What Are the True Values of the Terms in the Drake Equation?

On a November morning in 1961, eleven scientists gathered at the Green Bank observatory in West Virginia to talk about the possibility that someone, somewhere, was listening. Frank Drake, a young radio astronomer who had just spent months pointing a telescope at two nearby stars in search of an artificial signal, needed an agenda for the meeting. So he wrote a single line of algebra on a blackboard — seven terms multiplied together, promising to spit out N, the number of communicating civilizations in the Milky Way right now. It was less an equation than a to-do list disguised as arithmetic. Sixty-plus years later, that blackboard sketch is still the organizing scaffold for how we think about our cosmic company, and most of its terms remain almost exactly as uncertain as they were the day Drake chalked them up.

This is a strange kind of scientific object. It's not a law, not a derived result, not even really a testable hypothesis in its current form — it's a factorization of ignorance, a way of being precise about what we don't know. And that precision is oddly comforting: it turns a vague, overwhelming mystery ("are we alone?") into seven smaller mysteries, each of which looks, in principle, solvable. The trouble is that "in principle" has been doing an enormous amount of work for six decades. Some terms have tightened by orders of magnitude in just the last twenty years. Others are as wide open as they were when Sputnik was still a novelty, held hostage by the fact that we have exactly one confirmed data point for the origin and evolution of life, and that data point is us.

svg[The known and unknown terms of the Drake equation, sketched as a fading gradient from measured to guessed]

What this diagram is meant to convey honestly: the left half of the equation has become genuine observational science, built on statistics over hundreds of thousands of stars. The right half remains, in a very literal sense, a story built from a sample size of one planet.

Here is the uncomfortable core of the problem, and it is worth stating without euphemism: every term from f_l onward is constrained by exactly one data point, Earth, and Earth is not a random sample — it's the only sample we have, drawn under anthropic-fine-tuning conditions we can't fully disentangle from selection effects. We know life arose here. We do not know whether that event was a near-inevitability of chemistry given enough time and the right ingredients, or an almost inconceivably rare fluke that happened to occur on the one planet capable of noticing it happened.

f_l, the fraction of habitable planets on which life actually arises. abiogenesis — the transition from prebiotic chemistry to the first self-replicating, evolving systems — remains one of the deepest open problems in all of science, and it sits squarely inside this term. We have plausible chemical pathways: the Miller-Urey experiments of the 1950s showed that amino acids form readily from simple gases and energy under early-Earth-like conditions; the "RNA world" hypothesis proposes that self-replicating RNA molecules, capable of both storing information and catalyzing reactions, could have bootstrapped the whole process before DNA and proteins divided the labor; hydrothermal vent chemistry offers plausible energy gradients and mineral catalysts. We have geochemical evidence, contested but suggestive, that life on Earth may have appeared astonishingly fast — within a few hundred million years of the planet's surface becoming habitable at all, some readings of the isotopic and microfossil record push it earlier still. Fast emergence is often cited as evidence that abiogenesis is easy: if it happened almost as soon as it could have, maybe it happens whenever conditions allow.

But this argument has a well-known flaw, one that mixes uncomfortably with bayesian-inference: we are observers who can only exist on a planet where life did, in fact, arise, and arose early enough to leave time for intelligence to evolve before the sun's habitable window closes. This is an observation selection effect. Even if abiogenesis were spectacularly rare — say, expected to take on average a trillion years per habitable planet — we would still, on any planet where it eventually happened to occur quickly by sheer chance, find ourselves marveling at how fast life appeared, because slow cases never produce anyone to marvel at all. The philosopher Nick Bostrom and others have shown, with reasonably careful probability arguments, that the observed rapidity of life's origin on Earth is close to uninformative about the true rate of abiogenesis across the universe, precisely because of this selection bias. It's a sobering result: a piece of evidence that looks, intuitively, like strong support for "life is easy" turns out, on careful Bayesian analysis, to tell us almost nothing. f_l could plausibly range from something close to 1 — life springs up whenever chemistry allows — down to something so small that Earth is the only instance in the observable universe. No experiment, observation, or argument currently rules out either extreme.

f_i, the fraction of life-bearing worlds that evolve intelligence. If f_l is uncertain by unknown orders of magnitude, f_i may be worse, because here we have a genuinely troubling piece of evidence from Earth's own history: life has existed on this planet for roughly 3.5 to 4 billion years, but technological, tool-building, language-using, radio-transmitting intelligence has existed for perhaps a few tens of thousands of years — an eyeblink. Multicellularity itself appears to have evolved independently several times, complex nervous systems likewise, but something like human-level general intelligence — the capacity for abstract reasoning, cumulative culture, and technology — appears, on Earth's own record, to be a spectacular outlier even among the outliers. Convergent evolution gives some grounds for optimism: eyes have evolved independently dozens of times, as has flight, as has some degree of problem-solving cognition in crows, octopuses, elephants, and great apes. But nothing else on Earth has built a radio telescope, and it's entirely unclear whether that reflects something close to inevitable given enough evolutionary time, or an accident of a particular evolutionary lineage — bipedalism freeing the hands, a particular climate pressure, a chance encephalization event — that need not repeat.

f_c, the fraction of intelligent species that develop detectable technology. This term folds in an additional, thornier uncertainty: even granting intelligence, will it choose, or stumble into, technologies that broadcast? Dolphins and octopuses may be highly intelligent without hands or fire; a fully aquatic intelligent species might never develop metallurgy or radio at all, regardless of how clever it is. This term quietly smuggles in assumptions about what "detectable" even means — a question that belongs as much to biosignatures-detection as to Drake's original 1961 framing, which was concerned specifically with radio.

L, the lifetime of a detectable civilization. This is the term Drake himself considered the most important and the most uncertain, and it remains exactly that. It measures, in years, how long a civilization continues broadcasting after it becomes capable of doing so. Our own data point is unsettling: we have been leaking strong radio signals, and briefly), deliberately transmitting a few, for less than a century, and the trend in our transmission technology is toward less detectability — cable and fiber-optic replacing broadcast radio, efficient tight-beam and encrypted digital signals replacing the leaky, omnidirectional analog broadcasts of mid-twentieth-century television and radar. It's entirely possible that L is short not because civilizations destroy themselves, but simply because the "loud radio" phase of any technological civilization is a brief technological adolescence, quickly succeeded by quieter, more efficient communication that doesn't spill into space. Separately, there is the darker reading of L: that it measures survival against self-inflicted catastrophe — nuclear war, engineered pandemics, uncontrolled artificial intelligence, runaway climate disruption — the kinds of existential risks explored by researchers like Toby Ord. Estimates for L that different researchers have plugged into the equation over the decades span from decades to tens of millions of years, a range of six or more orders of magnitude, and there is no data that discriminates among them, because we have never observed the death — or the enduring survival — of a single other technological civilization.

Given that four or five of the seven terms are essentially unconstrained, what do working scientists actually do? Broadly, the field has split into a few distinct research programs, each trying to make progress on the unmeasurable terms by different routes.

The astrobiology-empirical program treats f_l, f_i, and f_c as ultimately answerable by better data, just not yet. Its adherents point to the coming generation of biosignature science: telescopes like JWST and, eventually, dedicated mission concepts such as the Habitable Worlds Observatory, which aim to take spectra of the atmospheres of rocky exoplanets and look for gases — oxygen alongside methane, in disequilibrium combinations hard to explain without biology — that would constitute indirect evidence of life. If such biosignatures are found on even one or two nearby exoplanets, f_l stops being a philosophical argument about one data point and becomes an actual, if still small, statistical sample. This program's appeal is that it's genuinely tractable with foreseeable technology; its trouble is that "genuinely tractable" still means years to decades of investment, and even a positive detection would only firmly establish microbial life, saying nothing directly about f_i or f_c.

The astrochemistry / origins-of-life program tries to constrain f_l from the other direction: not by finding alien life, but by understanding, mechanistically, how likely abiogenesis chemistry is under realistic early-planet conditions, in the lab, on Earth. If researchers can show that RNA-world-like self-replicators, or some plausible alternative chemistry, emerge robustly and repeatedly under a wide range of plausible prebiotic conditions, that would push f_l toward "high" independent of any Bayesian selection-effect worries about Earth's own history. Conversely, if decades of increasingly sophisticated experiments continue to show that every step from simple chemistry to the first replicator requires exquisitely specific, low-probability coincidences, that pushes f_l down. This program's appeal is that it's real, falsifiable laboratory science; its trouble is that a laboratory result about possible chemical pathways doesn't tell you the actual rate at which those pathways were realized across billions of planets over billions of years — a gap between "possible" and "probable" that chemistry alone can't close.

The Bayesian / statistical-modeling program, associated with researchers like Anders Sandberg, Eric Drexler, and Toby Ord, doesn't try to pin down each term individually. Instead it treats the whole equation probabilistically: rather than multiplying seven best-guess point values, it assigns each term a probability distribution — reflecting the genuine, honest spread of expert opinion in the literature, often spanning many orders of magnitude — and then propagates those distributions through the equation using Monte Carlo methods. The striking result of this approach, published by Sandberg, Drexler, and Ord in 2018, is that even without any new data, simply being honest about how uncertain each term is (rather than plugging in single "best guess" numbers, which silently discard the width of that uncertainty) dramatically changes the picture. Because the equation is a product of terms each spanning orders of magnitude, and because our uncertainty is better described as roughly log-uniform (uniform across factors of ten) than as uniform on a linear scale, propagating that uncertainty honestly reveals a wide range that includes both "we are likely alone in the galaxy, and possibly in the observable universe" and "the galaxy should have many communicating civilizations" as live possibilities — with a substantial fraction of the probability mass actually favoring solitude, once you take the uncertainty in the biological terms fully seriously, rather than the moderate, comfortable middle-ground numbers the equation is often used to produce. This is arguably the single most important recent contribution to Drake-equation scholarship: not a new number, but a rigorous demonstration that our usual way of using the equation — plugging in single "reasonable" values — was hiding just how uninformative our current knowledge really is.

The "great filter" reframing, developed by economist Robin Hanson and others, doesn't try to solve for individual Drake terms so much as ask a different, related question: given that the sky appears empty, somewhere along the chain from simple chemistry to a galaxy-spanning civilization, there must be at least one step — one term, or one sub-step within a term — that is extraordinarily improbable, a "Great Filter." The unresolved and genuinely important question is whether that filter lies behind us (abiogenesis itself, or the leap to complex eukaryotic life, or the leap to intelligence, all of which would be reassuring, since it would mean we've already cleared the hardest part) or ahead of us (something like the typical self-destruction of technological civilizations, which would be a profoundly worrying implication buried inside the value of L). This program connects the Drake equation most directly to the fermi-paradox: it reframes "what are the term values" as "where, in this seven-step chain, does nature apply the brakes?"

Each of these programs has genuine appeal, and genuine trouble. The empirical astrobiology program is the most scientifically satisfying because it promises actual measurements, but its timeline is long and its reach — even in the best case — may only ever touch the easiest terms (f_l, maybe) while leaving f_i, f_c, and L essentially untouched by any observation we can currently imagine making. The chemistry program does rigorous, checkable science but can't bridge the gap from "possible" to "the actual historical frequency across the galaxy." The Bayesian program is intellectually honest almost to a fault — its chief output is a rigorous demonstration of how little we know, which is valuable but can feel like progress that consists entirely of admitting non-progress. And the Great Filter framing is a powerful way to organize the question, but it is not itself a method for measuring anything; it's a lens, not a instrument.

There is no single experiment that resolves the Drake equation, because it isn't a single question. But there are concrete, foreseeable steps that would move specific terms from "guessed" to "measured," and it's worth being precise about what those look like, because this is where the honest optimism lives.

For n_e, ongoing and near-future transit and radial-velocity surveys, combined with better statistical de-biasing techniques, should tighten η⊕ to within a factor of two or so within the next decade — a real, incremental improvement, not a revolution, but genuine progress on a term that used to be pure guesswork.

For f_l, there are two independent and complementary paths. One is atmospheric biosignature detection: JWST has already begun probing the atmospheres of rocky exoplanets, though detecting a robust biosignature on an Earth-sized, temperate rocky world is at the very edge of its capability and will more plausibly fall to the next generation of large space telescopes, planned for the 2040s, purpose-built to directly image and take spectra of Earth-like planets around Sun-like stars. A single unambiguous positive detection — say, an atmosphere with oxygen and methane coexisting in a ratio that has no plausible abiotic explanation — would not prove f_l is "high" in any statistically rigorous sense from one example, but it would transform f_l from a term inferred from a sample of one to a term inferred from a sample of two, which is an infinitely larger relative improvement than it sounds. The second path is closer to home: missions to Europa (Europa Clipper, already en route) and concepts for landing on or drilling into Enceladus's ice could, in principle, detect microbial life or its biochemical signatures within our own solar system. A second independent genesis of life within one planetary system — Earth's — would be enormously informative, because it would suggest life arises readily whenever liquid water and chemical energy coexist, rather than requiring the specific, possibly rare, surface conditions of early Earth.

For f_i and f_c, there is, honestly, no foreseeable direct observational path. We cannot currently detect the presence of unintelligent life at interstellar distance except through crude atmospheric proxies, let alone detect the presence or absence of intelligence. Any progress on these terms in the coming decades will likely be indirect: better understanding of the evolutionary contingencies that produced human cognition, better models of convergent evolution, and continued (so far entirely negative) results from SETI radio and optical searches, which constrain f_c · L jointly — not separately — by ruling out certain kinds of loud, long-lived, nearby transmitting civilizations without telling us whether the silence reflects rarity or brevity or both.

For L, the most likely route to any constraint is sociological and historical rather than astronomical: the more we learn about the actual base rates of civilizational collapse, technological transitions, and existential risk from our own single case — and from careful theoretical work on the dynamics of self-replicating technological systems — the better we can bound plausible ranges for L, even though we will likely never observe a second civilization's L directly. This is progress of a philosophical and statistical character, not an instrument-and-telescope character, and it is unlikely ever to converge the way R* has converged.

So the honest forecast is: within the next twenty to thirty years, we can reasonably expect f_p (already essentially done), n_e, and possibly f_l to move from "wide uncertainty" to "reasonably well constrained." The remaining terms — f_i, f_c, and L — may remain fundamentally unconstrained by direct evidence for the foreseeable future, perhaps indefinitely, unless we actually detect a second technological civilization, at which point the entire character of the problem changes from statistical inference to direct observation.

It's tempting to treat this as an charming but inconsequential puzzle — stargazing dressed up in algebra. It isn't. The values of these terms bear on questions that reach into how we understand our own existence and plan our own future.

If f_l and f_i turn out to be high — if life and intelligence are cosmic commonplaces — then the silence of the sky becomes deeply troubling, because it would mean the filter lies ahead of us, in L: something tends to end technological civilizations, or silence them, reliably and early, and we would be wise to ask urgently what that something is. If f_l or f_i turn out to be catastrophically low, then Earth's biosphere, and human intelligence in particular, is not a routine outcome of cosmic chemistry but an almost unrepeatable accident — a fact that would reframe humanity's moral weight in a universe otherwise sterile, and would make the long-term survival of Earth-originating life and mind a stake of nearly unbounded value, since its extinction might mean the permanent extinction of the only such experiment for a very long time or ever.

There are practical stakes too. Space agencies allocate real budgets — hundreds of millions of dollars in mission concepts — based partly on where in this chain of terms they believe the highest-value uncertainty lies; the argument for prioritizing Europa Clipper, for a Habitable Worlds Observatory, for continued SETI funding despite forty years without a confirmed detection, is fundamentally an argument about which Drake terms are both important and newly tractable. And there is a philosophical stake that connects directly to anthropic-fine-tuning: the Drake equation forces us to confront, rigorously, how much of what we observe about our own existence is generic cosmic fact and how much is an artifact of the observer-selection effects built into the very act of asking the question — a discipline of thought that has become essential far beyond astrobiology, in cosmology, in the interpretation of physical constants, wherever "why do we find ourselves in conditions suitable for us" threatens to masquerade as an explanation rather than a tautology.

history

• 1953 — Miller-Urey experiment shows amino acids form readily from simple prebiotic chemistry.
• 1961 — Frank Drake formulates the equation for the Green Bank SETI meeting.
• 1995 — First exoplanet around a Sun-like star, 51 Pegasi b, detected by Mayor and Queloz.
• 2009–2018 — Kepler mission delivers the first statistical census of planets, largely fixing f_p and constraining n_e.

What makes the Drake equation endure isn't that it answers a question — it's that it survives being wrong about the answer, decade after decade, while still organizing the search usefully. Every term that has moved from guess to measurement (R*, f_p, and increasingly n_e) has done so through patient, unglamorous observational astronomy: counting stars, timing transits, correcting for selection effects in survey data. There's no reason to think the remaining terms are permanently unmeasurable in the same way — only that they are harder, because they ask about biology and history and minds rather than points of light, and because we have exactly one biosphere and one technological civilization to learn from.

The honest state of the field is this: we have converted two Drake terms from speculation to measurement, we are in the process of converting a third, and we do not currently know how — or whether — the remaining three or four will ever leave the realm of philosophical argument and enter the realm of data. That is not a comfortable place to leave a scientific question, but it is where this one actually sits, and pretending otherwise — quoting a confident value of N, as popular accounts sometimes do — misrepresents just how early we still are.

• Is the rapid emergence of life on Earth genuine evidence that abiogenesis is common, or is that inference entirely destroyed by observer-selection bias, as Bostrom's argument suggests?
• Would a second, independent origin of life within our own solar system — on Europa or Enceladus — tell us that life is common, or merely that panspermia moved it from one body to another?
• Is human-level intelligence a near-inevitable outcome of complex nervous systems given enough time, or a wildly improbable evolutionary accident specific to our lineage?
• Does the trend toward quieter, more efficient communication technology mean L is short simply because loud radio eras are brief — not because civilizations self-destruct?
• If we ever do detect a single confirmed biosignature or technosignature, how much should that one new data point be allowed to move our estimate of every other unmeasured term?

Ask the tutor: If we detected one unambiguous microbial biosignature on a single exoplanet tomorrow, how much should that really move our estimate of f_l — and how would you defend that number against the same observer-selection objection Bostrom raised about Earth itself?