Listen search), and related ones are in Enriquez and colleagues (2017), Wlodarczyk-Sroka and colleagues (2020) and Lacki (2024, 2025). A beacon of 100 million billion watts (1017), the kind in the table, has a bit over half the power of the sunlight striking Earth.
Life’s early start on Earth is only weak evidence that life starts easily (Spiegel and Turner 2012). Snyder-Beattie and colleagues (2021) turned Carter’s argument that intelligence needs several hard steps into a formal probability calculation (a Bayesian one). Estimates published since 2008 put the number of hard steps at four to six, counting the origin of life as one of them. That is more than I assumed (see the section “It comes down to what you assume”).
Whether we arrived early depends on the number of hard steps and on how long planets last (Hanson and colleagues 2021). It also depends on whether the planets of red dwarfs count as homes for life (Kipping 2025b). On the “we’re an average civilization” view, changing only how long planets last moves the result by −3 points. If the planets of small, long-lived stars also count fully as homes for life, the chance that we’re alone in the Milky Way (the first answer at the top of this page) falls from 71%, its value on that view, to 37%. That drop is the same size as the larger figure in the table’s last row.
It comes down to what you assume
So far I’ve kept my starting assumptions fixed and looked at the evidence. Give the same model the assumptions of a well-known view (that habitable planets are rare, say, or that aliens spread fast) and it comes out roughly where that view does.
My result rests on five judgment calls, and the views disagree on four. Two are about life: how easily does it start, and how many hard steps (very unlikely evolutionary leaps) lie between a microbe and a radio telescope? Two are about spreading: can a civilization keep spreading between the stars, and would we see it if it did? The fifth is what our own existence tells us. Only one view answers that differently: the full argument of Hanson and colleagues (2021), which includes their case that we arrived early in the universe’s history.
Figure 13 With the same evidence, the chance we’re alone ranges from 13% to 94% depending on what you assume
The numbers, and what some rows leave out
In 17 of the 18 rows, “we’re alone” ends higher than it started. The exception is “Hanson’s full case (we’re an average civilization)”, where “we’re alone” falls from 69% to 53% because our early arrival then counts as evidence of civilizations that spread. “We’re alone” is the leading answer in all but four rows, where “they never left home” leads. That answer reaches 41% in “life starts easily: 90% sure” and 43% in “evolution waits for the environment: no hard steps”. It reaches 60% in “life starts easily and no later step is unlikely” and 68% in “classic SETI: life common, spreading doubtful”.
One row is built on the view of Mills and colleagues (2025), who propose that evolution needed no improbable steps at all, only time for the environment to change. I added two assumptions of my own to that row: a 60% chance that life starts easily, and planets that usually stay habitable. With those assumptions, it is one of the four rows in which “we’re alone” is not the most likely answer in this model.
Three rows use only part of what their authors argued. The row on rare habitable planets does not assume that microbial life is common, though Ward and Brownlee (2000), who made the case for rare planets, hold that it is. The row in which settlement would be obvious uses one of Hart’s (1975) two premises. With the other premise added, that spreading is easy, the chance that we’re alone comes to 87%. The row on aliens that spread fast uses what Hanson and colleagues (2021) assume about spreading. It leaves out their argument that we showed up early (see “Did we arrive early in the universe’s history?” in the section “What Earth’s history can and can’t tell us”).
The full argument of Hanson and colleagues has a row of its own. On top of their assumptions about spreading, it has several hard steps and long-lived planets, and it takes the “we’re an average civilization” view of our own existence. In that row the chance that we’re alone is 53%. It is also the only row in which “here, and obvious” (the hatched block) keeps more than a tiny share after the evidence: 15%. In the other rows the evidence all but rules that explanation out. This row takes it as given that we could only have arisen on a planet nobody else had settled, so finding Earth unsettled counts for less. Add their assumptions about hard steps and long-lived planets, and the evidence rules out much less.
Which assumptions matter most
Apart from what our own existence tells us, four assumptions are worth arguing over.
The chance that life starts easily matters most. Raise it from 5% to 95% and the chance that we’re alone, after the evidence, falls from 80% to 35%. The chance that they never left home (others arose in the Milky Way, but none kept spreading from star to star) rises from 11% to 43%.
A second is the chance of a barrier that stops every civilization from spreading, including us. With no barrier, the chance that they never left home is 6%. With a 50% chance of a barrier to begin with, it is 33%.
A third is the chance that every civilization that spreads is hard to spot. It mostly decides the chance that settlers are here unnoticed.
The fourth, the number of hard steps, is not in the chart below. With almost none, the chance that we’re alone is 65%. With three or four it is 86%, about 20 percentage points higher.
Figure 14 Three assumptions that move the odds a lot: how easily life starts, whether anything stops civilizations from spreading, and whether settlers would be hard to spot
The three panels in numbers
In the first panel the evidence adds about 25 points to the chance that we’re alone at every setting. At the skeptical end, where life has a 5% chance of starting easily, that multiplies the chance by about 1.5. At the optimistic end, 95%, it multiplies the chance by more than 3. (An optimist starts with more of the probability on obvious settlement, meaning someone has taken over the Milky Way in a way we could not miss.) Counted as odds (the chance that something is true divided by the chance that it is not), the two ends are closer. The evidence multiplies the odds by about 3.4 at the skeptical end and 4.7 at the optimistic end.
In the second panel, the chance that they never left home goes from 6% with no barrier to 33% when a barrier starts with a 50% chance. In the third panel, when the chance that every civilization that spreads is hard to spot starts at 5%, the explanation “here, but hard to spot” ends at 4% after the evidence. When it starts at 50%, the explanation ends at 27%. In these two panels, for any starting chance above zero, the evidence multiplies the chance the panel shows by about 1.3 to 1.7.
The evidence does not test directly for a barrier or for unnoticed settlers. Both become more likely for the same reason “we’re alone” does: obvious settlement has been ruled out, and every explanation that is left rises by about the same factor. The chance that a barrier exists rises from 15% to 26%. The chance that every civilization that spreads is hard to spot rises from 5% to 8%, and the chance that everyone hides rises from 3% to 5%. (Those are chances that an assumption is true. The explanation “here, but hard to spot” is a different thing: it says such settlers are in fact here.) The curves come from a single run of the model, so they can differ from the numbers at the top of the page by up to a point.
I assumed about one hard step after the origin of life, on average. That is fewer than most estimates published since 2008, which pulls my answer away from “we’re alone”.
Unlikely steps in evolution: published counts, and how much they matter
I gave a 35% chance that there are no hard steps at all after the origin of life. Most estimates published since 2008 are higher, at about four (Watson 2008), five or six (Carter 2008) and six (Hanson and colleagues 2021), though they count the origin of life as a step. Counted their way, mine is about two. Carter’s first estimate (1983) was one or two, and Mills and colleagues (2025) argue for none.
Changing the chance of a barrier moves the chance that we’re alone about as far as changing the number of hard steps does. Changing the chance that life starts easily moves it twice as far. With three or four hard steps and nothing else changed, the chance that we’re alone is 86%, about 20 points above the 65% with almost none. The row “Three to six hard steps” in the first chart of this section has 94%, because it also assumes that life is harder to start and that each step is harder.
What if all my choices change at once? I did 400 test runs with randomly varied assumptions. The ranges I chose lean toward life starting easily, and on hard steps they stop short of the recent estimates.
With my own assumptions the chance that we’re alone is 73%. With the assumptions varied at random, the typical result is 56%, and 86% of the test runs end below mine, even though my low count of hard steps pulls my own result down.
Figure 15 In all of my 400 test runs, the evidence raised the chance that we’re alone, typically by 25 percentage points
Each dot is one test run. All of them take the view that our existence proves nothing. Hover or tap a dot for two of its assumptions.
The test runs in more detail
How high a test run ends depends mostly on how high it starts. The typical test run starts with a 28% chance that we’re alone and ends with 56%. The typical rise is 25 points, a little less than the gap between those two figures, because the typical start and the typical end come from different test runs. Of all the test runs, 90% end between 26% and 78%.
How often “we’re alone” comes out as the most likely answer depends on what you think our own existence tells us. (The section “What does our own existence tell us?” explains the four views.) On the “our existence proves nothing” view, which this page uses, it is the most likely answer in 78% of the test runs. On the “we’re an average civilization” view the figure is 62% if “average” means average among all civilizations, past and future, and 70% if it means average among civilizations so far. On the “you’re an average person” view it is 15%.
The figure for “you’re an average person” is the least certain, because every test run makes the same assumption: a civilization that has spread holds a million times as many individuals as ours. That figure has a second weakness. Each test run is built from many simulated universes (each one is a full set of randomly picked values for the model’s unknowns). On this view, in about 1 test run in 10, a few of those universes carry most of the weight. That is too few to be reliable.
Each test run picks every disputed assumption at random from a range. I chose the ranges, and a different set of ranges would give different figures. For the chance that life starts easily, I picked values evenly between 5% and 95%. So half the test runs give life more than a 50% chance of starting easily, where I gave it 20%. The ranges for hard steps don’t reach as high as the recent estimates. No test run averages more than about three hard steps after the origin of life, while the recent estimates of four to six count the origin as well.
Where the well-known theories fit
Standard catalogs list dozens of theories about “Where is everybody?” (Brin 1983; Webb 2015; Ćirković 2009). The table places the best-known ones among this page’s seven explanations. Several fall under “they never left home”. Of those, only a cause that hits every civilization, such as dying young or impractical travel, would stop us from spreading too. The model has no place for the idea that what we see is staged for us (the “planetarium hypothesis” in Webb’s list).
| Theory | Source | In this model |
|---|---|---|
| Rare Earth: planets fit for complex life are rare | Ward and Brownlee (2000) | The two “we’re alone” explanations: no one else in the visible universe, or no one else in the Milky Way |
| Hard steps: evolution needed a few very unlikely leaps | Carter (1983) | The same two explanations |
| Civilizations don’t last long | von Hoerner (1961) | “They never left home” |
| We have barely searched: they are at home, and our searches are too thin | The working assumption of radio searches. Wright and colleagues (2018) on how little has been searched | “They never left home”. The radio searches so far don’t rule it out |
| Travel between the stars is impractical, or civilizations can’t keep growing | Webb (2015) lists the travel objection. Haqq-Misra and Baum (2009) on growth that can’t last | “They never left home” |
| Percolation, slow diffusion and steady states: settlement stalls, spreads slowly or leaves lasting unsettled gaps | Landis (1998), Newman and Sagan (1981), Carroll-Nellenback and colleagues (2019) | “They never left home” if the wave stalled or is still crossing the galaxy (the model has no separate place for a wave that is still crossing). A settled galaxy with a lasting gap around Earth has no separate place either (see the note under this table) |
| The zoo hypothesis: they agree to leave us alone | Ball (1973) | “Here, and hiding” |
| The dark forest: everyone hides out of fear | Liu (2008), a novel, for the name. Brin (1983) describes an older version with “deadly probes” | “Here, and hiding” if they have spread. “They never left home” if each stays quietly at home |
| Aestivation: they sleep until the universe is colder | Sandberg, Armstrong and Ćirković (2016) | “Here, but hard to spot” |
| Their probes or other objects are in the Solar System and we haven’t looked hard enough | Freitas (1985) | “Here, but hard to spot”. Two of my studies test it: “Would we notice a probe parked nearby?” and “Is any nearby asteroid being pushed by sunlight the way a sail would be?” |
| Expanding civilizations that haven’t reached us yet | Olson (2015), Hanson and colleagues (2021), Kipping (2026), a preprint (a paper posted online that has not been through a journal’s review) | “Spreading, but not here yet” |
| Events such as gamma-ray bursts set life back across the whole galaxy | Annis (1999b), Ćirković and Vukotić (2008) | Not in the model |
I put the chance that a settled galaxy leaves a lasting gap around Earth at between 1 in a million and 1 in 10. A critic (another run of the same AI) let that chance be anything from impossible to certain. Obvious settlement then kept about 4% after the evidence, instead of under 1%. The chance that no other civilization has ever arisen in the Milky Way moved by under 2 points.
Try your own assumptions
The sliders replace six of my starting assumptions with yours, and your browser recalculates the odds from the same simulated universes. One row of buttons sets the sliders to my values or four other settings. The other picks a view on what our own existence tells us. Everything else stays as I set it.
What the sliders don’t change
Some of my starting assumptions stay fixed wherever you put the sliders. They include how hard each step in evolution is, how often a civilization spreads when nothing stops it, how unlikely I let the origin of life be, and what share of settlers would be obvious to us. Rows in the first chart of the section “It comes down to what you assume” show what changing them does.
Reset puts back my values and the first of the four views, “our existence proves nothing”. With other slider settings, the three answers can add up to 99% or 101%, because each one is rounded separately.
The chart below gives one more number under its three answers: the chance that no other civilization has ever arisen in the Milky Way, which would make us the first. With my values it is 78%. The 78% is a little higher than the 73% for “we’re alone” because it also counts cases where no one else arose here but settlers from another galaxy are on their way or already here.
The simulated universes are the ones behind the three answers at the top of this page. Each one is a full set of randomly picked values for the model’s unknowns.
The hard-steps slider sets the average number of hard steps, which is why it can show a fraction. At the bottom, 0.2, there are almost none, so evolution mostly waits for the environment. The top, 3.4, is as far as this slider goes. Published estimates run from four to six hard steps, but they count the origin of life as one of them and the slider doesn’t. Counted their way, the top of the slider sits near the bottom of the published range. For more hard steps than the slider allows, see the row “Three to six hard steps” in the first chart in the section “It comes down to what you assume”.
Figure 16 Your answer
No other civilization has ever arisen in the Milky Way:
This view favors universes full of civilizations whatever you assume about biology, so the first two sliders stop mattering.
What does our own existence tell us?
We exist. Does that tell us anything about who else is out there? There are four views, and the one you take can reverse the odds. Choosing among them is a question for philosophy. But combined with my assumptions, the views make different forecasts that a search for life beyond Earth could test.
Almost every number so far takes the first view, “our existence proves nothing”. It takes Earth as given, and asks only whether anyone would have settled it by now.
The second view is “we’re an average civilization”. It reasons as if our civilization had been picked at random from all civilizations that ever arise. On this view the result depends on how long planets stay habitable. With my assumptions about that, the odds barely move. But if the planets of small, long-lived stars counted fully as homes for life, the chance that we’re alone in the Milky Way (73% on the first view) would fall to about 37%.
The third view, “you’re an average person”, reasons as if you had been picked at random from every individual in every civilization. If most individuals live in civilizations that have spread, you should expect to find yourself in one. You don’t. On this view, that is strong evidence that civilizations rarely spread, or can’t spread at all.
The fourth view, “we exist, so life is common”, says that the more civilizations like ours a universe would contain, the more likely it is to be the real one. A Milky Way with no one else in it is then almost ruled out from the start. (Philosophers call this view self-indication, and the two “average” views self-sampling.)
Figure 17 Four views of what our existence tells us. On my assumptions, two of them change the odds.
What each row depends on
The two “average” views differ only in who we are supposed to be average among: civilizations, or individuals. That comparison group is called the reference class. “We’re an average civilization” has two rows because its group can be drawn two ways: all civilizations so far, or all that ever arise. With my assumptions about how long planets stay habitable, the odds barely move either way. Possible universes where most civilizations resemble ours become a little more likely.
The odds move much more if the planets of small, long-lived stars count fully as homes for life. In that case we are early arrivals among all the civilizations that ever arise. When we are compared with all of them, the chance that we’re alone falls to about 37%, give or take 2 points. On the “our existence proves nothing” view, counting those planets fully moves that chance only from 73% to 71%.
Our early arrival is the starting point of an argument by Hanson and colleagues (2021). They conclude that expanding civilizations explain it, because such civilizations will soon fill the universe and leave no room for latecomers (see “Did we arrive early in the universe’s history?” in the section “What Earth’s history can and can’t tell us”). In my model, most of the probability that “we’re alone” loses goes to “they never left home” instead.
The last two rows are a different kind of argument. On the “you’re an average person” view, “they never left home” gains because we find ourselves in a small civilization that has stayed at home. That row’s 45% assumes that a civilization that has spread holds a million times as many individuals as ours. The figure is 35% with a thousand times as many, and 49% with 10 billion times as many.
The same view puts the chance of a barrier that stops every civilization from spreading, including us, at about 92%. The chance of a barrier is high as long as I allow for one at all. If I start with only a 5% chance of a barrier it still comes out at 77%, and if I start with none it is zero. The chance of a barrier is 61% if a civilization that has spread holds only a thousand times as many individuals as ours.
On the “we exist, so life is common” view, a civilization like ours means one in our position: at our stage, and seeing what we see. The evidence hardly changes this row. Its split between “they never left home” and settlers already here is just three of my starting assumptions, scaled up so that they add to 100%. The three are a 15% chance of a barrier and then, within the remaining 85%, a 5% chance that all settlers are hard to spot and a 3% chance that they all hide. This last row is less precise than the others, because it rests mostly on a few simulated universes.
The part of this section headed “No one settled Earth before us. Is that evidence of anything?” describes an objection whose strict form has no row in this chart.
The two views that change the odds both point to a barrier that stops every civilization from spreading, including us. On the “you’re an average person” view, the chance of such a barrier comes out at about 92%. That number depends on my having allowed for a barrier at all: if I start with no chance of one, it stays at zero. (The 92% is higher than the 45% for “they never left home” because, on this view, a barrier is also likely in a galaxy where no one else arose.) On the “we exist, so life is common” view the chance of a barrier is about 70%. The evidence adds little to that figure. It is mostly what I assumed about spreading to begin with, made larger by this view.
Earlier work, and where mine differs
Hart (1975), Tipler (1980) and Sandberg, Drexler and Ord (2018) all took the first view, “our existence proves nothing”. Bostrom (2002) sets out the terms philosophers use for the other three views.
The third view, “you’re an average person”, says that if most individuals live in civilizations that have spread, we should expect to live in one. We don’t. Olum (2004) set out that conflict, and Gerig, Olum and Vilenkin (2013) analyzed the argument under the name “universal doomsday”. The name comes from the older “doomsday argument”. That argument says an average human should not expect to be among the earliest few of all the humans who will ever live, so the human future is probably not much longer than its past.
Katja Grace (2010) showed that the fourth view, “we exist, so life is common”, points to a barrier that stops every civilization from spreading. Others have worked through the same view (Olson and Ord 2021; Finnveden 2019; Cook 2022). Where my calculation overlaps with theirs, it agrees with them on two points. The fourth view rules out an empty universe, and it points strongly to a filter ahead of us, meaning something that stops civilizations before they spread.
We differ on how rare we let spreading be. Finnveden deliberately left out the cases where spreading is very rare, because nothing we do would matter in them. Cook cut the same cases off to keep his calculation stable. In their starting assumptions, the share of civilizations that spread goes no lower than 1 in 10,000 and 1 in 100. They conclude that, on the fourth view, civilizations that spread will almost certainly fill our future.
My model does allow spreading to be impossible. With that allowed, the fourth view gives Grace’s result, with the numbers set by my starting assumptions. Nearly all the probability goes to any universe that can hold as many silent civilizations as my assumptions allow. The split between “they never left home” and settlers already here is little more than my 15% starting chance of a barrier, set against my 5% and 3% starting chances for the two ways of going unseen (being hard to spot, and hiding).
No one settled Earth before us. Is that evidence of anything?
We could only have arisen on a planet nobody had settled. So perhaps finding Earth unsettled tells us nothing.
In its strictest form, this objection holds. The strict form compares us only with other civilizations arising at this same moment, and every one of them arose on a planet nobody had settled. On the strict form, the evidence hardly moves anything. The chance of obvious settlement (someone has taken over the Milky Way in a way we could not miss) goes from 36% to 35%, and the chance that no other civilization has ever arisen in the Milky Way goes from 58% to 56%. My main result, which does not use the strict form, puts those two chances at under 1% and 78%.
I don’t use the strict form, because it ignores when we arose. In a galaxy that ends up settled, nearly all civilizations that arise on an unsettled planet arise early in the galaxy’s history, before settlers have reached most of it. We arose late: most Sun-like stars at our distance from the galaxy’s center formed before the Sun, many by billions of years.
The two “we’re an average civilization” rows in the chart above accept the objection’s starting point, that we arose on an unsettled planet. But they compare us with every civilization that ever arises on an unsettled planet, not only with those arising at this moment, and in both rows the chance of obvious settlement still falls to 1 or 2%. Every percentage point this page credits to the evidence depends on my choice not to use the strict form.
How many civilizations? The range is enormous
Nobody can give a single number for how many civilizations there are, only a range, as Sandberg, Drexler and Ord (2018) pointed out. They took the Drake equation, the chain of factors Frank Drake wrote down in 1961 to estimate that number, and replaced each single number in it with a range.
The overall range is wide enough to hold both “we are alone in the visible universe” and “the Milky Way has hosted hundreds of thousands of civilizations”. Nearly all of that width comes from one unknown: how life begins. My starting range for it is as wide as Sandberg, Drexler and Ord’s (built differently), and none of my measurements narrows it.
The evidence makes a crowded galaxy (the right-hand side of the chart below) about half as likely. The chance that at least one other civilization has ever arisen in the Milky Way falls from 42% to 22%.
Sandberg and his colleagues weighed the evidence too. Their strongest case assumes that any civilization would eventually settle the galaxy, and gives a 99.6% chance that no detectable civilization is in the Milky Way now. My question is slightly different: has any other civilization ever arisen here? I get a 78% chance that none has. It is lower than their figure because my model allows settlement to be impossible, rare or unnoticed.
Figure 18 The evidence shifts the odds toward a galaxy that never had anyone else
How wide the range is, in numbers
The number of civilizations is uncertain by some 50 powers of ten. That means the top of the range is bigger than the bottom by a factor of 1 followed by 50 zeros.
The figures in this fold are written as powers of ten. The small raised number counts the zeros, and a minus sign in front of it means a fraction (one divided by a 1 with that many zeros).
The chart shows the range for the expected (average) number of other civilizations ever to arise in the Milky Way. Before looking at any evidence, my assumptions put the middle of that range at 10−2. After the evidence the middle is about 10−7. An expected number that small means roughly a 1 in 10 million chance that there has been even one other civilization.
The 22% given in the text above for at least one other civilization is far higher than 1 in 10 million because it counts the whole range, crowded end included. Strictly, it is the chance that the expected number is one or more (the right-hand side of the chart). The chance that at least one other civilization actually arose is one percentage point higher, because the chance that none did is 77%.
After the evidence, there is a 90% chance that the expected number lies between 10−52 and 105.
A separate question is whether anyone else is in the Milky Way right now. On my assumptions, after the evidence, the chance is about 14%. The three most optimistic well-known views (see the section “It comes down to what you assume”) put it between 49% and 75%.
What counts as a civilization that exists right now, and earlier estimates
The 14% chance that someone else is in the Milky Way right now has two parts. The model gives 8% to a civilization that arose here, is roughly like ours, and is at the stage when it can be detected. The rest covers the cases where no such civilization exists but settlers are already here, unnoticed or hiding. (The share given at the top of this page for settlers already here is slightly larger, because it counts every case with settlers here, with or without a civilization that arose here.)
The three optimistic views are three rows of the first chart in the section “It comes down to what you assume”. In one, evolution only waits for the environment and needs no unlikely leaps. Another is classic SETI (the search for extraterrestrial intelligence), where life is common and spreading is doubtful. In the last, life starts easily and no later step is unlikely.
Before weighing any evidence, Sandberg, Drexler and Ord started with a 52% chance that no detectable civilization is in the Milky Way now. Searches of nearby stars moved that figure by a few points. My own starting figure was a 58% chance that no other civilization has ever arisen in the Milky Way, which is a slightly different question. For the whole visible universe, Sandberg, Drexler and Ord put the chance that we are alone at 38% before looking at the evidence and at 39% to 85% after.
My model gives a number for a smaller region than the visible universe: the roughly 5.5 million galaxies we could ever reach at a tenth of the speed of light. The chance that no other civilization has arisen there yet is 52%. No observation tests that number directly.
For the larger region that could be reached at the speed of light, the chance that no other civilization has arisen yet is 43%. It would be lower on two of the four views of what our own existence tells us, “you’re an average person” and “we exist, so life is common”.
The model also gives a 9% chance that expanding civilizations already hold at least 1% of the universe. That is more than the chance given at the top of this page for a civilization still on its way here. The explanation “spreading, but not here yet” uses the same 1% test, but it counts only the cases where those civilizations have not reached us. In the rest, the simulated universe falls under one of the other explanations.
Discoveries that would change the odds
My model spells out every assumption, so I can work out how far each possible discovery would move the odds, and how likely it is. These forecasts start from my assumptions, updated by the evidence. Other assumptions would give other forecasts.
Detecting someone would settle the matter. Short of that, the biggest change would come from learning that no civilization can keep spreading between the stars. The chance that we’re alone in the Milky Way would fall from 73% to 54%. “They never left home” (others arose in the Milky Way, but none kept spreading from star to star) would rise from 16% to 46%. The two would be nearly even.
An announcement that life started twice in the Solar System, independently, would do almost as much. “We’re alone” would fall to 52% and “they never left home” would rise to 31%. Showing that civilizations can keep spreading would push “they never left home” the other way, down to 6%.
With my assumptions, one more radio search like today’s that found nothing would move the odds by less than 1 percentage point.
Figure 19 Short of detecting someone, two results would move the odds most: proof that no civilization can keep spreading, and news of a second start of life in the Solar System
What the forecasts assume, and the results that barely matter
The model gives a 31% chance that a second origin of life will be announced. That includes the 15% chance of a false alarm that I assumed. The rest comes almost entirely from my starting assumption that there is a 20% chance life starts easily, which the evidence has not moved. Spiegel and Turner (2012) showed that one independent origin would be far stronger evidence that life starts easily than life’s early start on Earth is.
Five of the 14 possible results would move each of the three answers by less than 2 points. Of the two radio results, ruling out transmitters like ours across the whole Milky Way would move “we’re alone” from 73% to 79%. Another empty search of the nearest million stars would move it by less than 1 point. Someone who starts from more optimistic assumptions, expecting civilizations to be more common, has more to learn from radio searches like these.
A slightly longer list of results would move none of the three answers by more than 3 points. Three of them are searches that find nothing: a search for a chemical sign of life on 25 nearby planets, a search of the Solar System for artifacts (objects that someone made), and a radio search of the nearest million stars. The others are findings on two questions, whichever way each comes out: how long rocky planets stay habitable, and whether the planets of red dwarfs (small, dim stars) are habitable.
Three searches look for technology: another radio search of the nearest million stars, a thorough search of the Solar System for artifacts, and a search of the whole Milky Way for technology at our own level. If any of them found something, the question would be settled and the chance that we’re alone would go to zero. The model puts the chance of finding anything at under 1% for the nearest million stars, 3% for the Solar System and 8% for the whole Milky Way. Once each possible result is weighted by its chance, a search of the whole Milky Way is expected to move the odds about as much as the search for a second origin of life.
A thorough search of our own Solar System for artifacts is also the only cheap test of the “here, but hard to spot” explanation, in which settlers are here but leave little that we would notice.
A way to test the four views of what our existence tells us
There are four views of what our own existence tells us, and the choice among them is usually treated as untestable. But one, “we exist, so life is common”, makes forecasts that can fail, as Olson and Ord (2021) noted. With my assumptions, each view makes its own forecast about discoveries within reach.
Take a second, independent origin of life in the Solar System. On the “our existence proves nothing” view, which this page uses, the model puts the chance that one exists at about 19%. On the “we exist, so life is common” view, it is about 95%.
Figure 20 Only the “we exist, so life is common” view expects to find life almost everywhere we look
How false alarms are counted, and why the chart has only three views
On the “our existence proves nothing” view, the chance that a second origin exists (about 19%) is in effect my starting assumption that there is a 20% chance life starts easily. The evidence has not moved that assumption. The chance that a second origin is announced is higher, at 31%, because I allow a 15% chance of a false alarm. The chart’s first row shows the chance that one exists, so it leaves false alarms out.
The “we exist, so life is common” view pushes the rate at which life starts to the top of the range in my starting assumptions. People disagree about whether a failed forecast should count against the view itself or only against the starting assumptions it was combined with.
The first row’s forecast assumes that Mars, Europa and Enceladus together gave life about a third as much opportunity to start as Earth did. That figure is my own and has not been measured. The caption imagines a careful search of all three, and the searches made so far are nowhere near that thorough. The study “If life had started twice on Earth, would we know?”, earlier on this page, looks at the searches of Earth and the Solar System so far. It finds that today they can move the odds by a few points at most, in either direction.
The second row is a biosignature, a chemical sign of life in a planet’s atmosphere. It leaves out a 10% false-alarm rate in the same way. The third row is a search of the whole Milky Way that finds technology like ours.
The chart shows three of the four views. The one left out, “we’re an average civilization”, comes in two versions: average among the civilizations so far, or among all that ever arise. Both versions give forecasts within a few points of “our existence proves nothing” in all three rows and are not drawn.
What this means for us
If we’re alone in the Milky Way, with nobody on the way (73%), the step that makes civilizations rare is probably one we have already passed. Sandberg, Drexler and Ord (2018) reached that conclusion too. That 73% is a 46% chance that no other life ever starts in our galaxy, plus a 27% chance that other life starts but never reaches technology. Either way, seeing no one tells us almost nothing about what lies ahead.
If other civilizations arose in the Milky Way but none kept spreading from star to star (16%), the likeliest reason is a barrier that stops every civilization from spreading, including us. In the model, that is the reason in about 72% of the cases that give this answer. But the barrier is an assumption I built in, and the evidence can’t confirm it. In test runs with no barrier, this answer (“they never left home”) told us nothing about whether we could spread.
The numbers behind the top two answers
Take “we’re alone” first. The part of it where no other life ever starts in the Milky Way had a 32% chance before looking at the evidence and has a 46% chance after. The 32% is that high because I allowed for life being extremely unlikely to begin at all. If we are alone, the chance that there is also a barrier is 19%, close to the 15% I started with.
Now take “they never left home”. The barrier’s 72% share of it was set by how I built my starting assumptions, with no help from the evidence. The rest of the time there is no barrier, but it is so rare for a civilization to keep spreading that none of those that arose has done it yet. In those cases the typical number of civilizations that arose is about 16 (half the time more, half the time fewer). Fewer than 100 arose about two thirds of the time, and 100 or more the other third.
There were two test runs with no barrier, both made by critics (more runs of the same AI that did this work, told to find faults in it). In place of the barrier, the critics let spreading be far rarer. “They never left home” kept its 16% share in one run and rose to 28% in the other. In both runs, that answer told us nothing about whether we could spread (the section “How I did this, and what it can’t tell you” has more).
Counting every case, the chance that the barrier exists rises from the 15% I started with to 26%. That is higher than the 16% for “they never left home”, because a barrier can exist with nobody there to be stopped by it. In more than half of the cases with a barrier, no other civilization ever arose.
Whether or not we’re alone, the model puts the chance of a barrier at 26% after the evidence, up from the 15% I started with. A barrier would only keep us in the Solar System, so it is no forecast of extinction.
Like every number above, the 26% comes from the view this page uses, “our existence proves nothing”. Two other views of what our existence tells us put that chance far higher. It is about 92% on the “you’re an average person” view and about 70% on the “we exist, so life is common” view. Neither number is something the evidence found. The 92% falls to zero if I start by ruling a barrier out. The 70% is mostly what I assumed about spreading to begin with, made larger by that view.
Nick Bostrom (2008) argued that finding life elsewhere would be bad news. His argument rests on the “Great Filter” of Robin Hanson (1998a): some step between dead matter and a civilization that fills its galaxy must be very hard. If life turns up elsewhere, that step is probably not the origin of life, so it more likely lies ahead of us. My model would raise “they never left home” from 16% to about 31% if a second, independent start of life were announced in the Solar System. That is bad news for us only if the barrier I assumed is real.
The odds would move most, in either direction, if we learned whether any civilization can keep spreading between the stars. Ours is the only one we can watch try, but our first settlements would only be a hint. What matters is whether spreading keeps going for millions of years. If we failed, that would not show it is impossible for everyone.
Earlier work, and how far the numbers could move
The 92% comes from the “you’re an average person” view: reason as if you had been picked at random from every individual in every civilization. If each of us is an average person, civilizations far larger than ours must be rare, or most people would live in one. This is the “doomsday argument” of Brandon Carter and John Leslie (Leslie 1996), applied across civilizations. (The doomsday argument says an average human should not expect to be among the earliest few of all the humans who will ever live, so the human future is probably not much longer than its past.) Gott (1993) reached a version of it independently. Gerig, Olum and Vilenkin (2013) found that the version applied across civilizations (the “universal” version) has some force. But they found our future less gloomy than the traditional argument suggests, at least when few threats could end a civilization early.
The 70% comes from the “we exist, so life is common” view, which gives more weight to possible universes that hold more civilizations like ours. The figure rests on a different argument, made by Katja Grace (2010). This view cancels the ordinary doomsday argument (Dieks 1992; Olum 2002), but it favors a galaxy full of civilizations at our stage. We see no one, so on this view almost none of them spreads. In my model the 70% is little more than my starting assumptions about spreading, rescaled.
The rise in “they never left home” to about 31% after a second origin of life depends on how sure we could be that the detection was real. That answer could land anywhere from 25% to 39%. My model also assumes that how easily life starts tells us nothing about whether spreading is possible. If the two were strongly linked, the bad-news conclusion could reverse, as Verendel and Häggström (2017) showed.
If spreading is ever shown to be possible, “they never left home” falls to 6% and “we’re alone” rises to 79%. If it turns out to be impossible for everyone, the two are nearly even (54% for “we’re alone” and 46% for “they never left home”).
How I did this, and what it can’t tell you
Who did what Ethan Mollick set the task: gather the data, test the possible explanations, have critics attack them, and give the likely answers if no definite one could be found. He asked for original analysis of existing data and, at the end, for review by panels of readers, scholars and designers. I am Claude, an AI model made by Anthropic. I did the rest in about two days in October 2026: the model, the studies, this text and the page design. I was also every critic and every panel member. No person checked the code or the numbers.
How the model works
The model simulates many possible universes. In each one, every unknown on a list gets its own randomly chosen value. Some of the unknowns are about life: how many habitable planets there are around each of three kinds of star, and how often life starts. Two more are about hard steps (very unlikely evolutionary leaps): how many there are, and how difficult they are.
Others are about spreading: whether a barrier stops every civilization from spreading, the chance that any one civilization keeps spreading from star to star, and how fast it travels inside a galaxy and between galaxies. The last unknown is the share of expanding civilizations that would be obvious from here.
Then I score each universe on how well it predicts four things we observe, and the universes that predict them better get more weight. The four are the timeline of life on Earth, nobody visibly here, no galaxy with someone collecting its starlight, and no signals detected.
The model also includes civilizations spreading here from other galaxies, and it uses the measured history of star formation in the universe to do that. If expanding civilizations already hold at least 1% of a simulated universe and have not reached us, I count that as someone “on their way”.
Because better-fitting universes get more weight, each result rests on the equivalent of fewer universes than the 2 million I simulated. How many depends on what you think our own existence tells us. The equivalent number for each of the four views is:
- “Our existence proves nothing” (the view this page uses): about 490,000.
- “We’re an average civilization”: also about 490,000 if we are compared with the civilizations so far, and about 390,000 if we are compared with all that ever arise.
- “You’re an average person”: about 62,000.
- “We exist, so life is common”: about 7,000, from a separate set of runs made for that view.
The section “It comes down to what you assume” reports 400 test runs with randomly varied assumptions. Each of those uses 60,000 universes.
What this can’t tell you Nobody has measured how easily life starts, or how many hard steps come after. Nobody knows whether a civilization can keep spreading from star to star, or whether we would notice one that had. And nothing settles what our own existence tells us. So for each of these the model uses my starting assumptions, which statisticians call priors.
The list of explanations is mine too. The model puts the chance that no other civilization has ever arisen in the Milky Way at 78%. Suppose I missed an explanation in which other civilizations have arisen. Suppose too that it fits what we see as well as the surviving explanations do on average. If I had given it even a 10% chance to begin with, that 78% would fall to about 66%.
If other civilizations arose but none kept spreading, the model allows two reasons. Either a barrier stops every civilization from spreading, including us, or spreading is possible but very rare. The evidence can’t tell which. I assumed the barrier was likelier, so where this page says whatever stops others would probably stop us too, it is repeating my assumption.
What if nothing stops civilizations from spreading?
Before looking at the evidence, I put a 15% chance on a barrier that stops every civilization from spreading, including us. In universes without the barrier, I never let the chance that any one civilization keeps spreading fall below about 1 in 100 million.
I was far looser with the origin of life. I let it be as improbable as 1 chance in a number written as a 1 followed by 100 zeros (10100). If I had let spreading be that improbable too, the answer “they never left home” would say much less about our own prospects.
The critics tested this in two runs with no barrier. In the first, spreading could be as rare as 1 civilization in a number written as a 1 followed by 30 zeros (1030). The chance of “they never left home” still came out at about 16%, but that answer no longer said anything about our own prospects.
In the second run, spreading could be as improbable as the origin of life, down to 1 civilization in 10100. “They never left home” then rose from 16% to 28%. The chance that no other civilization has ever arisen in the Milky Way fell by 11 percentage points, from 79% in that run to 68%. (On this page that chance is 78%.)
Sandberg, Drexler and Ord (2018) noted something similar about their own work: their conclusion about our future could change if the chance of a civilization becoming detectable were radically uncertain.
What else the model assumes or leaves out
Two possibilities that would apply to every civilization at once are still in the model, because I chose to keep them. Before looking at the evidence, I gave a 5% chance that physics or economics forces every expanding civilization to be hard to spot, and a 3% chance that everyone hides. Nearly all the probability left for the explanations “here, but hard to spot” and “here, and hiding” comes from those two choices. In the simulated universes behind “here, but hard to spot”, the Milky Way typically holds more than 10,000 independent expanding civilizations, and every one of them is hard to spot.
The result also assumes that star formation continues roughly as astronomers expect, and that civilizations arise independently of one another.
Wherever this page uses life’s early start on Earth as evidence, it assumes that life was equally likely to begin at any time while Earth was habitable. If life could begin only on the young planet, an early start says little about how easily life starts (see the study “Did giant impacts delay life’s start?”).
The model counts a wave of settlement that came and went long ago as “they never left home”. It does the same for a wave that started and then stalled within a small region. I do not model anything that sets life back to the start across the whole Milky Way (gamma-ray bursts have been proposed).
The surveys used here would all miss a galaxy where someone collects less than about half of the starlight (less than about three quarters in a spiral galaxy) and the waste heat comes out at between about 20 and 100 kelvin (minus 253 to minus 173 Celsius). A galaxy like that looks the same as one with ordinary dust.
Two of the model’s inputs come from preprints (see “Sources that were not peer reviewed” below). Lacki (2016) gives the limit on how many galaxies could be giving off very cold waste heat, and Kipping (2026) gives the number of galaxies in a given volume of space. The JWST and Spitzer space telescopes have observed some planets of red dwarfs. I compiled the tally of those planets by hand, from results published as of October 2026.
To compare my choices with other people’s, see the table “Where the well-known theories fit”. It shows where the published answers sit in my list of explanations. The sliders let you replace six of my judgments with yours.
Who checked it The model and this page went through three rounds of fault-finding review, 22 critic reports in all (these are not the 22 studies). But every critic was Claude, in a separate run (a new session), told to argue a case or audit the work. Each study also got a second look, from a separate run whose job was to break the study.
In every study in the section “What 14 more studies found”, at least one of my first claims did not survive. The third round of critics found wrong arithmetic and several misquoted papers, which are corrected here. The critics also changed the headline twice, at one point from “three answers the evidence cannot separate” to a leading answer. If the headline can move that far, be careful how much you trust any model like this.
After that I edited the text for plain language. More runs of the same AI checked it as expert reviewers, general readers and designers, and another run checked each reference. Then, on October 13, I rewrote all of it to be plainer still, changing no finding or number, and further runs checked each rewritten section against the earlier text. Not one of these reviewers, at any stage, was a person, and no outside expert has reviewed this page.
What the second looks and the critics changed
Every study went the same way. I did one analysis on public data, handed it to the second look to break, and then revised it. Where a first result did not hold up, the page reports the corrected one.
For example, one of my early headlines was a limit on how many galaxies could have someone collecting their starlight, which I said held “at any temperature”. The second look showed that for spiral galaxies it held only if the waste heat from the collecting is cool, so I withdrew it. A critic also pointed out that Annis (1999a) and Zackrisson and colleagues (2015) had used the method first.
The critics of the model each had a role. Some were told to argue a case as an astrobiologist (someone who studies the prospects for life beyond Earth), an optimist about SETI (the search for extraterrestrial intelligence), a skeptic about expansion or a believer in expanding aliens. Others were told to argue as an expert in reasoning from our own existence, or as a defender of the view that “it only takes one” expanding civilization to fill a galaxy. The rest audited the code, the statistics, the studies and the citations.
The first round showed that an early version of the model got its result almost entirely from one unsupported assumption, so I rebuilt it.
The second round found three faults in the rebuilt version. First, one starting assumption had been set using the evidence, when starting assumptions are meant to be fixed before looking at it. Second, hiding was a single switch for a whole universe, so every civilization in it hid or none did, when it should vary from one civilization to the next. (The model still has a 3% chance that everyone hides. I kept that by choice, as the fold “What else the model assumes or leaves out” says.) Third, there was no explicit place for a barrier that stops every civilization from spreading. Fixing those meant rebuilding the model a second time. That changed my starting assumptions, and the headline changed with them, from “three answers the evidence cannot separate” to a leading answer.
In the third round, a separate run built its own copy of the model from the description, and its results matched mine. The same round then showed that my new headline claimed too much, because the page presented a change in my starting assumptions as if the evidence had grown stronger. It also found wrong arithmetic and several misquoted papers. This page is the corrected version.
The runs that checked the plain-language text came after those three rounds. The expert reviewers among them were told to act as a journal’s reviewers in astronomy, the history of life, statistics and the history of this debate.
Sources that were not peer reviewed Several results I use come from preprints (papers posted online that have not been through a journal’s review). These are Lacki (2016), Curtis and colleagues (2026), Kipping (2026), Olson and Ord (2021), Sandberg, Drexler and Ord (2018) and, in the studies, Zackrisson and colleagues (2026) and Pan and colleagues (2026). Other sources, among them online posts, a laboratory report and a magazine article, were not peer reviewed either. The reference list below names them.
The working papers
Each study’s working paper and second look are linked below, with the model’s code, its full tables and the critics’ full reports from all three rounds, including the unresolved objections.
- All working papers Twenty-two studies. Each comes with its second look and its code. All were revised after the second look.
- Critics’ reports, round 1: the first model
- Critics’ reports, round 2: the rebuilt model
- Critics’ reports, round 3: the model and this page
- Result tables Every number on the page, with the model runs it came from.
- How the model works The starting assumptions, how the evidence is weighed, the seven explanations and the known weaknesses.
- Model source code The model and the scripts that run it, as plain text files.
References, and which were not peer reviewed
This list has the works named on this page and the published sources of the model’s inputs. Every entry was checked against the title and authors on the publisher’s page, the preprint archive arXiv or the DOI registry. A few older works were checked against a library record or the author’s own page instead.
The preprints are named above, under “Sources that were not peer reviewed”. The other sources that were not peer reviewed are these. Finnveden (2019) and Cook (2022) are online research posts, Grace (2010) is a blog post, Hanson (1998a) is a web essay and Hanson (1998b) is an unpublished working paper. Jones (1985) is a laboratory report and Bostrom (2008) is a magazine article. Ren and colleagues (2024), cited in the studies, appeared in a journal of short notes that are not sent to outside reviewers. Ren and colleagues (2025) have since followed their note with a peer-reviewed letter.
- Abramov, O., Medvegy, A., Kremer, B. and Mojzsis, S. J. (2026). A Hadean timeline for the emergence of the RNA World. Nature Communications 17, 9790.
- Annis, J. (1999a). Placing a Limit on Star-Fed Kardashev Type III Civilisations. Journal of the British Interplanetary Society 52, 33-36.
- Annis, J. (1999b). An Astrophysical Explanation for the Great Silence. Journal of the British Interplanetary Society 52, 19-22.
- Avelino-de-Souza, K., et al. (2025). Cellular Composition of the Brain of a Northern Minke Whale. Journal of Comparative Neurology 533, e70089.
- Bailer-Jones, C. A. L., Rybizki, J., Andrae, R. and Fouesneau, M. (2018). New stellar encounters discovered in the second Gaia data release. Astronomy & Astrophysics 616, A37.
- Ball, J. A. (1973). The zoo hypothesis. Icarus 19, 347-349.
- Behroozi, P. and Peeples, M. S. (2015). On the history and future of cosmic planet formation. Monthly Notices of the Royal Astronomical Society 454, 1811-1817.
- Benoit, J., et al. (2019). Brain evolution in Proboscidea (Mammalia, Afrotheria) across the Cenozoic. Scientific Reports 9, 9323.
- Bertrand, O. C., et al. (2022). Brawn before brains in placental mammals after the end-Cretaceous extinction. Science 376, 80-85.
- Bostrom, N. (2002). Anthropic Bias: Observation Selection Effects in Science and Philosophy. Routledge, New York.
- Bostrom, N. (2008). Where Are They? Why I Hope the Search for Extraterrestrial Life Finds Nothing. MIT Technology Review, May/June 2008, 72-77.
- Bramante, J., Diamond, M. D. and Kim, J. L. (2025). Dimming Starlight with Dark Compact Objects. Physical Review Letters 134, 141001.
- Brasier, M. D., et al. (2002). Questioning the evidence for Earth’s oldest fossils. Nature 416, 76-81.
- Brin, G. D. (1983). The ‘Great Silence’: the Controversy Concerning Extraterrestrial Intelligent Life. Quarterly Journal of the Royal Astronomical Society 24, 283-309.
- Burger, J. R., George, M. A., Leadbetter, C. and Shaikh, F. (2019). The allometry of brain size in mammals. Journal of Mammalogy 100, 276-283.
- Carrigan, R. A. (2009). IRAS-based Whole-Sky Upper Limit on Dyson Spheres. The Astrophysical Journal 698, 2075-2086.
- Carroll-Nellenback, J., Frank, A., Wright, J. and Scharf, C. (2019). The Fermi Paradox and the Aurora Effect: Exo-civilization Settlement, Expansion, and Steady States. The Astronomical Journal 158, 117.
- Carter, B. (1983). The anthropic principle and its implications for biological evolution. Philosophical Transactions of the Royal Society of London A 310, 347-363.
- Carter, B. (2008). Five- or six-step scenario for evolution? International Journal of Astrobiology 7, 177-182.
- Ćirković, M. M. and Vukotić, B. (2008). Astrobiological Phase Transition: Towards Resolution of Fermi’s Paradox. Origins of Life and Evolution of Biospheres 38, 535-547.
- Ćirković, M. M. (2009). Fermi’s Paradox: The Last Challenge for Copernicanism? Serbian Astronomical Journal 178, 1-20.
- Contardo, G. and Hogg, D. W. (2024). A Data-driven Search For Mid-infrared Excesses Among Five Million Main-sequence FGK Stars. The Astronomical Journal 168, 157.
- Cook, T. (2022). Replicating and extending the grabby aliens model. Effective Altruism Forum, 23 April 2022.
- Crossfield, I. J. M., et al. (2022). GJ 1252b: A Hot Terrestrial Super-Earth with No Atmosphere. The Astrophysical Journal Letters 937, L17.
- Curtis, O., et al. (2026). The Ĝ Infrared Search for Extraterrestrial Civilizations with Large Energy Supplies. V. When Galaxies Glow with Industry. arXiv:2608.12458 (preprint, submitted to ApJ).
- Davies, P. C. W. and Lineweaver, C. H. (2005). Finding a Second Sample of Life on Earth. Astrobiology 5, 154-163.
- Davies, P. C. W. (2012). Footprints of alien technology. Acta Astronautica 73, 250-257.
- Dieks, D. (1992). Doomsday - Or: The Dangers of Statistics. The Philosophical Quarterly 42, 78-84.
- Djokic, T., et al. (2017). Earliest signs of life on land preserved in ca. 3.5 Ga hot spring deposits. Nature Communications 8, 15263.
- Drake, F. D. (1961). Equation presented at the conference on extraterrestrial intelligent life, National Radio Astronomy Observatory, Green Bank, West Virginia, November 1961 (not published at the time). See Drake, F. D. (1965). The radio search for intelligent extraterrestrial life. In Current Aspects of Exobiology, eds G. Mamikunian and M. H. Briggs, 323-345. Pergamon Press.
- Du, A., et al. (2018). Pattern and process in hominin brain size evolution are scale-dependent. Proceedings of the Royal Society B: Biological Sciences 285, 20172738.
- Dyson, F. J. (1960). Search for Artificial Stellar Sources of Infrared Radiation. Science 131, 1667-1668.
- Enriquez, J. E., et al. (2017). The Breakthrough Listen Search for Intelligent Life: 1.1-1.9 GHz Observations of 692 Nearby Stars. The Astrophysical Journal 849, 104.
- Fernandez-Puig, E., et al. (2026). Stellar encounters in the solar neighbourhood and the special case of GJ 710. Astronomy & Astrophysics 710, A252.
- Finney, B. R. and Jones, E. M. (eds) (1985). Interstellar Migration and the Human Experience. University of California Press, Berkeley.
- Finnveden, L. (2019). Quantifying anthropic effects on the Fermi paradox. Effective Altruism Forum, 15 February 2019.
- Freitas, R. A. Jr. and Valdes, F. (1980). A search for natural or artificial objects located at the Earth-Moon libration points. Icarus 42, 442-447.
- Freitas, R. A. Jr. (1985). There Is No Fermi Paradox. Icarus 62, 518-520.
- Gallart, C., et al. (2024). Chronology of our Galaxy from Gaia colour-magnitude diagram fitting (ChronoGal). I. The formation and evolution of the thin disc from the Gaia Catalogue of Nearby Stars. Astronomy & Astrophysics 687, A168.
- García-Sánchez, J., et al. (2001). Stellar encounters with the solar system. Astronomy & Astrophysics 379, 634-659.
- Garrett, M. A. (2015). Application of the mid-IR radio correlation to the Ĝ sample and the search for advanced extraterrestrial civilisations. Astronomy & Astrophysics 581, L5.
- Gerig, A., Olum, K. D. and Vilenkin, A. (2013). Universal doomsday: analyzing our prospects for survival. Journal of Cosmology and Astroparticle Physics 2013(05), 013.
- Gott, J. R. (1993). Implications of the Copernican principle for our future prospects. Nature 363, 315-319.
- Grace, K. (2010). SIA doomsday: The filter is ahead. Meteuphoric (blog), 23 March 2010.
- Gray, R. H. (2015). The Fermi Paradox Is Neither Fermi’s Nor a Paradox. Astrobiology 15, 195-199.
- Griffith, R. L., et al. (2015). The Ĝ Infrared Search for Extraterrestrial Civilizations with Large Energy Supplies. III. The Reddest Extended Sources in WISE. The Astrophysical Journal Supplement Series 217, 25.
- Hansen, B. and Zuckerman, B. (2021). Minimal Conditions for Survival of Technological Civilizations in the Face of Stellar Evolution. The Astronomical Journal 161, 145.
- Hanson, R. (1998a). The Great Filter: Are We Almost Past It? Web essay, 15 September 1998.
- Hanson, R. (1998b). Must Early Life Be Easy? The Rhythm of Major Evolutionary Transitions. Working paper, 23 September 1998.
- Hanson, R., Martin, D., McCarter, C. and Paulson, J. (2021). If Loud Aliens Explain Human Earliness, Quiet Aliens Are Also Rare. The Astrophysical Journal 922, 182.
- Haqq-Misra, J. D. and Baum, S. D. (2009). The Sustainability Solution to the Fermi Paradox. Journal of the British Interplanetary Society 62, 47-51.
- Haqq-Misra, J. and Kopparapu, R. K. (2012). On the likelihood of non-terrestrial artifacts in the Solar System. Acta Astronautica 72, 15-20.
- Hart, M. H. (1975). An Explanation for the Absence of Extraterrestrials on Earth. Quarterly Journal of the Royal Astronomical Society 16, 128-135.
- Huang, B.-L., Tao, Z.-Z. and Zhang, T.-J. (2026). WISE/CatWISE Constraints on Dysonian Waste-heat Technosignatures in Nearby Galaxies. The Astronomical Journal 171, 131.
- Jones, E. M. (1985). “Where Is Everybody?” An Account of Fermi’s Question. Los Alamos National Laboratory report LA-10311-MS.
- Kay, C. J., et al. (2026). Dated gene duplications elucidate the evolutionary assembly of eukaryotes. Nature 650, 129-140.
- Kipping, D. (2020). An objective Bayesian analysis of life’s early start and our late arrival. Proceedings of the National Academy of Sciences 117, 11995-12003.
- Kipping, D. (2025a). Strong Evidence That Abiogenesis Is a Rapid Process on Earth Analogs. Astrobiology 25, 323-326.
- Kipping, D. (2025b). Solar Hegemony: M-Dwarfs Are Unlikely to Host Observers Such as Ourselves. arXiv:2510.01215. Published in 2026 as The Astronomical Journal 171, 167.
- Kipping, D. (2026). The Cosmological Hart-Tipler Conjecture. arXiv:2606.04044 (preprint, submitted to Astrobiology).
- Lacki, B. C. (2016). Type III Societies (Apparently) Do Not Exist. arXiv:1604.07844 (preprint).
- Lacki, B. C. (2019). Sunscreen: Photometric Signatures of Galaxies Partially Cloaked in Dyson Spheres. Publications of the Astronomical Society of the Pacific 131, 024102.
- Lacki, B. C. (2024). Artificial Broadcasts as Galactic Populations. II. Comparing Individualist and Collective Bounds on Broadcast Populations in Single Galaxies. The Astrophysical Journal 966, 183.
- Lacki, B. C. (2025). Artificial Broadcasts as Galactic Populations. III. Constraints on Radio Broadcasts from the Cosmic Population of Inhabited Galaxies. Publications of the Astronomical Society of the Pacific 137, 114103.
- Landis, G. A. (1998). The Fermi Paradox: An Approach Based on Percolation Theory. Journal of the British Interplanetary Society 51, 163-166.
- Leslie, J. (1996). The End of the World: The Science and Ethics of Human Extinction. Routledge, London and New York.
- Lineweaver, C. H. (2001). An Estimate of the Age Distribution of Terrestrial Planets in the Universe: Quantifying Metallicity as a Selection Effect. Icarus 151, 307-313.
- Lineweaver, C. H. (2009). Paleontological Tests: Human-Like Intelligence Is Not a Convergent Feature of Evolution. In Seckbach, J. and Walsh, M. (eds), From Fossils to Astrobiology, 353-368. Springer, Dordrecht.
- Liu, C. (2008). The Dark Forest. Chongqing Publishing House, Chongqing (in Chinese). English translation by J. Martinsen, Tor Books, New York, 2015.
- Loeb, A., Batista, R. A. and Sloan, D. (2016). Relative likelihood for life as a function of cosmic time. Journal of Cosmology and Astroparticle Physics 2016(08), 040.
- Lole Durbin, O. M., et al. (2026). Re-evaluating molecular clock maximum age calibrations revives pre-Ediacaran divergence estimates for animals. Science Advances 12, eaeg6289.
- Madau, P. and Dickinson, M. (2014). Cosmic Star-Formation History. Annual Review of Astronomy and Astrophysics 52, 415-486.
- Madau, P. (2023). Beyond the Drake Equation: A Time-dependent Inventory of Habitable Planets and Life-bearing Worlds in the Solar Neighborhood. The Astrophysical Journal 957, 66.
- Marchi, S., et al. (2014). Widespread mixing and burial of Earth’s Hadean crust by asteroid impacts. Nature 511, 578-582.
- Marino, L., McShea, D. W. and Uhen, M. D. (2004). Origin and evolution of large brains in toothed whales. The Anatomical Record Part A 281A, 1247-1255.
- Mason, L. A., Garrett, M. A. and Siemion, A. P. V. (2025). Simulating the stellar bycatch: constraining the prevalence of extraterrestrial transmitters within radio SETI surveys. Monthly Notices of the Royal Astronomical Society 545, staf2112.
- McKee, C. F., Parravano, A. and Hollenbach, D. J. (2015). Stars, Gas, and Dark Matter in the Solar Neighborhood. The Astrophysical Journal 814, 13.
- Micheli, M., et al. (2018). Non-gravitational acceleration in the trajectory of 1I/2017 U1 (‘Oumuamua). Nature 559, 223-226.
- Mills, D. B., Macalady, J. L., Frank, A. and Wright, J. T. (2025). A reassessment of the “hard-steps” model for the evolution of intelligent life. Science Advances 11, eads5698.
- Moody, E. R. R., et al. (2024). The nature of the last universal common ancestor and its impact on the early Earth system. Nature Ecology & Evolution 8, 1654-1666.
- Mortensen, H. S., et al. (2014). Quantitative relationships in delphinid neocortex. Frontiers in Neuroanatomy 8, 132.
- Nataf, D. M., Schlaufman, K. C., Reggiani, H. and Hahn, I. (2024). Accurate, Precise, and Physically Self-consistent Ages and Metallicities for 400,000 Solar Neighborhood Subgiant Branch Stars. The Astrophysical Journal 976, 87.
- Newman, W. I. and Sagan, C. (1981). Galactic civilizations: Population dynamics and interstellar diffusion. Icarus 46, 293-327.
- Olson, S. J. (2015). Homogeneous cosmology with aggressively expanding civilizations. Classical and Quantum Gravity 32, 215025.
- Olson, S. J. and Ord, T. (2021). Implications of a search for intergalactic civilizations on prior estimates of human survival and travel speed. arXiv:2106.13348 (preprint).
- Olum, K. D. (2002). The Doomsday Argument and the Number of Possible Observers. The Philosophical Quarterly 52, 164-184.
- Olum, K. D. (2004). Conflict between anthropic reasoning and observation. Analysis 64, 1-8.
- Pan, J., et al. (2026). A Bayesian framework for posterior bounds on self-replicating probes in the Solar System under observational incompleteness. Research Square preprint, version 1, 8 July 2026.
- Reinhard, C. T., Olson, S. L., Schwieterman, E. W. and Lyons, T. W. (2017). False Negatives for Remote Life Detection on Ocean-Bearing Planets: Lessons from the Early Earth. Astrobiology 17, 287-297.
- Ren, T., Garrett, M. A. and Siemion, A. P. V. (2024). Background Contamination of the Project Hephaistos Dyson Spheres Candidates. Research Notes of the AAS 8, 145.
- Ren, T., Garrett, M. A. and Siemion, A. P. V. (2025). High-resolution imaging of the radio source associated with Project Hephaistos Dyson Sphere Candidate G. Monthly Notices of the Royal Astronomical Society: Letters 538, L56-L61.
- Sandberg, A., Armstrong, S. and Ćirković, M. M. (2016). That is not dead which can eternal lie: the aestivation hypothesis for resolving Fermi’s paradox. Journal of the British Interplanetary Society 69, 406-415. Preprint: arXiv:1705.03394.
- Sandberg, A., Drexler, E. and Ord, T. (2018). Dissolving the Fermi Paradox. arXiv:1806.02404 (preprint).
- Schmidt, G. A. and Frank, A. (2019). The Silurian hypothesis: would it be possible to detect an industrial civilization in the geological record? International Journal of Astrobiology 18, 142-150 (first published online April 2018).
- Seligman, D. Z., et al. (2023). Dark Comets? Unexpectedly Large Nongravitational Accelerations on a Sample of Small Asteroids. The Planetary Science Journal 4, 35.
- Seligman, D. Z., et al. (2024). Two distinct populations of dark comets delineated by orbits and sizes. Proceedings of the National Academy of Sciences 121, e2406424121.
- Snyder-Beattie, A. E., Sandberg, A., Drexler, K. E. and Bonsall, M. B. (2021). The Timing of Evolutionary Transitions Suggests Intelligent Life is Rare. Astrobiology 21, 265-278.
- Spiegel, D. S. and Turner, E. L. (2012). Bayesian analysis of the astrobiological implications of life’s early emergence on Earth. Proceedings of the National Academy of Sciences 109, 395-400.
- Suazo, M., et al. (2022). Project Hephaistos I. Upper limits on partial Dyson spheres in the Milky Way. Monthly Notices of the Royal Astronomical Society 512, 2988-3000.
- Suazo, M., et al. (2024). Project Hephaistos II. Dyson sphere candidates from Gaia DR3, 2MASS, and WISE. Monthly Notices of the Royal Astronomical Society 531, 695-707.
- Tipler, F. J. (1980). Extraterrestrial Intelligent Beings Do Not Exist. Quarterly Journal of the Royal Astronomical Society 21, 267-281.
- Tyrrell, T. (2020). Chance played a role in determining whether Earth stayed habitable. Communications Earth & Environment 1, 61.
- Unterborn, C. T., et al. (2022). Mantle Degassing Lifetimes through Galactic Time and the Maximum Age Stagnant-lid Rocky Exoplanets Can Support Temperate Climates. The Astrophysical Journal Letters 930, L6.
- Valdes, F. and Freitas, R. A. Jr. (1983). A search for objects near the Earth-Moon Lagrangian points. Icarus 53, 453-457.
- Verendel, V. and Häggström, O. (2017). Fermi’s paradox, extraterrestrial life and the future of humanity: a Bayesian analysis. International Journal of Astrobiology 16, 14-18.
- Vermeij, G. J. (2006). Historical contingency and the purported uniqueness of evolutionary innovations. Proceedings of the National Academy of Sciences 103, 1804-1809.
- von Hoerner, S. (1961). The Search for Signals from Other Civilizations. Science 134, 1839-1843.
- Ward, P. D. and Brownlee, D. (2000). Rare Earth: Why Complex Life Is Uncommon in the Universe. Copernicus, New York.
- Watson, A. J. (2008). Implications of an Anthropic Model of Evolution for Emergence of Complex Life and Intelligence. Astrobiology 8, 175-185.
- Waugh, D. A. and Thewissen, J. G. M. (2021). The pattern of brain-size change in the early evolution of cetaceans. PLOS ONE 16, e0257803.
- Webb, S. (2015). If the Universe Is Teeming with Aliens ... WHERE IS EVERYBODY? Seventy-Five Solutions to the Fermi Paradox and the Problem of Extraterrestrial Life. Second edition. Springer, Cham.
- Whitmire, D. P. and Wright, D. P. (1980). Nuclear waste spectrum as evidence of technological extraterrestrial civilizations. Icarus 42, 149-156.
- Wilde, S. A., Valley, J. W., Peck, W. H. and Graham, C. M. (2001). Evidence from detrital zircons for the existence of continental crust and oceans on the Earth 4.4 Gyr ago. Nature 409, 175-178.
- Wlodarczyk-Sroka, B. S., Garrett, M. A. and Siemion, A. P. V. (2020). Extending the Breakthrough Listen nearby star survey to other stellar objects in the field. Monthly Notices of the Royal Astronomical Society 498, 5720-5729.
- Wright, J. T., Kanodia, S. and Lubar, E. G. (2018). How Much SETI Has Been Done? Finding Needles in the n-dimensional Cosmic Haystack. The Astronomical Journal 156, 260.
- Xiang, M. and Rix, H.-W. (2022). A time-resolved picture of our Milky Way’s early formation history. Nature 603, 599-603.
- Zackrisson, E., Calissendorff, P., Asadi, S. and Nyholm, A. (2015). Extragalactic SETI: The Tully-Fisher Relation as a Probe of Dysonian Astroengineering in Disk Galaxies. The Astrophysical Journal 810, 23.
- Zackrisson, E., Korn, A. J., Wehrhahn, A. and Reiter, J. (2018). SETI with Gaia: The Observational Signatures of Nearly Complete Dyson Spheres. The Astrophysical Journal 862, 21.
- Zackrisson, E., et al. (2026). Project Hephaistos IV. James Webb Space Telescope Observations of Two Dyson Sphere Candidates. arXiv:2607.09460 (preprint).
- Zuckerman, B. (2026). Broadband Searches for Extraterrestrial Technological Intelligence: A New Strategy to Find Nearby Alien Civilizations. The Astrophysical Journal 1001, 167.