Is intelligent life rare in the universe, or did it arise on Earth because planets like ours reliably produce it given enough time? Nobody knows, and the question is harder than it sounds, because we have exactly one example and we are that example. In 1983 the physicist Brandon Carter turned this awkwardness into an argument. If the steps that led from chemistry to minds were each very unlikely, he reasoned, then the timeline of life on Earth should carry a recognizable signature, and he thought he could see it. That argument, now called the hard-steps model, still anchors much of the scientific case that we may be alone, and it has recently come under its sharpest challenge yet.
Interest is not confined to journals. Popular coverage has run pieces such as Smithsonian’s report that intelligent life may be more likely elsewhere than previously thought, and Universe Today asking whether intelligent life is inevitable. On forums concerned with long-run risk, the idea sits inside the “Great Filter” discussion, where the worry is whether the improbable step lies behind us or ahead.
The question, stated precisely
Loose versions of the question mix up several different claims. A precise version runs like this: on a planet that stays habitable for a finite window, is the expected waiting time for intelligent, technology-capable life much longer than that window, or comparable to it, or much shorter?
If the expected wait is far longer than the window, almost every habitable planet runs out of time first, and intelligence is a fluke in the literal sense: a lucky draw from a distribution whose typical outcome is failure. Should the wait instead be comparable to or shorter than the window, intelligence is something like a regular product of planetary evolution, delayed mainly by how quickly environments change. Neither claim settles whether simple life is rare; microbes could be everywhere while minds are not.
Why it matters is fairly direct. Estimates of how many civilizations might exist hinge on one factor that nobody can measure from the outside: the fraction of inhabited worlds where intelligence appears. The hard-steps model is one of the few attempts to infer that fraction from data we do have, namely the dates in Earth’s own history.

What is established
Some things here are not in dispute. Earth is about 4.6 billion years old, and life shows up early: the oldest purported evidence of metabolism is about 3.7 billion years old, as summarized in the 2025 reassessment discussed below. The Sun brightens over time, and models of the Earth system predict that rising temperatures and falling carbon dioxide will end the biosphere, or at least land plant ecosystems, somewhere around 1.0 to 1.9 billion years from now; that range is the one compiled by Mills and colleagues (2025) from earlier modeling work. Carter’s original paper had used a much longer remaining window, based on the Sun’s main-sequence lifetime.
It is also clear that some major innovations appear only once, or seem to. Eukaryotic cells, with their nucleus and mitochondria, share a single common ancestor. Human-style technological intelligence has appeared in one lineage. Whether “appeared once” means “was improbable” is exactly the point at dispute.
A third established point is mathematical rather than biological. Given the standard assumptions of the model (each step happens at a constant small rate, steps happen in sequence, and success means finishing all of them before the window closes), the statistics of successful runs follow from probability theory. Robin Hanson’s 1998 working paper and later Snyder-Beattie, Sandberg, Drexler and Bonsall (2021) worked these out in detail, and the simulation below reproduces the core result.
Carter’s argument in brief
Carter’s observation, in his 1983 paper in the Philosophical Transactions of the Royal Society A, was a coincidence of timescales. The time it took intelligence to appear on Earth is of the same order as the total time the Sun allows. Nothing obvious ties the biological timescale to the stellar one.
His explanation used an observer-selection effect. Suppose the expected time to evolve intelligence were far shorter than the Sun’s lifetime. Then intelligence would typically appear early, and we would expect to find ourselves near the start of the window. Suppose instead the expected time were far longer. Most planets would fail, but on the rare planet that succeeds, observers would find that success arrived late, near the end of the available time, because a lucky fast run still has to squeeze several improbable events into a fixed interval. We exist, so we are necessarily on a successful planet. A late arrival is what the “very hard” case predicts.
Carter then used a model with a small number of critical steps to argue that their number was low, and that most of the developments in between were not themselves bottlenecks. The quantitative core is simple: with n hard steps in a window of length L, the expected time left after the final step is about L/(n+1). Snyder-Beattie and colleagues restate this in their review of the Carter argument as the claim that the time remaining “will be roughly equal to τ⊙/(n + 1)”.
Hypothesis one: a few genuinely hard steps
The strongest version of the rare-intelligence view comes from Snyder-Beattie and colleagues. Rather than counting steps, they fixed a candidate sequence (origin of life, eukaryotes, sexual reproduction, intelligence) and asked which step rates are consistent with Earth’s dates once the selection effect is built in. Their Bayesian analysis renormalizes the likelihood by the probability of finishing within the habitable window, which is the formal version of Carter’s point.
The result is lopsided. The posterior peaks near rates matching the fossil dates, but it does not fall away for very slow rates; it flattens to a nonzero plateau. Translated, Earth’s record cannot rule out steps that are astronomically unlikely, because a successful run looks much the same either way. In their model with three or four steps, about 99 percent of the posterior weight sits on parameters giving less than a 1 percent chance of finishing in time. Their summary:
“expected evolutionary transition times likely exceed the lifetime of Earth, perhaps by many orders of magnitude”
Snyder-Beattie et al., 2021, Astrobiology
The steelman is genuine: any inference about how common we are, drawn from the fact that we exist, must correct for selection. The early origin of life, often taken as evidence that life is easy, is weaker evidence than it looks: Hanson showed that, conditional on success, the durations of truly hard steps tend to look alike regardless of how hard each one is. A fast first step is consistent with an easy origin and also with a hard one.
Against this view, the plateau cuts both ways. If the data cannot distinguish rates of one in ten billion years from one in a trillion, they also offer limited positive support for rarity; the conclusion leans heavily on the prior, which the authors chose to be log-uniform. The candidate steps are also picked by the analysts, and the window length is an input that has shifted as Earth-system models improved.
Hypothesis two: no hard steps, only opening windows
The challenge published in Science Advances by Daniel Mills, Jennifer Macalady, Adam Frank and Jason Wright (2025; preprint on arXiv) attacks the model at its assumptions rather than its algebra. Their headline is blunt:
“we advance a potential alternative model where there are no hard steps”
Mills, Macalady, Frank and Wright, 2025 (arXiv abstract)
Their argument has three strands. First, the remaining-time input: with the biosphere ending perhaps 1.0 to 1.9 billion years from now rather than about 5 billion, humans arrive late in the window but not in the striking way Carter described. Second, uniqueness: an innovation may look like a one-off because competing lineages went extinct and left little trace, because an early innovator crowded out later ones, or because the innovation changed the environment so that it could not recur. Oxygenic photosynthesis is their central case of the last.
The third strand carries the most weight. Each step had environmental preconditions, chiefly oxygen. Eukaryotes and animals were probably excluded before the Great Oxidation Event, and an atmosphere that a human could breathe long-term was reliably present only after roughly 420 to 400 million years ago. On this reading, the long pauses in Earth’s history were waits for the planet, not waits for luck, and humans arrived “on time.” They also point out that their five candidate steps are rather evenly spaced, with an average gap of about 0.84 billion years and a longest gap only two to three times the shortest.
The weakness of this hypothesis is the mirror image of its rival’s. It explains delays by environmental change, but oxygenation itself was a biological product, so the timing of photosynthesis and its spread still needs explaining. Calling that predictable is plausible but unmeasured.

A middle position and the timeline problem
These camps are not exhaustive. A mixed picture, in which one or two steps are genuinely improbable while the rest are paced by geochemistry, fits much of the evidence on both sides. Hanson’s analysis already allowed for this: easy steps take roughly their own expected time, while hard steps stretch to fill the window in a characteristic way.
Everything then depends on the dates, and the dates are soft. The origin of eukaryotes is not pinned to a narrow interval; Mills and colleagues place it somewhere in the Proterozoic, a span of roughly two billion years. Early life’s start depends on when the surface became habitable, which they put between about 4.5 and 3.9 billion years ago. Any test that relies on spacing inherits all of this uncertainty, which is worth keeping in mind for the next section.
A small simulation built for this article
The model makes concrete predictions that can be computed directly. The script below treats each step as an exponential waiting time and conditions on finishing all of them inside a window normalized to length 1. For equal rates there is an exact sampler: draw the total time from a gamma distribution truncated at the window, then place the earlier completions as sorted uniform points below it. Part B mixes hard and easy steps by rejection sampling; part C compares Earth with the model. The seed is fixed; the run takes about seven seconds.
import numpy as np
from scipy import stats
rng = np.random.default_rng(23)
L = 1.0 # habitable window, normalized
def conditional_equal(n, mu, N):
"""Exact draw: n exponential steps (mean mu) conditioned on finishing by L."""
g = stats.gamma(a=n, scale=mu)
S = g.ppf(rng.uniform(0, g.cdf(L), N)) # truncated total time
u = np.sort(rng.uniform(0, 1, (N, n - 1)), axis=1) # partial sums given S
t = np.hstack([u, np.ones((N, 1))]) * S[:, None]
gaps = np.diff(np.hstack([np.zeros((N, 1)), t]), axis=1)
return gaps, L - S, g.cdf(L)
print("A: mean gap and mean time left (fractions of window)")
for n in (1, 2, 3, 5):
for mu in (0.3, 1, 10, 1000):
gaps, rem, p = conditional_equal(n, mu, 200_000)
print(f"n={n} mu={mu:>6}: P(success)={p:.2e} gap={gaps.mean():.3f} "
f"left={rem.mean():.3f} 1/(n+1)={1/(n+1):.3f}")
print("B: 3 hard (mu=10) + 2 easy (mu=0.02) steps, rejection sampling")
mus = np.array([10, 0.02, 10, 0.02, 10]); keep = []
for _ in range(40):
x = rng.exponential(mus, (1_000_000, 5))
keep.append(x[x.sum(1) < L])
x = np.vstack(keep)
print("accepted", len(x), "mean durations", x.mean(0).round(3),
"left", (L - x.sum(1)).mean().round(3))
print("C: Earth comparison")
for R, W in ((1.0, 5.6), (1.9, 6.5)): # Gyr left, window from formation
r = R / W
print(f"left={R} window={W}: fraction={r:.3f} ML n={-1/np.log(1-r):.1f}")
gaps, rem, _ = conditional_equal(5, 1000, 500_000)
for k, label in ((gaps, "5 inter-step gaps"),
(np.hstack([gaps, rem[:, None]]), "5 gaps + time left")):
ratio = k.max(1) / k.min(1)
print(label, "median max/min =", np.median(ratio).round(1),
"P(max/min<=3) =", (ratio <= 3).mean().round(4))
Selected results from that run (μ is the mean waiting time per step, as a multiple of the window):
| Steps n | μ / window | P(all finish) | Mean gap | Mean time left | 1/(n+1) |
|---|---|---|---|---|---|
| 1 | 0.3 | 0.964 | 0.262 | 0.738 | 0.500 |
| 1 | 1000 | 0.001 | 0.500 | 0.500 | 0.500 |
| 3 | 1 | 0.080 | 0.237 | 0.290 | 0.250 |
| 3 | 10 | 1.55×10-4 | 0.249 | 0.253 | 0.250 |
| 5 | 1 | 3.66×10-3 | 0.162 | 0.188 | 0.167 |
| 5 | 10 | 7.67×10-8 | 0.166 | 0.169 | 0.167 |
| 5 | 1000 | 8.33×10-18 | 0.167 | 0.167 | 0.167 |
Three things come out. Once steps are hard (μ well above the window), successful runs look the same no matter how hard: the gaps and the time left after the last step each converge on 1/(n+1) of the window, while the chance of success itself drops by many orders of magnitude. That is Carter’s spacing law and Hanson’s “rhythm,” reproduced. In the mixed run, 5,453 of 40 million histories succeeded; the two easy steps took about 0.019 of the window each, close to their own mean of 0.02, while the three hard steps took 0.239, 0.238 and 0.240, with 0.246 left over. Easy steps reveal their difficulty, whereas hard ones all look alike.
For the comparison with Earth, take Mills and colleagues’ numbers at face value: about 4.6 billion years elapsed and 1.0 to 1.9 billion left. Treating the time left as a single draw from the model’s distribution, the maximum-likelihood number of hard steps is about 5.1 if 1.0 billion years remain and about 2.9 if 1.9 billion remain. Both fall in the range usually proposed, so this test alone does not falsify the model.
Spacing is a stiffer test. With five hard steps, uniform-spacing statistics make some gaps short and others long. In 500,000 simulated successful histories, the median ratio of longest to shortest inter-step gap was 14.4, and only 4.2 percent had a ratio of 3 or less. Including the time left as a sixth interval, the median ratio rose to 18.9 and only 1.7 percent were that even. Earth’s five steps, on Mills and colleagues’ accounting, sit at a ratio of about two to three.
What this does and does not show needs stating carefully. It shows that, if those five events were all hard steps and the dates are right, Earth’s history is more regular than a pure hard-steps model usually produces; something like a p-value of a few percent. It does not show that the steps were easy. Regular spacing also arises if a few hard steps mix with environmental delays, or if analysts unconsciously pick evenly spread milestones. The test is also sensitive to where the window starts and which events are counted.
Where the evidence points, and what would settle it
What follows is opinion, not established result. The hard-steps model is a sound piece of probability attached to fragile inputs. Its strongest lesson is negative: Earth’s early origin of life and late arrival of intelligence do not, by themselves, show that either step was easy, because selection hides the difference. The case that intelligence is very rare rests heavily on priors and on window estimates that have moved. The environmental-windows view seems the better description of the last two billion years, but it has not explained away the possibility that one or two early transitions were truly improbable. A mixed model is the best-supported guess.
Several observations or studies could move the question, and some are falsifiable in a useful way:
- Independent origins. A second, unrelated origin of life on Mars, Europa or Enceladus would make abiogenesis hard to treat as a hard step. Finding no life at all on several clearly habitable worlds would push the other way.
- Biosignatures by stellar age. If oxygen-rich atmospheres turn up on exoplanets, the windows hypothesis predicts they should be common on old, long-habitable worlds and track planetary age; a hard-steps view predicts they would be scattered rarely and without a clear age trend.
- Tighter dates. A narrower date for the last eukaryotic common ancestor, and for the onset of oxygenic photosynthesis, would make the spacing test above meaningful instead of illustrative. If refined gaps grow uneven, the hard-steps reading gains.
- Earth-system modeling. A firm estimate of the biosphere’s remaining lifetime directly moves the estimated number of hard steps, as the simulation shows.
- Pre-registered step lists. Agreeing on candidate transitions before testing their spacing would remove the worry that milestones are chosen to fit.
Related reading on this site: the leading theories of how life began covers the first candidate step, and whether evolution is random explains why chance and selection are not opposites, which matters for any claim that a transition was “lucky.” For evidence that the process continues in our own lineage, see whether humans are still evolving.

Common questions
What is a “hard step” in evolution?
It is a transition whose expected waiting time is much longer than the time a planet stays habitable. On most planets it would never happen. The term comes from Carter’s model and the later work of Hanson and others.
Does life appearing early on Earth prove life is common?
Not by itself. If the origin of life were a hard step, successful planets would still tend to show it occupying about the same share of the window as any other hard step. An early start is somewhat more likely if life is easy, but the evidence is weaker than intuition suggests.
Does the hard-steps model say we are alone?
No. It says that if intelligence required several hard steps, it would be rare per habitable planet. With enough planets, rare could still mean many civilizations across the universe, just far apart.
Why do recent critics say humans evolved “on time”?
Mills and colleagues argue that each major step had to wait for Earth’s environment, especially oxygen levels, to allow it. On that view the long delays reflect a planet slowly becoming suitable, not a string of lucky accidents.
Could this question ever be answered?
Partly. Finding independent life elsewhere, or surveying many exoplanet atmospheres, would test the early steps. The final step, intelligence, is much harder to survey.
Further reading
- Rare Earth: Why Complex Life Is Uncommon in the Universe by Peter D. Ward and Donald Brownlee. The best-known popular case that complex life is rare, the view the hard-steps model formalizes. See it on Amazon (affiliate link)
- The Vital Question: Energy, Evolution, and the Origins of Complex Life by Nick Lane. A readable account of why the origin of eukaryotic cells may have been an improbable bottleneck. See it on Amazon (affiliate link)
As an Amazon Associate I earn from qualifying purchases. A commission may be earned if you buy through this link, at no extra cost to you.
References
- Carter, B. (1983). The anthropic principle and its implications for biological evolution. Philosophical Transactions of the Royal Society of London A, 310, 347–363. PDF
- Hanson, R. (1998). Must early life be easy? The rhythm of major evolutionary transitions. Working paper, George Mason University. PDF
- Snyder-Beattie, A. E., Sandberg, A., Drexler, K. E., & Bonsall, M. B. (2021). The timing of evolutionary transitions suggests intelligent life is rare. Astrobiology, 21(3), 265–278. doi:10.1089/ast.2019.2149. PMC full text
- Mills, D. B., Macalady, J. L., Frank, A., & Wright, J. T. (2025). A reassessment of the “hard-steps” model for the evolution of intelligent life. Science Advances. doi:10.1126/sciadv.ads5698; preprint arXiv:2408.10293
- Smithsonian Magazine. Intelligent life may be more likely to exist on other planets than previously thought, scientists say. Article
- Universe Today. Is intelligent life inevitable? Article
- LessWrong. Great Filter (wiki entry). Page
