Epistemic transparency: Confidence in conclusions varies throughout. I give rough indicators of my confidence at the section level by indicating the amount of time I spent researching/thinking about each particular subtopic, plus a qualitative description of the types of sources I rely on. In general, I consider it a first step toward understanding this threat from civilizational collapse — not a final or decisive one.
Acknowledgements
This research was funded by the Forethought Foundation. It was written by Luisa Rodriguez under the supervision of Arden Koehler and Lewis Dartnell. Thanks to Arden Koehler, Max Daniel, Michael Aird, Matthew van der Merwe, Rob Wiblin, Howie Lempel, and Kit Harris who provided valuable comments. Thanks also to William MacAskill for providing guidance and feedback on the larger project.
Summary
In this post, I explore the probability that if various kinds of catastrophe caused civilizational collapse, this collapse would fairly directly lead to human extinction. I don’t assess the probability of those catastrophes occurring in the first place, the probability they’d lead to indefinite technological stagnation, or the probability that they’d lead to non-extinction existential catastrophes (e.g., unrecoverable dystopias). I hope to address the latter two outcomes in separate posts (forthcoming).
My analysis is organized into case studies: I take three possible catastrophes, defined in terms of the direct damage they would cause, and assess the probability that they would lead to extinction within a generation. There is a lot more someone could do to systematically assess the probability that a catastrophe of some kind would lead to human extinction, and what I’ve written up is certainly not conclusive. But I hope my discussion here can serve as a starting point as well as lay out some of the main considerations and preliminary results.
Note: Throughout this document, I’ll use the following language to express my best guess at the likelihood of the outcomes discussed:
Case 1: I think it’s exceedingly unlikely that humanity would go extinct (within ~a generation) as a direct result of a catastrophe that causes the deaths of 50% of the world’s population, but causes no major infrastructure damage (e.g. damaged roads, destroyed bridges, collapsed buildings, damaged power lines, etc.) or extreme changes in the climate (e.g. cooling). The main reasons for this are:
Although civilization’s critical infrastructure systems (e.g. food, water, power) might collapse, I expect that several billions of people would survive without critical systems (e.g. industrial food, water, and energy systems) by relying on goods already in grocery stores, food stocks, and fresh water sources.
After a period of hoarding and violent conflict over those supplies and other resources, I expect those basic goods would keep a smaller number of remaining survivors alive for somewhere between a year and a decade (which I call the grace period, following Lewis Dartnell’s The Knowledge).
After those supplies ran out, I expect several tens of millions of people to survive indefinitely by hunting, gathering, and practicing subsistence agriculture (having learned during the grace period any necessary skills they didn’t possess already).
Case 2: I think it’s very unlikely that humanity would go extinct as a direct result of a catastrophe that caused the deaths of 90% of the world’s population (leaving 800 million survivors), major infrastructure damage, and severe climate change (e.g. nuclear winter/asteroid impact).
While I expect that millions would starve to death in the wake of something like a global nuclear winter, my best guess is that hundreds of thousands to hundreds of millions of the survivors of the initial catastrophe would be able to subsist using leftover food stocks and supplies, before eventually working out how to feed themselves through traditional agriculture (in regions where this was still possible) and fishing and/or modified agriculture (using methods that don’t rely on climate factors like warm temperatures and regular precipitation).
For catastrophes that cause temporary climate effects, survivors would only need to subsist using non-traditional agriculture for a decade at most. After that, survivors could revert to traditional agriculture, which would be much easier.
Survivors might also face threats from radioactive fallout from nuclear detonations and power plant meltdowns. While this would increase the death toll in the years following the collapse, I expect it wouldn’t be widespread enough to be immediately fatal to everyone, nor would it cause fertility rates or life expectancy to decrease enough to threaten extinction.
Case 3: I think it’s fairly unlikely that humanity would go extinct as a direct result of a catastrophe that caused the deaths of 99.99% of people (leaving 800 thousand survivors), extensive infrastructure damage, and temporary climate change (e.g. a more severe nuclear winter/asteroid impact, plus the use of biological weapons).
I expect hundreds of thousands of the survivors of the initial catastrophe would be able to subsist using leftover food stocks and supplies, before eventually working out how to feed themselves through traditional agriculture and fishing and/or modified agriculture (similar to the case of a nuclear war-only catastrophe).
A naive back of the envelope calculation (BOTEC) makes me think that surviving populations would either stay above the minimal viable population (the minimum number of individuals necessary to maintain the level of genetic diversity to repopulate Earth), or find each other (by crossing distances as large as England or Wales). My confidence in this conclusion is higher for scenarios where survivors have long-distance communication, long-distance transportation, or are confined to a smaller geographic region (e.g. the Southern Hemisphere).
My main uncertainties are around 1) whether I’m underestimating the frequency and severity of shocks the survivors might face in the post-collapse environment, and 2) whether surviving groups might deliberately avoid each other — perhaps because of fears about pathogen spread — allowing their populations to dip/stay below the minimum viable population. While it seems likely to me that natural shocks would reduce the number of people that would survive in the decades following the initial catastrophe — likely to be somewhere in the tens of thousands — I think it’s fairly unlikely that either of these would actually lead to extinction. But I’m unsure.
My best guess is that the turning point at which extinction goes from under 50% to over 50% is between 99.999% population death (80,000) and 99.9999% (8,000) population death (even before considering additional starting conditions like infrastructure damage or climate change). When the surviving population is in the tens of thousands and lower, it starts to seem more likely that the obstacles the survivors would face — difficulties growing food, random natural shocks — would be enough to wipe out the remaining groups. The probability of human extinction at these levels of population death would of course be even higher if the initial catastrophe caused additional environmental obstacles (like climate change, radiation).
General arguments and considerations that kept coming up and had a big impact on my conclusions:
Historical survival and resilience. In fact, we know that humans have survived for thousands of years “without technology,” and in a very large variety of climates and environments.
The grace period. The period after the initial catastrophe — during which survivors would be able to access leftover supplies, goods, and some infrastructure — would be enormously helpful in ensuring humanity’s survival before the survivors have learned basic survival skills.
With population loss comes “decorrelation” of survivors. A population reduced to hundreds of thousands of survivors would likely be spread out in relatively small bands or groups of hundreds or thousands of people. Given this kind of distribution of survivors, many, many failures would need to happen — at roughly the same time, and all over the world — for humanity to go extinct. But the types of failures the survivors are likely to face would only be weakly or not at all “correlated” between groups of survivors or across time (e.g. multiple generations). In other words, even if some failure mode is very likely per group and time period (a drought may be the nail in the coffin for groups of survivors in one area), this still doesn’t suffice for human extinction, as many other groups would be unaffected.
Historically, the fact that different groups of humans were relatively independent (spread out, affected by different social and environmental pressures) seems to have made humanity relatively robust. We can point to lots of societies that collapsed and didn’t recover, but humanity as a whole has been very resilient — in part because it’s been made up of lots of heterogeneous groups spread out all over the planet.
Non-uniformity of the initial catastrophe’s impacts. One of the ways that different survivor groups will become decorrelated stems from the fact that groups are likely to end up in regions that have been differentially affected by the initial catastrophe. For example, some regions might be completely destroyed, while others will have more moderate infrastructure damage.
In fact, it’s extremely difficult to come up with a scenario in which a catastrophe would have uniform effects all over the world — where the entire world has the same degree of infrastructure damage, the same amount of goods and supplies left around, the same severity of any changes to the climate, the same rates of pathogen spread and mortality. So while it’s easy to imagine that a catastrophe that caused civilizational collapse might produce unlivable conditions in some regions, other regions wouldn’t be as badly affected (another example: a nuclear winter might make the Northern Hemisphere unsuitable for agriculture, but not all of the Southern Hemisphere). That geographic variation means that even if we can imagine a scenario that looks unsurvivable for some regions, people who find themselves in (or make their way to) less affected areas should have a much better chance of survival.
The population loss would have to be incredibly extreme to lead to extinction. The minimum viable population (100–1,000 people, assuming a sufficient balance of the sexes) is really, really small compared to the current population: Just 0.00001% of the people alive today would need to survive for repopulation to be possible. In a sense, there’s a really “high bar” for extinction. To get there from a catastrophe that caused 90%, 99%, or even 99.9% population death seems to actually require one to hold exceptionally pessimistic views about how survivors would fare in the event of civilizational collapse. Consider the following toy calculation:
Note: In the following calculation, I assume each group’s fate is independent of the fates of other groups. See footnotes and discussion in the body of the document for comments on the strength of this assumption.[1] The higher the true correlation between survivor groups, the more my toy calculations will cause me to underestimate the probability that all of the survivor groups would be wiped out.
With 8 million survivors, even extremely pessimistic views of the likelihood that any one group gets wiped out don’t necessarily lead to high probabilities for extinction in most scenarios. For example:
Even if you thought any one group of 100 or 1,000 survivors had a 99% chance of being wiped out (by failing to feed itself, to find water, to survive natural shocks), it would still be virtually guaranteed that at least one group of survivors would survive.
If you thought there was a 99% chance that each one of 800 groups of 10,000 people would be wiped out, there would still only be a 1 in 3,000 chance of extinction.
The probability of extinction would be higher (45%) if you believed that larger groups of 100,000 would also have a 99% chance of being wiped out. But to hold that view, you’d have to think that out of a group of 100,000 people, there’s a 99% chance basically none of them would overcome the obstacles of the post-collapse environment. I grant that there would be many obstacles: Survivors would have to work out how to grow their own food (potentially in the face of extreme global cooling), how to find and purify water, how to build shelter, potentially how to fish. They would have to survive disease outbreaks, hurricanes, tsunamis, conflict with other desperate survivors, possibly radiation contamination from nuclear war or the meltdown of nuclear reactors. But even so, my view is that the probability that a given population of 100,000 people would be completely wiped out — even facing all of those obstacles — wouldn’t be as high as 99%.
Similarly, the probability of extinction is very high (92%) if you think that any given group of 1 million survivors has a 99% chance of being wiped out. But again, to believe extinction risk was that high, you’d have to think that there would be a 99% chance that basically none of those 1 million people would survive and reproduce. I don’t find that plausible.
What is the likelihood that civilizational collapse would cause human extinction?
Introduction
Even if a catastrophe were not so destructive as to cause immediate extinction, it could be destructive enough to cause civilizational collapse. By this I mean something like the total failure of the world’s systems of governance and critical resource systems (e.g., industrial food, water, and power systems). Some have argued that these conditions could drive humanity to extinction by making it impossible for the survivors to maintain numbers above the minimum viable population — especially if the catastrophe had other longer-lasting effects (e.g., severe warming or cooling). Below, I discuss some of the approaches I’ve taken to estimating the likelihood that civilizational collapse would cause human populations to fall below the minimum viable population.
Importantly, different catastrophes could cause civilizational collapse, and would leave survivors in dramatically different conditions. The dimensions along which these conditions vary that I see as most important are: levels of population loss, infrastructure damage (e.g. damaged roads, destroyed bridges, collapsed buildings, damaged power lines, etc.), the severity and persistence of any climate impacts, and the severity and persistence of worldwide conflict. In what follows, I’ll use three case studies to discuss how I think each of these factors would affect the likelihood that civilizational collapse would cause human extinction.
Before diving in, a few important notes:
First, when I say “critical systems” in this piece I mean industrial agriculture (but not non-industrial “traditional” agriculture), water transport and purification, power production and distribution, transportation, telecommunications, and effective governance, unless otherwise noted. And when I discuss the systems having “collapsed,” I mean they are producing little to none of their usual outputs. A civilizational collapse, as I will use the term, is an event in which approximately all of the entire world’s critical systems are collapsed simultaneously for over six months.
Second, this piece focuses on whether civilizational collapse would lead more or less directly to human extinction without additional catalyzing conditions, like a second catastrophe (either soon or long after the first) or centuries of economic stagnation that eventually end in human extinction. I hope to explore both of these other scenarios in future posts.
Third, even if the population dropped below the minimum viable population, human extinction would not be inevitable. Rather, the likelihood of recessive genes that lead to lower fitness would go up, but these wouldn’t necessarily lead to extinction (though they would increase the risk of it). To illustrate, consider that the 10 million Ashkenazi Jews living today descended from a population of just 350 individuals who lived between 600 and 800 years ago (Commun, 2014). While Ashkenazi Jews suffer from higher rates of breast cancer, ovarian cancer, congenital deafness, cystic fibrosis, gaucher disease, Tay-Sachs, Parkison’s, and other diseases, their population grew rapidly over the last 500 years nonetheless. Given this, my bottom-line extinction probability estimates are therefore an upper bound on the likelihood of the probability of extinction.
Case 1: 50% population loss, no infrastructure damage, no climate change (e.g. a limited pandemic)
I’ll start by discussing what might happen in the wake of a relatively moderate catastrophe — one that causes the deaths of about half the population (4 billion survivors), but which doesn’t arise from global conflict, and which doesn’t cause any major infrastructure damage or climate change.
What is the likelihood that civilization would collapse under these conditions?
Before getting into the likelihood that society would recover from civilizational collapse under these starting conditions, I’ll briefly discuss whether we should expect human civilization to actually collapse in my sense in this scenario.
Expert judgment: One expert told me that some resilience experts hypothesize that the economy would collapse (i.e. currency might lose its value, international trade might cease, etc.) in a case where 40% or more people had died or were not working[2] (though I haven’t been able to find a written source corroborating this). Even if that’s true, it’s not obvious that civilization would completely collapse if the world economy did. But it certainly seems possible: The goods and services required for most industries to run are not produced locally, but globally — they are accessed through international trade that presumably wouldn’t exist if the global economy collapsed.
There are people who think civilizational collapse is likely even at (or below) 50% population loss. (This includes Lewis Dartnell, author of The Knowledge, and several of the academics in the PIIRS Global Systemic Risk Research Community at Princeton University.) They often appeal to the fragile and interconnected nature of the modern globalized economy: Many critical systems have features that make them more efficient in the short run but less robust to large shocks (examples in footnote).[3] People who are pessimistic about civilizational robustness often appeal to this kind of fragility.
I’m more optimistic. Despite our increasingly interconnected systems, I believe that even in a catastrophe of this scale, our systems would be resilient enough to stay functioning — at least in some places. At current population levels, this catastrophe would still leave four billion survivors. I find it hard to imagine that four billion people working together wouldn’t be sufficiently adaptive, creative, and knowledgeable enough to keep critical systems running in some places, especially if certain regions are less affected than others. Without infrastructure damage, climate change, or major ongoing conflict, I strongly suspect that at least some regions would keep their critical systems online.
Indeed, while a few disagreed, most experts I spoke with shared the view that, while it seemed plausible that critical systems might go down temporarily at the regional level, it seemed exceedingly unlikely that all critical systems would fail all over the world in Case 1.
I explore the possibility of civilizational collapse from this sort of catastrophe in any case, because 1) I could be wrong, and 2) I think it’s a useful baseline with which to compare more severe catastrophes.
In a scenario in which a catastrophe of this scale does cause civilizational collapse, whether humanity goes extinct as a direct(ish) result of that collapse would depend on whether humanity could survive without critical systems, and for how long, and how quickly humans would be able to bring critical systems back online (if at all). I discuss this next.
What is the likelihood that humanity would survive without key systems (food, water, power) in the years immediately after a catastrophe?
Time spent on this section: 5-10 hours
Types of sources: Popular non-fiction books, op-eds in well-regarded newspapers, and expert interviews
If the survivors of the catastrophes aren’t able to get critical systems back online very quickly after the initial catastrophe, would they be able to survive? I took a mix of approaches to answering this question:
Historical track record: The genus homo survived for thousands of years without complex support systems, practicing hunting and gathering for hundreds of thousands of years, and Homo sapiens practiced settled agriculture for over ten thousand years. They did this in the face of changing climates and predators, with only rudimentary tools. This is at least some evidence that the survivors of a catastrophe could do this again — plausibly for thousands of years (remembering that the scenario we’re considering assumes the catastrophe didn’t cause any meaningful changes to the climate or natural environment).
On the other hand, most humans living today wouldn’t know how to survive without the support of our modern food and water systems. This might appear to suggest that the historical track record doesn’t tell us as much about the likelihood of survival among humans who face the collapse of civilization today.
Case study: To illustrate how humans might be unable to survive without critical systems, we can take a look at a few case studies. There are many examples of humans failing to survive in new environments, and/or after losing access to support systems on which they were reliant. One set of cases I looked into were those of stranded or lost European explorers who died in large numbers when forced to try to survive in the wilderness.
In the Secret to Our Success, Joseph Henrich describes several cases where groups of explorers — although physically fit, highly trained, and equipped with extensive supplies — died of disease and starvation because they couldn’t exercise basic survival skills in the environments in which they found themselves. At the same time, local populations were thriving in the environments in which the explorers perished, because of their large body of highly specific, non-obvious, detailed cultural knowledge. For example, the Franklin Expedition, a group of over 100 explorers who became stranded in the Arctic, slowly died because they could not practice the hunting techniques practiced by the local Inuit for millenia. Even lost expeditions in much more forgiving and food-rich environments — the Burke and Wills expedition in Australia and Narvaez Expedition in Florida, for example — saw entire expeditions of explorers perish due to ignorance of their environment.[4]
Disanalogies from those case studies: All that said, many humans likely would know how to survive without modern food and water systems. For example, I expect that most subsistence farmers — of which there are probably between 1 and 2 billion[5] — would at the very least have the knowledge and skills to survive by growing their own food. While flow-through effects of the catastrophe might mean that some proportion of these subsistence farmers wouldn’t be able to access key inputs for subsistence agriculture (or would be displaced from where they know how to survive), I expect at least 10% of them, at a minimum, would — leaving very likely more than 50 million people who would likely have the knowledge, skills, and resources to survive even without global infrastructure and centralized systems.
In addition to these people, there are apparently as many as 3.7 million Americans who identify as “preppers”. Given that most people who identify as preppers aren’t actually prepared to survive for years without critical systems (they mainy stockpile canned food and even build bunkers without necessarily learning how to acquire and prepare food and water in the natural environment), it seems plausible that as few as 1% of preppers (37,000 individuals) would survive. But even then, that 1% alone would keep us well above the minimum viable population of on the order of 100-1,000 survivors (assuming an even enough sex mix).
Even for people who don’t already know key survival skills — how to hunt, gather, and/or grow food — there would be opportunities to learn them. For example, there would likely be information sharing between those who did and those who didn’t, plus a great deal of written information on the topic in libraries and in homes. And this information would cover all of the world’s natural environments — so post-collapse humans wouldn’t be without key knowledge in new environments the way the European explorers were. To demonstrate, I searched the SAS Survival Handbook, Amazon’s best-selling wilderness survival guide, for the key pieces of information Henrich notes would have been critical for the explorers’ survival in the stories above. I found almost all of the relevant bits of information and techniques easily (and for bits that were missing, usually because they were incredibly specific, I found close substitutes).
Certainly, these survival skills would be hard to learn. Some survival skills, like starting a fire with only things found in the natural environment, are notoriously difficult to master. But many survivors wouldn’t have to learn the hardest skills: lighting fires with only sticks or flint, building shelters with only natural materials, making receptacles to hold water from natural resources alone, processing dead animals (making clothes from tanned hides, making fuel from their fat, carving bones into knives, needles, or buttons). They’d have access to knives, lighters, matches, tarps and plastic bags (for shelter) wool socks, plastic water bottles (to purify and carry water), salt (to preserve foods), alcohol (to treat wounds), thread/rope, and cooking oil (to burn). Some survivors would find iodine tablets, fishing rods, and compasses. Indeed, most survival skills look much less difficult with modern technologies.
The grace period: What’s more, at least some of these survivors would have time to learn survival skills during a period in the immediate aftermath of the catastrophe when systems have failed, but there is still a considerable stock of goods available for immediate consumption: food, water, batteries, and vitamin supplements in grocery stores; clothes in malls; cars, planes, ships, and computers that still work; gasoline at the pumps that could be siphoned out even if the power grid was down; medications in pharmacies; generators; books in libraries and in houses, etc. During this period, I expect that at least some survivors (maybe 5–25%) without experience practicing subsistence farming, hunting, and gathering would be able to meet their basic needs while learning those skills.
Without this grace period, I think it’s still fairly likely that hundreds of thousands or millions of people (largely subsistence farmers) would be able to survive the collapse, but I’m less confident. So I decided it was worth doing a BOTEC to understand the likelihood that there would be such a grace period, and how that period would last.
Note: The following BOTEC relies on particularly poor sources, makes a bunch of dubious assumptions (discussed more below), and I’m not confident I’ve thought of all of the most important supplies. It should be considered very rough.
BOTEC: To estimate whether and how long this kind of grace period might last given different population sizes, I investigated the amount of various goods/supplies that would be available in the post-collapse environment. In addition to goods that would facilitate human survival, I also considered goods that might be essential (or at least very helpful) in making use of the goods that are essential to human survival (e.g., humans don’t need petrol to survive, but that might facilitate movement, which might be essential to accessing a distributed food supply).
Note some important assumptions made here, that would meaningfully affect how realistic these estimates are:
I’m assuming food stocks wouldn’t perish before they were eaten (or that food stocks would be deliberately consumed in order of perishability, to ensure stocks last as long as possible). If this assumption doesn’t hold, the grace period would be shorter, perhaps by a lot.
I’m ignoring animal agriculture and cannibalism, in part because without a functioning agriculture system, it’s not clear to me whether enough people would be able to consume living beings. If this assumption doesn’t hold, the grace period could be a fair bit longer.
I didn’t look into whether the food stocks include stocks intended for animal agriculture. If they don’t, it’s possible I’m missing out on a huge amount of grain stock that could be consumed by humans — lengthening the grace period considerably.[6]
I find that the goods and supplies left after an initial catastrophe would only support a population in the billions for several months to a year — not very long at all. Given this, I expect a large fraction of the survivors who survive this initial catastrophe to die during a period of violent competition following the collapse.
I expect this period of violence to last until the survivors reach some sort of equilibrium, where they perceive there to be sufficient food, water, and shelter. The reason for this view is that it seems very irrational for groups to keep fighting if there are enough resources to go around. Of course, humans aren’t always rational, so I imagine some fighting might persist. But I think it’s very likely that many survivors would correctly conclude it would be more dangerous to fight each other than to accept a relatively low risk of not being able to meet their basic needs because of resource scarcity, if resources weren’t in fact too scarce to support the population.
Carrying capacity for hunter-gatherers and subsistence farmers: Importantly, without the technologies that enable humans to produce enormous amounts of food using a fraction of the land and water as our ancestors, the Earth would only be able to sustain a fraction of the human population it does now. Just how large a population it could support depends on the means of subsistence employed by the survivors. The carrying capacity of the Earth for hunter-gatherers is thought to be around 10 million if the survivors regress to pre-paleolithic levels of technology (if they lose, for example, flakes, handaxes, controlled use of fire, and wooden spears) (Taiz, 2013). The carrying capacity is hypothesized to be in the low billions for settled subsistence farmers — possibly as high as three billion for mostly vegetarian populations with particularly advanced agricultural systems (Ellis, 2013).
Bottom line: Taken together, if a Case 1 scenario leads to collapse (but not to additional catastrophes), I expect that millions to hundreds of millions — possibly even a few billion — of the 4 billion survivors would be able to learn enough survival skills to survive indefinitely even after the grace period. My biggest uncertainty is whether I’m underestimating the difficulty of surviving in a post-collapse world with only basic knowledge of subsistence farming and/or survival skills.
What is the likelihood that humanity would survive in the event of conflict immediately following the catastrophe?
Time spent on this section: 1–2 hours
Types of sources: Academic literature, expert interviews, and speculation
Literature review: Sociologists trying to understand human behavior in the aftermath of shocks like natural and manmade disasters largely agree that violence is not the most common reaction to emergencies (Drabek and McEntire, 2003). Rather than looting, exploitation, and other antisocial behavior, academics have found that groups tend to cooperate in the aftermath of disasters.
But a catastrophe of this magnitude would be unprecedented, and meaningfully different from the disasters that researchers have been able to observe. Unlike smaller scale disasters, where even those most impacted by the disaster could reasonably expect that help was on the way, post-collapse survivors would have no reason to expect this. As I concluded above, the leftover goods and supplies wouldn’t sustain all of the initial 4 billion survivors for very long. Without the hope of support — and facing the threat of starvation and other dangers — I strongly suspect that the limited availability of leftover supplies would almost certainly cause violent conflict at both the group and individual level. Specifically, I expect that subsets of the survivors — likely groups with access to weapons — would use violence (or the threat thereof) to hoard the surviving supplies, killing their competitors or letting them starve to death.
But as discussed, I expect any violence would mostly subside once some equilibrium had been reached — at the point where ~all of the surviving population could reasonably expect to be supported by the available resources (in other words, at the point at which the population dropped to or below whatever the Earth’s carrying capacity was). It’s at that point that regular conflict and violence would actually make the environment more dangerous than coexisting non-violently and sharing the available resources. And while resources would be somewhat more depleted by an initially large group of survivors all consuming them — and by the use of some resources directly in inter-group violence — therefore shortening the grace period, I don’t expect this would change my bottom line (but I’m uncertain about this).
My best guess is that this violent competition would whittle the number of survivors down to the hundreds of millions or fewer — a population for whom the supplies might last up to a decade or more. This is based on my best guess about the population size that I expect could be robustly supported by the supplies left during the grace period and the approximate hunter-gatherer carrying capacity of the Earth.
Bottom line: I don’t expect violence to drive the surviving population so low that humanity would go extinct. Rather, I think it’s extremely likely that violence would mostly disappear once the population had gotten small enough that all of the remaining survivors could expect to survive indefinitely on the available resources — probably hundreds of millions or fewer. My key uncertainty is whether initially large numbers of survivors, many fighting each other, would shorten the grace period enough that survivors wouldn’t have enough time to learn survival skills they’d need to survive after the grace period.
Concrete example: A moderate pandemic
So what, concretely, do I think would happen in the event of a catastrophe like a “moderate” pandemic — one that killed 50% of people, but didn’t cause infrastructure damage or climate change?
My best guess is that civilization wouldn’t actually collapse everywhere. But if it did, I expect that the ~4 billion survivors would shrink to a group of 10–100 million survivors during a period of violent competition for surviving goods in grocery stores/distribution centers, food stocks, and fresh water sources. I expect these supplies would last somewhere between a year and a decade or more — on the shorter side if the period of violence were prolonged, causing a large proportion of the supplies to be used up by larger populations before the population stabilizes at a lower number. After those supplies ran out, I expect that up to 10 million of those survivors would survive if practicing hunting and gathering, or as many as hundreds of millions if practicing advanced subsistence agriculture (after learning any skills they didn’t possess already during the grace period).
It would be reasonable to be skeptical of any or even several of the individual claims I’ve made here. But a toy calculation gives at least a rough indication of how wrong they would have to be for my bottom line to change.
Note: I assume the likelihood that any one group gets wiped out is independent from the probability that another group does. I think this is somewhat reasonable, given that there are many things that affect individual groups or subsets of groups without affecting all groups. For example, a contaminated water source used by a few groups might cause disease rates in that group to go up significantly, perhaps even causing the groups to be wiped out, but groups far away wouldn’t be affected. The higher the true correlation between survivor groups, the more my toy calculations will cause me to underestimate the probability that all of the survivor groups would be wiped out.
With 4 billion survivors, the degree of pessimism you have to have about their ability to survive to end up believing that no groups would survive indefinitely is actually very extreme. The exact beliefs you’d have to have would depend on whether survivors were concentrated into a few big groups, or distributed across many smaller ones. Specifically:
Even if you thought any given group of between 10 survivors and 1 million survivors had a 99% chance of being wiped out, it would still be virtually guaranteed that at least one group would survive.
If you thought every single one of 400 groups of 10 million people had a 99% chance of being wiped out, you would still only think there was a 1.7% chance of extinction. And, to hold that belief, you’d have to think that despite having extremely large populations (larger than any US city), none of those people would be able to work out how to practice very basic farming or fishing, or that none of them would survive conflict within the group, or that a natural disaster would kill all of them. Personally, I find this view implausible.
Similarly, the probability of extinction is pretty high if you think that any of 400 groups of 100 million survivors has a 99% chance of being wiped out. Again, though, to believe extinction risk was that high, you’d have to think that there’s a 99% chance that none of the 100 million people would overcome obstacles posed by the absence of critical systems and violence. (For reference, only 14 countries have a population of 100 million or higher.)
Given this, I believe it’s exceedingly unlikely that this scenario would lead to human extinction.
Case 2: 90% population loss, infrastructure damage, and extreme climate change (e.g. nuclear war that caused nuclear winter)
In a scenario in which a catastrophe causes the deaths of 90% of the population (800 million survivors), major infrastructure damage, and climate change — for example, a severe, global nuclear war that caused a nuclear winter — I believe the question of whether humans would be able to meet their basic needs becomes more difficult.[14] The questions I consider for this scenario are:
What is the likelihood that survivors are able to continue to survive using traditional forms of agriculture, given a catastrophe that causes severe infrastructure damage and climate change?
What is the likelihood that radiation causes extinction?
What is the likelihood that humanity would survive in the event of conflict immediately following the catastrophe?
What is the likelihood that survivors are able to continue to survive using traditional forms of agriculture?
Time spent on this section: 2–3 hours
Types of sources: Academic literature, non-academic reports, and expert interviews
Expert judgment: Several experts, including ALLFED director David Denkenberger, have affirmed this conclusion — they do not expect humanity to dip below the minimum viable population even in relatively extreme sun-blocking scenarios.
Literature review: The nature of all of the catastrophes we know of that would cause extreme global cooling (e.g. nuclear winter, asteroid impacts) would have unevenly distributed impacts — causing extreme global cooling in some parts of the world, but more moderate cooling in others. For example, in the case of a nuclear war between the US and Russia, nuclear winter models suggest that the most severe climate effects would be limited to the Northern Hemisphere, where temperatures would fall by 10–30 degrees C. But in the Southern Hemisphere, and especially at the equator, those effects would be much less severe: between 5–10 degrees Celsius. With heterogeneous impacts like this, it’s likely that agriculture would still be possible in some regions — especially in New Zealand and Australia, and possibly in South America and Central Africa.[15]
To be clear, I’m describing a very grim scenario, in which basically everyone in the Northern Hemisphere — and in many parts of the Southern Hemisphere — would be unable to grow food using standard agricultural techniques. Given this, I expect there would be mass starvation and violent competition and conflict until a new equilibrium was reached, one where the remaining survivors didn’t exceed the Earth’s carrying capacity. While I expect this would be a truly terrible period of widespread suffering, I believe this equilibrium would be reached long before the population got anywhere near the minimum viable population. My best guess is the population would fall to hundreds of thousands to tens of millions, but not much lower.
While I haven’t looked into this much, I feel fairly convinced that hundreds of thousands or millions of people could survive using traditional approaches to agriculture in parts of the world with more moderate climate effects (and basic mitigation strategies, like switching to crop types that are more resilient to temperature and precipitation fluctuations). And as with Case 1, at least some of the survivors in a Case 2 scenario would probably be able to survive the immediate aftermath of a catastrophe that caused civilizational collapse by exploiting food and other supplies in stores and larger stockpiles. This would give survivors some buffer time to learn additional skills required to survive once those supplies run out (e.g. fishing) or develop the techniques necessary to produce food using methods that don’t rely on climate factors like warm temperatures and regular precipitation.
BOTEC: The longer the buffer time, the more likely humanity would be to subsequently survive. But there are a number of different considerations (relative to Case 1) that affect the calculus of just how long such a grace period would be in the context of a catastrophic event like a nuclear war that killed 90% of people and caused a nuclear winter. So I’ve done a similar exercise to the one above where I try to account for some of those differences.
Note: As above, the following BOTEC relies on particularly poor sources, makes a bunch of dubious assumptions (discussed more below), and I’m not confident I’ve thought of all of the most important supplies. It should be considered very rough.
Bottom line: I think it’s extremely likely that these supplies would last somewhere between around a year and a decade or more. I expect it would be closer to the lower end, given that competition and violence could lead to the depletion of supplies more quickly than if the population were reduced to a smaller number by the catastrophe directly.
All this in mind, I think it is very likely that the survivors would be able to learn enough during the grace period to be able to feed and shelter themselves ~indefinitely.
What is the likelihood that radiation causes extinction?
Time spent on this section: 2–3 hours
Types of sources: Academic papers, Wikipedia, and interviews with experts
Literature review: In the aftermath of a nuclear war, radioactive fallout from the nuclear detonations would have long-lasting health impacts. In the most extreme nuclear war scenarios considered by academics (a nuclear war between the US and Russia and their allies, using 10,000 megatons (MT) of nuclear bombs), approximately 30% of the geographic area in the Northern Hemisphere would have enough fallout to be lethal to any adult in the area (Ehrlich et al., 1983). The current US and Russian nuclear arsenals don’t currently have that kind of megatonnage (they currently have closer to 2,500 MT). If we naively assume that radiation scales linearly, we might expect a modern day US-Russia nuclear war to contaminate up to 7.5% of the land area of the Northern Hemisphere. This may not sound like much, but consider that 95% of the world’s population lives on just 10% of its land area — meaning that 7.5% of land area could be home to millions or even billions of people. What’s more, tens to hundreds of millions more might be exposed to enough radiation to be more susceptible to cancer for the rest of their lives.
On top of this, there are currently around 440 civilian nuclear power reactors scattered around the world, and likely tens or hundreds more military reactors. These have fail-safes and automatic shut down measures that are designed to ensure that all of the nuclear material in these reactors would be safely contained in the event of a global catastrophe that meant people stopped attending to them. Concretely, these fail safes make sure that water continues to be circulated around the nuclear fuel to ensure it doesn’t get so hot it causes a meltdown — i.e., an event where the nuclear core partially or completely melts, which might allow the nuclear fuel to breach its multiple layers of containment and leak out into the environment. If fuel did reach the environment, the radioactive fallout could spread across continents, creating exposure levels ranging from immediately fatal (in areas ranging from tens to thousands of square kilometers) to non-lethal but causing potential higher rates of cancer and infertility.
But some of these fail-safes could plausibly fail during a catastrophe that caused infrastructure damage (or afterward, if any components of the fail system degraded). For example, some nuclear reactors rely on backup generators to power the pumps that keep water circulating in the core of the reactor. If those backup generators eventually all broke down, the reactor might melt down.
I currently don’t have a good sense of how likely these failures would be. Newer nuclear reactors rely on more robust safety systems, with parts that wouldn’t break down as easily. And all nuclear reactor safety systems are designed to account for infrastructure damage caused by earthquakes and other physical shocks.
But in a large-scale nuclear war, it seems very plausible that at least some nuclear reactors would melt down. My best guess is that this wouldn’t happen at a large scale, but even if it did, some areas would likely be far enough away from reactors to be spared the radioactive contamination. For example, Australia has just one nuclear reactor. Even if that reactor were to melt down, much of Australia would likely remain uncontaminated (Australia is just under 3 million square miles, and the Chernobyl meltdown is estimated to have contaminated under 60,000 square miles; and only a much smaller fraction of that area was sufficiently contaminated as to be lethal to humans).
Bottom line: While radioactive fallout from nuclear detonations and power plant meltdowns would increase the death toll in the years following the collapse, I expect it wouldn’t be widespread enough to be immediately fatal to everyone, nor would it cause fertility rates or life expectancy to decrease enough to threaten extinction. And at the very least, some areas are sufficiently far away as to be relatively safe from radioactive fallout.
What is the likelihood that humanity would survive in the event of conflict immediately following the catastrophe?
Time spent on this section: 1–2 hours
Types of sources: Academic literature, expert interviews, and speculation
Historical base rate: In Case 2, it seems slightly more plausible to me that violence would lead to human extinction than in Case 1, but still fairly unlikely. I don’t think human extinction could be caused by a conflict fought with conventional weapons; there would just be too many survivors (~800 million) to be killed in conventional warfare (compare this to WWI and WWII, during which ~20 million and ~75 million people were killed, respectively).
Weapons of mass destruction: My best guess is that the only way violence in the wake of a Case 2 civilizational collapse could directly lead to human extinction is if one group of survivors had access to and deployed weapons of mass destruction. This seems unlikely to me, first because it seems hard to imagine a group of survivors incapable of recovering critical infrastructure — and barely capable of meeting even their basic needs — would be able to successfully deploy weapons of mass destruction (though I’m not very confident about this).
Second, it’s hard to imagine a scenario where the use of weapons of mass destruction kills millions of survivors, spread all over the world, without modern technologies like transportation. For example, with potentially many survivor groups, it seems hard to imagine how nuclear detonations would kill ~everyone despite the fact that the groups would likely be spread out all over the world, potentially in small bands that can’t each be individually targeted. Similarly, it’s hard to imagine how a pathogen could spread ~everywhere when survivors would likely have greatly reduced mobility (the latter isn’t obviously impossible, but it at least seems exceedingly difficult to me).
There’s one counterargument I find somewhat persuasive, which is that it seems possible that all of the survivors might be confined to a relatively small area (for example, if only a small fraction of the Earth’s land area is habitable), making them more vulnerable to a single, large attack. If this were the case, it’s easier for me to imagine that the use of weapons of mass destruction could kill all of the remaining survivors. This would presumably mean the aggressors would be killing themselves, which makes it seem even less likely to me. But we’ve seen humans come dangerously close to threatening their own survival before, often because human aggressors aren’t always good at predicting how cascading effects could threaten their survival as well. A random example to make this concrete: If all of the survivors of a nuclear war were confined to Australia, which might be less impacted by a nuclear winter, one group might choose to use nuclear weapons against another group, not realizing that the radioactive fallout or further climate change could make Australia uninhabitable, even for them.
Bottom line: I expect the survivors in Case 2 would not deploy weapons of mass destruction against their competitors, as it would likely pose a pretty big risk to the aggressor as well as the target. But I’m uncertain about this — humans have come close to making similarly self-destructive choices before. Thankfully, even if one group did use weapons of mass destruction against their competitors, I still think it’s very unlikely that their use would cause human extinction. This is because except in a few very specific and very strange scenarios, I expect the survivors would be too geographically distributed and disconnected to be wiped out by a single act of aggression. I therefore expect the result would be a much higher death toll, but not extinction.
Concrete example: A large nuclear war that causes a nuclear winter
So what, concretely, do I think would happen in the event of a catastrophe like a nuclear war that led to the death of 90% of the population, and caused severe infrastructure damage and significant global cooling?
I expect that, in addition to the billions of people killed in the initial catastrophe, hundreds of millions or more would likely die in the famines and violent competition that followed. But my best guess is that hundreds of thousands to hundreds of millions of the survivors of the initial catastrophe would survive this violent period. I think it’s extremely likely these survivors would be able to support themselves using leftover food stocks and supplies, before eventually working out how to feed themselves through traditional agriculture and fishing and/or modified agriculture (using methods that don’t rely on climate factors like warm temperatures and regular precipitation).
All of the catastrophes we know of that would lead to extreme cooling would only do so for 1–10 years, and agriculture would become possible again once the climate began to return to normal. At that point, it seems even more likely that the surviving humans would be able to meet their own basic needs by returning to traditional forms of agriculture.
My key uncertainties are around whether I’m putting too much weight on the idea that humans would figure out how to subsist without traditional agriculture just because it’s technically possible, and whether conflict could lead to extinction through channels I haven’t foreseen.
Another toy calculation suggests that these uncertainties probably aren’t troubling enough to change my bottom line.
Note: I again assume each group’s fate is independent of the fates of other groups. I actually think this is a pretty reasonable assumption in this case. I expect that the survivors of a catastrophe like a severe nuclear war would end up somewhat spread out (at least across the Southern Hemisphere), as doing so would create less competition for resources within a smaller area (I discuss this more later). The farther apart the surviving groups are, the less likely they are to be affected by the same shocks (natural disasters, disease outbreaks, conflict).
Additionally, in the event of a catastrophe like a nuclear war, transportation, communication, and other technologies that facilitate contact between geographically distributed groups would be enormously limited. This would further limit the extent to which each group’s fate ended up relating to another’s. There would be other sources of variation between groups that made their fates less correlated: Some groups might be made up mostly of farmers, while others will be made up of lawyers, some groups will tend toward cooperation, while others toward conflict, plus pure randomness (e.g. some groups might have a high proportion of survivors with genetic immunity to a particular disease).
But there are also factors that point in the other direction — factors that suggest the surviving groups would be at least somewhat correlated. For example, nuclear winter climate conditions, while nonuniform, would nonetheless impact all surviving groups. Similarly, more severe natural disasters might affect large regions, meaning that at least all of the survivor groups at the regional level might end up experiencing very similar challenges to survival simultaneously. Likewise, there might be things about “human nature” that would be shared amongst all survivors. For example, it’s possible that all of the survivors, having witnessed the initial catastrophe, would have similar psychological experiences — like shock, stress, and social distrust, among others — that would make it more difficult to survive and cooperate.
As above, the higher the true correlation between survivor groups, the more my toy calculations will cause me to underestimate the probability that all of the survivor groups would be wiped out.
With 800 million survivors, the degree of pessimism you have to have about their ability to survive to end up believing that no groups would survive indefinitely is actually kind of extreme. The exact beliefs you’d have to have would depend on whether survivors were concentrated into a few big groups, or distributed in many smaller ones. Specifically:
Even if you thought any given group of 100, 1,000, or 10,000 survivors had a 99% chance of being wiped out, it would still be virtually guaranteed that at least one group would survive.
If you thought there was a 99% chance that any one of 800 groups of 100,000 people would be wiped out, there would still only be a 1 in 3,000 chance of extinction.
The probability of extinction is higher (45%) if you believe that larger groups of 10 million would also have a 99% chance of being wiped out. But, again, to hold that view, you’d have to think that out of a group of 10 million people (again, bigger than the largest US city), not even a few hundred of those people would overcome the obstacles of the post-collapse environment (how to fish, how to farm despite global cooling, avoiding being killed by a hurricane or drought). I do not find this view very plausible.
Similarly, the probability of extinction is very high indeed if you think that any given group of 100 million survivors has a 99% chance of being wiped out. Again, to believe extinction risk was that high, you’d have to think that there would be a 99% chance that none of the 100 million people would work out how to survive (for reference, only 14 countries have a population of 100 million or higher).
Given all of this, my subjective judgment is that it’s very unlikely that this scenario would more or less directly lead to human extinction.
Case 3: 99.99% population loss, infrastructure damage, and climate change (e.g. a global war where biological weapons and nuclear weapons were used)
In a scenario in which a catastrophe caused the deaths of 99.99% of the population (800,000 survivors), in addition to major infrastructure damage and climate change (e.g. war using both nuclear and biological weapons), the survivors would face many of the same challenges faced in Case 2, such as food insecurity and the potential for ongoing conflict. Below, I’ll explore the additional obstacles faced by survivors in this scenario caused by the drastically smaller population, as well as the possibility of a persistent pathogen from a biological attack. Specifically, I’ve tried to understand:
Given that 99.99% population loss is obviously extremely significant, what is the likelihood that the survivors are so spread out that each individual group is below the minimum viable population size?
What is the likelihood that humanity would survive without key systems (food, water, power) in the years immediately after a catastrophe (during and after the grace period)?
What is the likelihood that a human population of 800,000 would survive natural shocks like natural disasters, climate fluctuations (drought), and disease?
My answers to each of these questions increases my credence in a Case 3 catastrophe leading to human extinction, but together they only increase it 10–20% over my credence in a Case 2 catastrophe leading to extinction.
What is the likelihood that the survivors are so spread out that each individual group is below the minimum viable population size?
Time spent on this section: 1 hour
Types of sources: Academic literature, popular nonfiction books, and expert interviews
With only 800,000 survivors, it seems much more plausible than in Case 1 and Case 2 scenarios that many of the surviving groups would be below the minimum viable population of 100–1,000 people. For example, it seems at least plausible that there could be 80,000 groups of 10 survivors — especially if there were pressures to stay in small groups or spread out geographically. Some people have argued that, having fallen below the minimum viable population, these groups would run the risk of dying out unless they were able and willing to find each other, forming larger (genetically viable) groups. So a critical question might be: how likely is it that groups of survivors would be able to find each other within a generation? This question breaks down further into:
Would they find each other by chance?
Could they coordinate to find each other if not?
Could they physically reach each other?
Would they want to find each other?
When thinking about whether 800,000 survivors in lots of small groups would stumble across each other by chance, I realized I didn’t have an intuitive sense about how far away groups of survivors would be from each other. To get a very rough sense, I did a simplistic BOTEC, estimating how much land area a group of survivors of a particular size would have to cross in order to find another group of a similar size, in expectation.
BOTEC: In the following toy calculation, I explored two scenarios: one where the survivors were distributed evenly across the world, and another where they were distributed evenly across the Southern Hemisphere (which might be the case in the event of a nuclear war in the Northern Hemisphere).
The BOTEC relies on the following simplifying assumptions:
The survivors are mobile and actively trying to find each other, rather than avoiding it. If I’m wrong about this, my estimates will overstate the likelihood of inter-group contact.
Survivors are evenly distributed across different regions. If I’m wrong about this, my estimates will probably overstate the likelihood of inter-group contact, but will probably understate the size of each group.
I don’t account for obstacles like mountain ranges, nor do I account for the fact that some groups would be separated by oceans. Rather, the following best reflects something like: the ease of reaching groups on the same continent with no major geographic barriers. This assumption means my estimates will overstate the likelihood of inter-group contact.
There were a few things I took away while making this table. First, any groups with populations of 1,000 or above should have enough genetic diversity to repopulate. And for groups of survivors made up of 10 to 100 people, the approximate land area each would have to cross to happen upon another group of survivors is pretty small. If the survivors were in small bands of around 10 people, they’d only have to travel across an area the size of Oxfordshire (in expectation) to happen upon another band. If the survivors were in groups of 100, they’d have to travel across an area the size of the Bay Area.
If the survivors had no way of coordinating, it could plausibly take years for them to happen across each other. But there are reasons to think it would happen even more quickly than that. First, there are likely to be points of disproportionately high traffic — attractor points — like water sources, cities (where survivors will go to scavenge supplies), roads/railroad tracks, etc. Additionally, in the first 5–10 years after the collapse, I think it’s fairly likely that at least some survivors would find ways to retain some motorized transportation, drastically increasing the area that could be covered over time. Unless the initial catastrophe was something that wiped out electronics, many cars would still be drivable, and petrol and diesel could be accessed with relative ease (until they went off after sitting there for five and ten years, respectively). Given this, I think it’s very likely that more than a few groups of survivors would find each other within a generation even if they had no means of communication.
But even that I think is too pessimistic, as I think it’s extremely likely that survivors would be able to employ at least some forms of long-distance communication to contact and connect with other survivors. Even with no advanced technology, survivors could create beacons by lighting fires at high ground. And with some luck, survivors might work out how to communicate using radios — either rebooting existing ones or making their own.[18]
Given this, I actually think it’s extremely likely that survivors would form one or more groups that were well above the minimum viable population of 100–1,000 people (assuming a sufficient balance of the sexes) at this level of population loss — even if survivors were spread out across the whole world and limited to basic forms of transportation (foot, non-motorized wheeled vehicles).
Importantly, though, if these groups were trying to avoid each other — perhaps because of fears of conflict or disease — it becomes a bit more plausible to think that small groups of survivors might stay below the minimum viable population, potentially leading to human extinction. But while I think it’s actually pretty likely that there would be at least some inter-group avoidance and conflict for at least a period of months or years after the initial catastrophe, I think it’s extremely unlikely that this aversion would exist and persist among thousands of different groups of survivors for a full generation. This is because I’d expect the benefits of larger populations (specialization, division of labor, etc.) to eventually outweigh the risks (within a period of years or decades at most) — for example as a human-borne pathogen died out after being unable to spread, or after populations were small enough that resource competition — and by extension, violence — had decreased substantially.
Bottom line: Taking into account both the logistics of coordinating and connecting with other groups of survivors, as well as the potential for decades-long intergroup avoidance, I think it’s extremely unlikely that there wouldn’t be any groups that ended up above the minimum viable population in the event of a catastrophe that left 800,000 survivors.
What is the likelihood that humanity would survive without key systems (food, water, power) in the years after a catastrophe?
Time spent on this section: <1 hour on top of the time spent above
Types of sources: Popular non-fiction books, random unvetted websites, government reports, well-regarded news websites, and expert interviews
As above, I did a BOTEC to understand how long the 800,000 survivors would be able to rely on surviving goods and supplies to meet their basic needs. I found that even the scarcest supplies would last at least 50 years (in theory), with more abundant supplies like grain stocks lasting hundreds of years (again assuming that 55% of stocks are lost during the initial catastrophe and ignoring perishability). I don’t present the full results here, because the limiting factor on how long these resources would keep survivors alive isn’t the number of survivors, but rather the perishability of the supplies. Given this, I think it’s at least safe to conclude that the surviving supplies and goods would likely keep survivors alive for well over a decade, maybe more.
During that period, as above, I expect survivors would learn to subsist without critical systems by hunting, gathering, fishing, and practicing modified and traditional agriculture and animal husbandry.
Bottom line: I expect these supplies would last somewhere on the order of several years or decades, giving survivors plenty of time to learn the survival skills they’d need to meet their basic needs after they eventually ran out/went off. Given this, I think it’s extremely likely that humanity would be able to feed itself basically indefinitely after a catastrophe killing 99.99% of people.
What is the likelihood that humanity would survive natural shocks like natural disasters, climate fluctuations (drought), and disease in the decades/centuries after the collapse?
Time spent on this section: ~1 hour
Types of sources: Our World in Data and expert interviews
While the surviving population in Case 3 is very small (for the decades or centuries until it grows again), one might worry that it could be driven all the way down to extinction as a result of natural shocks. I think it’s very likely that many groups of survivors would be seriously threatened — and some wiped out — by shocks like drought, hurricanes, disease outbreaks, or pests. But in the aftermath even of a Case 3 civilizational collapse, I expect survivors will be somewhat distributed around the world (or hemisphere) for several reasons.
First, it seems likely that the survivors would start off in different regions of the world just by chance. Second, it seems likely that travel would be limited, meaning that at least some groups would be completely disconnected from others. And third, because natural resources and surviving supplies would likely be distributed all over the world, a strategy of spreading out — in order to draw on all of the resources available and to avoid competition — would likely maximize the possible number of survivors. Given these pressures toward geographic distribution, any single shock might affect some or even many groups of survivors in a single region, but it’s hard to imagine any shock that would come close to affecting an entire hemisphere or even continent all at once. We can therefore expect that regional shocks like droughts and disease outbreaks wouldn’t be enough to wipe out groups spread out all over.
Historical base rates: Historically, natural disasters like wildfires, earthquakes, and droughts have been responsible for a pretty small share of global deaths. There aren’t great estimates for the death tolls caused by natural disasters before the mid-nineteenth century, but the most deadly year in the last century was 1931, when over 3.5 million people died as a result of natural disasters — just under 0.2% of the world population at the time. My impression is that, before the 20th century, regional shocks sometimes decimated local populations (and occasionally played a significant role in the collapse of a single civilization), but that the geographic distribution of human societies across the world meant those shocks never threatened the entire species.
There are a few reasons to think that shocks might be more frequent after a collapse. For example, anthropogenic global warming will continue to get worse (even without added emissions), leading to more climate volatility, including more frequent and severe hurricanes. And if shocks became more frequent, it becomes more plausible to think that all of the survivors might be affected by various natural shocks all at once. To my knowledge, they would not be so frequent as to mean we should expect human survivors would fare much worse than our ancestors did. They’d have to be either much more severe or much more frequent — probably both — but this is an area of uncertainty for me.
One scenario that seems like it could more plausibly lead to human extinction is one where, as described above, the survivors of the initial catastrophe are confined to a small area (like a region of Australia). Were this to be the case, it seems more plausible to me that a single shock could threaten human existence.
Bottom line: If groups of survivors were geographically distributed, as I imagine they would be, it’s hard to imagine how even a very severe natural disaster would lead to the deaths of all of the remaining survivors. If ~all of the survivors were in a small area — if there were basically no groups of hundreds or thousands of survivors anywhere else in the world — it seems plausible that all of the survivors could be wiped out directly or indirectly by a natural shock. But as I said above, I think this degree of geographic concentration seems extremely unlikely. So I think it’s extremely likely that humanity would survive natural shocks indefinitely.
Concrete example: A large nuclear war and the use of biological weapons, resulting in the death of 99.99% of the population, and causing infrastructure damage and climate change
So what do I think would happen in the event of a catastrophe like a global war involving both nuclear and biological weapons that led to the death of 99.99% of the population, caused severe infrastructure damage, climate change?
I expect hundreds of thousands of the survivors of the initial catastrophe would be able to subsist using leftover food stocks and supplies, before eventually working out how to feed themselves through traditional agriculture and fishing and/or modified agriculture (using methods that don’t rely on climate factors like warm temperatures and regular precipitation) (similar to the case of a nuclear war-only catastrophe).
Like before, for catastrophes that cause temporary climate effects, survivors would only need to subsist using non-traditional agriculture for a decade at most. After that, survivors could revert to traditional agriculture, which would be much easier.
At this level of population loss, a naive BOTEC makes me think that surviving populations would either stay above the minimum viable population, or find each other (by crossing distances as large as England or Wales). My confidence in this conclusion is further strengthened by the fact that there would be scenarios where survivors would retain or recreate long-distance communication, long-distance transportation, or are confined to a smaller geographic region (e.g. the Southern Hemisphere), making contact even easier.
While it seems likely to me that natural shocks would reduce the number of people that would survive in the decades following the initial catastrophe — likely to somewhere in the tens of thousands — I think it’s fairly unlikely that either of these would actually lead to extinction. This is because neither pathogens nor natural disasters have ever come close to causing human extinction, and have only relatively rarely even caused the collapse of a single human civilization. This seems largely due to the geographic spread of humans across the planet, which makes it hard for any given disaster to affect all groups of humans. I expect the post-collapse conditions to lead to a similar or greater level of geographic distribution.
Given this, I think it’s fairly likely, though far from guaranteed, that a catastrophe that caused 99.99% population loss, infrastructure damage, and climate change (e.g. a megacatastrohe, like a global war where biological weapons and nuclear weapons were used) would more or less directly cause human extinction.
My main uncertainties in Case 3 are around 1) whether I’m underestimating the frequency and severity of shocks the survivors might face in the post-collapse environment, and 2) whether surviving groups might deliberately avoid each other — perhaps because of fears about pathogen spread — allowing their populations to dip/stay below the minimum viable population.
Another toy calculation gives an indication of how wrong I would have to be for this scenario to lead to human extinction.
Note: As above, I assume each group’s fate is independent of the fates of other groups. In Case 3, I think this is a pretty robust assumption. I would expect survivors to be in relatively small, scattered, isolated groups for the reasons discussed above.
As discussed, the higher the true correlation between survivor groups, the more my toy calculations will cause me to underestimate the probability that all of the survivor groups would be wiped out.
With 800,000 survivors, the most plausible scenarios still seem unlikely to lead to extinction, even under very pessimistic assumptions about the vulnerability of each group of survivors.
If you thought any given group of 10 or 100 survivors had a 99% chance of being wiped out, it would still be virtually guaranteed that at least one group would survive.
If you thought there was a 99% chance that any one of 800 groups of 1,000 people would be wiped out, there would still only be a 1 in 3,000 chance of extinction.
If the survivors were in groups of 10 thousand people (80 groups total), and each of those groups had a 99% chance of being wiped out, there would be a 45% chance that humanity would go extinct. I think a 99% failure rate among groups of 10 thousand survivors is too pessimistic. I can imagine scenarios in which a single event or shock like a drought, nuclear meltdown, or natural disaster might have a chance at causing an entire population of 10,000 people to die. But I don’t think these are likely or frequent enough to put the overall probability of the loss of everyone in a group of 10,000 dying at 99%. Personally, I would put the probability that any of these groups is wiped out at 90%, at the highest. At this failure rate, the chance of extinction is about 1 in 5,000. But if you think 99% is reasonable, then it would be reasonable to think there is a 45% chance that this catastrophe would cause human extinction.
If the survivors were in groups of 100,000, and you put the probability that any one of those groups got wiped out at 90–99%, you could conclude that human extinction was pretty likely. Again, I think a 99% chance that a population of 100,000 is wiped out is too high, but reasonable people might disagree with me. I would put the probability that any group of 100,000 survivors might be wiped out in a post-collapse environment at 80% at the highest, which yields a probability of human extinction just below 17%.
Given all of this, my subjective judgment is that it’s fairly unlikely that this scenario would more or less directly lead to human extinction.
Other research directions
A few notes on other approaches to answering some of the questions I discussed that I considered but didn’t pursue, plus a rough guess at how long each approach would take and whether it would require specialized knowledge or skills:
I’d love to see someone host something like a wargame, where people (ideally actual subject-matter experts) adopted particular roles — a power plant operator, the leaders of influential countries, journalists, military generals, business executives, etc. — and participated in simulating a catastrophe. If modeled after a traditional war game, participants would have a scenario described to them, then would take turns making decisions (“moves”), each with the aim of achieving their respective interests/acting on their values. I think such a game might teach us a lot about how individuals and institutions would behave in the aftermath of such a catastrophe. (50+ hours / Special skills required)
Looking at lessons from science fiction books and films. I’m not sure how much there is to gain here, but I think it’s possible that works of sci-fi offer useful insights into what types of conditions survivors would face after a catastrophe of this type. A list of some of the best civilizational collapse science fiction is here. (5–10 hours / No special skills required)
I think there’s literature on society-level factors that inform how people respond to disasters. For example, I think I’ve heard somewhere that places with higher levels of social trust have lower levels of looting, hoarding, and other antisocial disaster behavior. This might be worth looking into more. (2–5 hours / No special skills required)
In my discussion of Case 2, I consider one type of catastrophe that could lead to 90% population death, infrastructure damage, and climate change, where the climate change I’m referring to is nuclear winter (decreased temperature and rainfall) and the infrastructure damage I’m referring to isn’t worldwide. But there are other cases one could choose with different considerations worth exploring. For example, in a scenario with an increase in the frequency of natural disasters (like extreme anthropogenic climate change), each group of survivors would be at a higher risk of extinction from things like hurricanes and droughts. This could have important implications on the bottom line for this case that make it meaningfully different from this one. It could be useful for someone to break down the starting conditions beyond what I’ve done (e.g. rather than thinking about climate change as one parameter, one could think about temperature, precipitation, and natural disasters separately; or, instead of thinking of infrastructure damage as one parameter, one could think about electronics, transportation, etc., separately). (10–30 hours / No special skills required)
I’m not sure how informative this would be, but I wonder if it could be interesting to see if anyone could try to take a survival guide and some basic camping supplies and try out various survival skills like building a fire and gathering an edible meal (without advanced preparation, the internet, or help from others). (3–30 hours / Maybe some adventurousness required)
How much more frequent/severe would natural shocks have to be to threaten surviving populations of different sizes? (2–10hours / Some math required)
I think this is a pretty reasonable assumption in the event of 99.9% population death. I expect that the survivors of a catastrophe this extreme would end up somewhat spread out (because this will mean less competition for the resources within a smaller area). The farther apart surviving groups are, the less likely they are to be affected by the same shocks (natural disasters, outbreaks, conflict). Additionally, in the event of civilizational collapse, transportation, communication, and other technologies that facilitate contact between geographically distributed groups would be enormously limited. This would further limit the extent to which each group’s fate ended up relating to another’s. There would be other sources of variation between groups: Some groups might be made up mostly of farmers, while others will be made up of lawyers; some groups will tend toward cooperation, while others toward conflict, plus pure randomness (e.g. some groups might have a high proportion of survivors with genetic immunity to a particular disease).
On the other hand, there are also factors that point in the other direction — factors that suggest the surviving groups would be at least somewhat correlated. For example, nuclear winter climate conditions, while nonuniform, would impact all surviving groups. Similarly, larger natural disasters might affect large regions, meaning that at least all of the survivor groups at the regional level might end up experiencing very similar challenges to survival simultaneously.
This last bit suggests that we could see economic collapse even if we weren’t seeing massive population losses, but rather just an unwillingness or inability of enough people to work, as might happen in a pandemic with a disease enough people were unwilling to risk contracting (or which made people sick enough not to work, even if it didn’t kill them).
Resource-pooling aims to improve the efficiency and cost effectiveness of systems that need to meet unpredictable demand. For example, demand for firefighters is both unpredictable and high-variance — there are zero or few fires most of the time, but sometimes there are several. To be prepared to respond to several fires that might all happen at once, a precinct would have to employ a lot of firefighters. But this means a lot of firefighters who are idle most of the time. To reduce inefficiency, most precincts pool resources, employing fewer firefighters while building in agreements between precincts that allow one to call on reinforcements from neighboring precincts.
Resource pooling is widespread: For example, banks pool capital reserves, and power plants have agreements allowing them to shift power demands to another power plant if demands are unusually high at one plant. While this makes these systems much more efficient — and also more effective in the short term — it makes them more fragile when viewed over long time frames, during which there are more opportunities for rare but extreme events to cause particularly high demand. Arguably, this kind of resource pooling is at least partly responsible for the severe ventilator shortages during the COVID-19 pandemic.
Because the components in a system with pooled resources are “coupled” (the performance of one is linked to the performance of others), resource pooling means that if 50% of the population were killed, and then even just a few components of a system with pooled resources becomes compromised, the whole system could fail. In other words, the reason systems with pooled resources are more fragile is because it means the nodes in a system become coupled — the performance of one is dependent on the performance of another. This means that if one fails, it’s much more likely that many others fail. To make this more concrete: imagine that a set of government hospital pools medical supplies, sharing between them when demand is particularly high at one hospital. Most of the time, this saves hospitals money, by allowing them to work well at lower safety margins (lower levels of surplus supplies). But when demand is high at all of those hospitals, they’ll all fail to meet demand for that supply. When the failure of one node causes others to fail, this is called a cascading failure.
Hyper-specialization is when systems require several subsystems to all work in order for them to have any output at all, because each ingredient of the output is produced in a separate subsystem. For example, a microchip manufacturer — rather than having several factories with the capacity to build microchips from start to finish — might build each individual subcomponent in different factories, such that failure of one such specialized factory will destroy the company’s entire manufacturing capacity. If a system is hyper-specialized in this way, then losing even a small percentage of its component subsystems (which seems like it could plausibly happen if 50% of the population were killed) would mean it would fail altogether.
In experimental literature that explores the effects of these approaches and their propensity for cascading failure, it’s often found that tightly coupled systems perform poorly over periods long enough to observe rare but extreme shocks (e.g. Elga, n.d.). For example, one study assigned participants to the role of the CEO of a power company (and tasked with the goal of maximizing profits) that supplied the power to ten cities. Some of the participants were allowed to use resource pooling — shifting surplus power supply in one city to meet unexpectedly high demand in another city. The participants were then allowed to choose what safety margins to adopt (the amount of power produced to meet an imperfectly predictable demand). Computer simulations were then run to see how often the participants’ imaginary power supply companies would be able to meet the power demands of the city. Participants whose systems were coupled were 35 times more likely to experience a total system failure, failing to meet the power needs in all ten cities (Elga, n.d.).
If the shock is as extreme as 50% of the inputs disappearing, it seems plausible that the system could fail altogether.
There seems to be a large and fascinating literature on the interconnectedness of systems, and how particular connections between systems make them particularly vulnerable (some examples: Buldyrev et al., 2009, Barigozzi et al., 2010). Unfortunately, a lot of this literature was inaccessible to me (it was highly quantitative, used a lot of network theory I don’t understand, etc.). But I think it could be worth looking into more.
Henrich describes in detail the highly specific knowledge that allowed local peoples to thrive in these environments. In the Arctic, this included the ability to identify the breathing holes that seals make in the ice, and to identify older ice which was low enough in saline to drink when melted, to name just a few. In Australia, it was a complex, non-obvious, multi-step process for processing one of the only available plants in the desert — the nardoo plant, which will slowly deplete the human body of vitamin B1 if not processed correctly. Henrich emphasizes that it was that collective knowledge, discovered and then passed down over generations, that allowed local populations to survive in these environments. Given that most people have not been taught this kind of knowledge, it seems plausible that we’d not have enough knowledge of the environment and agriculture to survive post-collapse.
26.7% of the world population derives its income through agriculture (p.46); this FAO report claims that “About two-thirds of the developing world’s 3 billion rural people live in about 475 million small farm households, working on land plots smaller than 2 hectares.” (p. 1).
I expect that survivors would be able to work out how to consume grain stocks, despite the fact that they may be unprocessed, meaning that they may not be immediately ready for consumption. I believe this would be an overcomable obstacle, but if I’m wrong, it could limit the grace period to the shorter end of the range above (a few months).
Details of how I thought about food rations: I assume rations imposed wouldn’t reduce caloric intake below 2,100 per day — which is the number of calories required to be offered to refugees in refugee food programs, like that offered by the World Food Programme. If all countries that currently consume over 2,100 calories imposed food rationing that limited intake to 2,100 calories, there would be a ~20% caloric surplus on average. Calculations here.
Assumes that 30 days worth of water bottles, consumed at the rate of 1 million per minute under normal circumstances, would be stored and accessible. Also assumes that people drink ~3 water bottles per day.
Here, food stocks is defined in the source paper as a group of “92 products” used “to reconstruct 50 years of aggregated food reserves, expressed in caloric equivalent (kcal), at the regional and global scales.” (Laio et al. 2016)
Note that many people don’t actually rely on any medications, so these are only critical to some. Additionally, perishability depends on the specific medication, but medications generally last 1–10 years before expiring.
In some places, power outages would occur within 4–24 hours. In regions where power is supplied by particularly robust power plants (e.g. the Hoover Dam), it’s theoretically possible that the power supply could go on without human support for several years. In either case, generators could be scavenged from roadwork/construction sites once centralized power supplies stopped working.
Note: Extreme climate change can mean lots of things, including increases or decreases in temperature, precipitation, and frequency of natural disasters like hurricanes, among others. The considerations I discuss under Case 2 won’t apply to all of these types of climate changes. See the section on other research directions for possible research projects that could fill this gap.
And while it’d be important for survivors to make sure they could meet their nutritional requirements, this might not be as hard as it seems. For example, consider that one could survive indefinitely on a diet of white potatoes, sweet potatoes, and butter (Nunn, 2009).
Given that there would be major infrastructure damage, it seems likely that a substantial portion of food stocks and other supplies would be destroyed during the initial catastrophe. It’s hard to say how much. For just a very rough sense, I assume 55% of these resources would be destroyed, based on the fact that about 55% of people live in countries with nuclear weapons or in nuclear weapons alliances (so, perhaps we can assume that the supplies destroyed would be proportionate to the population of countries likely to be affected by a global nuclear war).
Making a makeshift radio might be a surprisingly accessible technology, Lewis Dartnell explains in The Knowledge:
“During the Second World War, soldiers holed up at POW camps built their own makeshift radio receivers for music or news of the war effort. These ingenious constructions reveal the sheer variety of scavenged materials that can be jury-rigged to create a working radio. Aerial wires were slung over trees, or disguised as clotheslines, and sometimes even barbed wire fences were appropriated for the task. A good grounding was achieved by connecting to cold-water pipes in the POW barracks. Inductors were constructed by winding coils around cardboard toilet rolls, the scavenged bare wire insulated by candle wax, or in Japanese POW camps by applying a paste of palm oil and flour. Capacitors for the tuning circuit were improvised out of layers of tinfoil or cigarette-pack lining, alternating with newspaper sheets for insulation; the wide, flat device was then curled like a jelly roll to make a more compact component. The earphone is a trickier component to improvise and so was often salvaged from wrecked vehicles… Perhaps the most ingenious improvisation of all, however, was in creating the all-important rectifier, needed to demodulate the audio signal from the carrier wave. Mineral crystals like iron pyrite or galena were unobtainable on the battlefield, but rusty razor blades and corroded copper pennies were discovered to serve just as well. The blade or coin was fixed to a scrap piece of wood alongside a safety pin bent upright. A sharpened graphite pencil was firmly attached to the point of the safety pin… allowing fine readjustment of the pencil graphite across the metal oxide surface until a working rectifying junction was found.”
What is the likelihood that civilizational collapse would directly lead to human extinction (within decades)?
Epistemic transparency: Confidence in conclusions varies throughout. I give rough indicators of my confidence at the section level by indicating the amount of time I spent researching/thinking about each particular subtopic, plus a qualitative description of the types of sources I rely on. In general, I consider it a first step toward understanding this threat from civilizational collapse — not a final or decisive one.
Acknowledgements
This research was funded by the Forethought Foundation. It was written by Luisa Rodriguez under the supervision of Arden Koehler and Lewis Dartnell. Thanks to Arden Koehler, Max Daniel, Michael Aird, Matthew van der Merwe, Rob Wiblin, Howie Lempel, and Kit Harris who provided valuable comments. Thanks also to William MacAskill for providing guidance and feedback on the larger project.
Summary
In this post, I explore the probability that if various kinds of catastrophe caused civilizational collapse, this collapse would fairly directly lead to human extinction. I don’t assess the probability of those catastrophes occurring in the first place, the probability they’d lead to indefinite technological stagnation, or the probability that they’d lead to non-extinction existential catastrophes (e.g., unrecoverable dystopias). I hope to address the latter two outcomes in separate posts (forthcoming).
My analysis is organized into case studies: I take three possible catastrophes, defined in terms of the direct damage they would cause, and assess the probability that they would lead to extinction within a generation. There is a lot more someone could do to systematically assess the probability that a catastrophe of some kind would lead to human extinction, and what I’ve written up is certainly not conclusive. But I hope my discussion here can serve as a starting point as well as lay out some of the main considerations and preliminary results.
Note: Throughout this document, I’ll use the following language to express my best guess at the likelihood of the outcomes discussed:
Case 1: I think it’s exceedingly unlikely that humanity would go extinct (within ~a generation) as a direct result of a catastrophe that causes the deaths of 50% of the world’s population, but causes no major infrastructure damage (e.g. damaged roads, destroyed bridges, collapsed buildings, damaged power lines, etc.) or extreme changes in the climate (e.g. cooling). The main reasons for this are:
Although civilization’s critical infrastructure systems (e.g. food, water, power) might collapse, I expect that several billions of people would survive without critical systems (e.g. industrial food, water, and energy systems) by relying on goods already in grocery stores, food stocks, and fresh water sources.
After a period of hoarding and violent conflict over those supplies and other resources, I expect those basic goods would keep a smaller number of remaining survivors alive for somewhere between a year and a decade (which I call the grace period, following Lewis Dartnell’s The Knowledge).
After those supplies ran out, I expect several tens of millions of people to survive indefinitely by hunting, gathering, and practicing subsistence agriculture (having learned during the grace period any necessary skills they didn’t possess already).
Case 2: I think it’s very unlikely that humanity would go extinct as a direct result of a catastrophe that caused the deaths of 90% of the world’s population (leaving 800 million survivors), major infrastructure damage, and severe climate change (e.g. nuclear winter/asteroid impact).
While I expect that millions would starve to death in the wake of something like a global nuclear winter, my best guess is that hundreds of thousands to hundreds of millions of the survivors of the initial catastrophe would be able to subsist using leftover food stocks and supplies, before eventually working out how to feed themselves through traditional agriculture (in regions where this was still possible) and fishing and/or modified agriculture (using methods that don’t rely on climate factors like warm temperatures and regular precipitation).
For catastrophes that cause temporary climate effects, survivors would only need to subsist using non-traditional agriculture for a decade at most. After that, survivors could revert to traditional agriculture, which would be much easier.
Survivors might also face threats from radioactive fallout from nuclear detonations and power plant meltdowns. While this would increase the death toll in the years following the collapse, I expect it wouldn’t be widespread enough to be immediately fatal to everyone, nor would it cause fertility rates or life expectancy to decrease enough to threaten extinction.
Case 3: I think it’s fairly unlikely that humanity would go extinct as a direct result of a catastrophe that caused the deaths of 99.99% of people (leaving 800 thousand survivors), extensive infrastructure damage, and temporary climate change (e.g. a more severe nuclear winter/asteroid impact, plus the use of biological weapons).
I expect hundreds of thousands of the survivors of the initial catastrophe would be able to subsist using leftover food stocks and supplies, before eventually working out how to feed themselves through traditional agriculture and fishing and/or modified agriculture (similar to the case of a nuclear war-only catastrophe).
A naive back of the envelope calculation (BOTEC) makes me think that surviving populations would either stay above the minimal viable population (the minimum number of individuals necessary to maintain the level of genetic diversity to repopulate Earth), or find each other (by crossing distances as large as England or Wales). My confidence in this conclusion is higher for scenarios where survivors have long-distance communication, long-distance transportation, or are confined to a smaller geographic region (e.g. the Southern Hemisphere).
My main uncertainties are around 1) whether I’m underestimating the frequency and severity of shocks the survivors might face in the post-collapse environment, and 2) whether surviving groups might deliberately avoid each other — perhaps because of fears about pathogen spread — allowing their populations to dip/stay below the minimum viable population. While it seems likely to me that natural shocks would reduce the number of people that would survive in the decades following the initial catastrophe — likely to be somewhere in the tens of thousands — I think it’s fairly unlikely that either of these would actually lead to extinction. But I’m unsure.
My best guess is that the turning point at which extinction goes from under 50% to over 50% is between 99.999% population death (80,000) and 99.9999% (8,000) population death (even before considering additional starting conditions like infrastructure damage or climate change). When the surviving population is in the tens of thousands and lower, it starts to seem more likely that the obstacles the survivors would face — difficulties growing food, random natural shocks — would be enough to wipe out the remaining groups. The probability of human extinction at these levels of population death would of course be even higher if the initial catastrophe caused additional environmental obstacles (like climate change, radiation).
General arguments and considerations that kept coming up and had a big impact on my conclusions:
Historical survival and resilience. In fact, we know that humans have survived for thousands of years “without technology,” and in a very large variety of climates and environments.
The grace period. The period after the initial catastrophe — during which survivors would be able to access leftover supplies, goods, and some infrastructure — would be enormously helpful in ensuring humanity’s survival before the survivors have learned basic survival skills.
With population loss comes “decorrelation” of survivors. A population reduced to hundreds of thousands of survivors would likely be spread out in relatively small bands or groups of hundreds or thousands of people. Given this kind of distribution of survivors, many, many failures would need to happen — at roughly the same time, and all over the world — for humanity to go extinct. But the types of failures the survivors are likely to face would only be weakly or not at all “correlated” between groups of survivors or across time (e.g. multiple generations). In other words, even if some failure mode is very likely per group and time period (a drought may be the nail in the coffin for groups of survivors in one area), this still doesn’t suffice for human extinction, as many other groups would be unaffected.
Historically, the fact that different groups of humans were relatively independent (spread out, affected by different social and environmental pressures) seems to have made humanity relatively robust. We can point to lots of societies that collapsed and didn’t recover, but humanity as a whole has been very resilient — in part because it’s been made up of lots of heterogeneous groups spread out all over the planet.
Non-uniformity of the initial catastrophe’s impacts. One of the ways that different survivor groups will become decorrelated stems from the fact that groups are likely to end up in regions that have been differentially affected by the initial catastrophe. For example, some regions might be completely destroyed, while others will have more moderate infrastructure damage.
In fact, it’s extremely difficult to come up with a scenario in which a catastrophe would have uniform effects all over the world — where the entire world has the same degree of infrastructure damage, the same amount of goods and supplies left around, the same severity of any changes to the climate, the same rates of pathogen spread and mortality. So while it’s easy to imagine that a catastrophe that caused civilizational collapse might produce unlivable conditions in some regions, other regions wouldn’t be as badly affected (another example: a nuclear winter might make the Northern Hemisphere unsuitable for agriculture, but not all of the Southern Hemisphere). That geographic variation means that even if we can imagine a scenario that looks unsurvivable for some regions, people who find themselves in (or make their way to) less affected areas should have a much better chance of survival.
The population loss would have to be incredibly extreme to lead to extinction. The minimum viable population (100–1,000 people, assuming a sufficient balance of the sexes) is really, really small compared to the current population: Just 0.00001% of the people alive today would need to survive for repopulation to be possible. In a sense, there’s a really “high bar” for extinction. To get there from a catastrophe that caused 90%, 99%, or even 99.9% population death seems to actually require one to hold exceptionally pessimistic views about how survivors would fare in the event of civilizational collapse. Consider the following toy calculation:
Note: In the following calculation, I assume each group’s fate is independent of the fates of other groups. See footnotes and discussion in the body of the document for comments on the strength of this assumption.[1] The higher the true correlation between survivor groups, the more my toy calculations will cause me to underestimate the probability that all of the survivor groups would be wiped out.
With 8 million survivors, even extremely pessimistic views of the likelihood that any one group gets wiped out don’t necessarily lead to high probabilities for extinction in most scenarios. For example:
Even if you thought any one group of 100 or 1,000 survivors had a 99% chance of being wiped out (by failing to feed itself, to find water, to survive natural shocks), it would still be virtually guaranteed that at least one group of survivors would survive.
If you thought there was a 99% chance that each one of 800 groups of 10,000 people would be wiped out, there would still only be a 1 in 3,000 chance of extinction.
The probability of extinction would be higher (45%) if you believed that larger groups of 100,000 would also have a 99% chance of being wiped out. But to hold that view, you’d have to think that out of a group of 100,000 people, there’s a 99% chance basically none of them would overcome the obstacles of the post-collapse environment. I grant that there would be many obstacles: Survivors would have to work out how to grow their own food (potentially in the face of extreme global cooling), how to find and purify water, how to build shelter, potentially how to fish. They would have to survive disease outbreaks, hurricanes, tsunamis, conflict with other desperate survivors, possibly radiation contamination from nuclear war or the meltdown of nuclear reactors. But even so, my view is that the probability that a given population of 100,000 people would be completely wiped out — even facing all of those obstacles — wouldn’t be as high as 99%.
Similarly, the probability of extinction is very high (92%) if you think that any given group of 1 million survivors has a 99% chance of being wiped out. But again, to believe extinction risk was that high, you’d have to think that there would be a 99% chance that basically none of those 1 million people would survive and reproduce. I don’t find that plausible.
What is the likelihood that civilizational collapse would cause human extinction?
Introduction
Even if a catastrophe were not so destructive as to cause immediate extinction, it could be destructive enough to cause civilizational collapse. By this I mean something like the total failure of the world’s systems of governance and critical resource systems (e.g., industrial food, water, and power systems). Some have argued that these conditions could drive humanity to extinction by making it impossible for the survivors to maintain numbers above the minimum viable population — especially if the catastrophe had other longer-lasting effects (e.g., severe warming or cooling). Below, I discuss some of the approaches I’ve taken to estimating the likelihood that civilizational collapse would cause human populations to fall below the minimum viable population.
Importantly, different catastrophes could cause civilizational collapse, and would leave survivors in dramatically different conditions. The dimensions along which these conditions vary that I see as most important are: levels of population loss, infrastructure damage (e.g. damaged roads, destroyed bridges, collapsed buildings, damaged power lines, etc.), the severity and persistence of any climate impacts, and the severity and persistence of worldwide conflict. In what follows, I’ll use three case studies to discuss how I think each of these factors would affect the likelihood that civilizational collapse would cause human extinction.
Before diving in, a few important notes:
First, when I say “critical systems” in this piece I mean industrial agriculture (but not non-industrial “traditional” agriculture), water transport and purification, power production and distribution, transportation, telecommunications, and effective governance, unless otherwise noted. And when I discuss the systems having “collapsed,” I mean they are producing little to none of their usual outputs. A civilizational collapse, as I will use the term, is an event in which approximately all of the entire world’s critical systems are collapsed simultaneously for over six months.
Second, this piece focuses on whether civilizational collapse would lead more or less directly to human extinction without additional catalyzing conditions, like a second catastrophe (either soon or long after the first) or centuries of economic stagnation that eventually end in human extinction. I hope to explore both of these other scenarios in future posts.
Third, even if the population dropped below the minimum viable population, human extinction would not be inevitable. Rather, the likelihood of recessive genes that lead to lower fitness would go up, but these wouldn’t necessarily lead to extinction (though they would increase the risk of it). To illustrate, consider that the 10 million Ashkenazi Jews living today descended from a population of just 350 individuals who lived between 600 and 800 years ago (Commun, 2014). While Ashkenazi Jews suffer from higher rates of breast cancer, ovarian cancer, congenital deafness, cystic fibrosis, gaucher disease, Tay-Sachs, Parkison’s, and other diseases, their population grew rapidly over the last 500 years nonetheless. Given this, my bottom-line extinction probability estimates are therefore an upper bound on the likelihood of the probability of extinction.
Case 1: 50% population loss, no infrastructure damage, no climate change (e.g. a limited pandemic)
I’ll start by discussing what might happen in the wake of a relatively moderate catastrophe — one that causes the deaths of about half the population (4 billion survivors), but which doesn’t arise from global conflict, and which doesn’t cause any major infrastructure damage or climate change.
What is the likelihood that civilization would collapse under these conditions?
Before getting into the likelihood that society would recover from civilizational collapse under these starting conditions, I’ll briefly discuss whether we should expect human civilization to actually collapse in my sense in this scenario.
Expert judgment: One expert told me that some resilience experts hypothesize that the economy would collapse (i.e. currency might lose its value, international trade might cease, etc.) in a case where 40% or more people had died or were not working[2] (though I haven’t been able to find a written source corroborating this). Even if that’s true, it’s not obvious that civilization would completely collapse if the world economy did. But it certainly seems possible: The goods and services required for most industries to run are not produced locally, but globally — they are accessed through international trade that presumably wouldn’t exist if the global economy collapsed.
There are people who think civilizational collapse is likely even at (or below) 50% population loss. (This includes Lewis Dartnell, author of The Knowledge, and several of the academics in the PIIRS Global Systemic Risk Research Community at Princeton University.) They often appeal to the fragile and interconnected nature of the modern globalized economy: Many critical systems have features that make them more efficient in the short run but less robust to large shocks (examples in footnote).[3] People who are pessimistic about civilizational robustness often appeal to this kind of fragility.
I’m more optimistic. Despite our increasingly interconnected systems, I believe that even in a catastrophe of this scale, our systems would be resilient enough to stay functioning — at least in some places. At current population levels, this catastrophe would still leave four billion survivors. I find it hard to imagine that four billion people working together wouldn’t be sufficiently adaptive, creative, and knowledgeable enough to keep critical systems running in some places, especially if certain regions are less affected than others. Without infrastructure damage, climate change, or major ongoing conflict, I strongly suspect that at least some regions would keep their critical systems online.
Indeed, while a few disagreed, most experts I spoke with shared the view that, while it seemed plausible that critical systems might go down temporarily at the regional level, it seemed exceedingly unlikely that all critical systems would fail all over the world in Case 1.
I explore the possibility of civilizational collapse from this sort of catastrophe in any case, because 1) I could be wrong, and 2) I think it’s a useful baseline with which to compare more severe catastrophes.
In a scenario in which a catastrophe of this scale does cause civilizational collapse, whether humanity goes extinct as a direct(ish) result of that collapse would depend on whether humanity could survive without critical systems, and for how long, and how quickly humans would be able to bring critical systems back online (if at all). I discuss this next.
What is the likelihood that humanity would survive without key systems (food, water, power) in the years immediately after a catastrophe?
Time spent on this section: 5-10 hours
Types of sources: Popular non-fiction books, op-eds in well-regarded newspapers, and expert interviews
If the survivors of the catastrophes aren’t able to get critical systems back online very quickly after the initial catastrophe, would they be able to survive? I took a mix of approaches to answering this question:
Historical track record: The genus homo survived for thousands of years without complex support systems, practicing hunting and gathering for hundreds of thousands of years, and Homo sapiens practiced settled agriculture for over ten thousand years. They did this in the face of changing climates and predators, with only rudimentary tools. This is at least some evidence that the survivors of a catastrophe could do this again — plausibly for thousands of years (remembering that the scenario we’re considering assumes the catastrophe didn’t cause any meaningful changes to the climate or natural environment).
On the other hand, most humans living today wouldn’t know how to survive without the support of our modern food and water systems. This might appear to suggest that the historical track record doesn’t tell us as much about the likelihood of survival among humans who face the collapse of civilization today.
Case study: To illustrate how humans might be unable to survive without critical systems, we can take a look at a few case studies. There are many examples of humans failing to survive in new environments, and/or after losing access to support systems on which they were reliant. One set of cases I looked into were those of stranded or lost European explorers who died in large numbers when forced to try to survive in the wilderness.
In the Secret to Our Success, Joseph Henrich describes several cases where groups of explorers — although physically fit, highly trained, and equipped with extensive supplies — died of disease and starvation because they couldn’t exercise basic survival skills in the environments in which they found themselves. At the same time, local populations were thriving in the environments in which the explorers perished, because of their large body of highly specific, non-obvious, detailed cultural knowledge. For example, the Franklin Expedition, a group of over 100 explorers who became stranded in the Arctic, slowly died because they could not practice the hunting techniques practiced by the local Inuit for millenia. Even lost expeditions in much more forgiving and food-rich environments — the Burke and Wills expedition in Australia and Narvaez Expedition in Florida, for example — saw entire expeditions of explorers perish due to ignorance of their environment.[4]
Disanalogies from those case studies: All that said, many humans likely would know how to survive without modern food and water systems. For example, I expect that most subsistence farmers — of which there are probably between 1 and 2 billion[5] — would at the very least have the knowledge and skills to survive by growing their own food. While flow-through effects of the catastrophe might mean that some proportion of these subsistence farmers wouldn’t be able to access key inputs for subsistence agriculture (or would be displaced from where they know how to survive), I expect at least 10% of them, at a minimum, would — leaving very likely more than 50 million people who would likely have the knowledge, skills, and resources to survive even without global infrastructure and centralized systems.
In addition to these people, there are apparently as many as 3.7 million Americans who identify as “preppers”. Given that most people who identify as preppers aren’t actually prepared to survive for years without critical systems (they mainy stockpile canned food and even build bunkers without necessarily learning how to acquire and prepare food and water in the natural environment), it seems plausible that as few as 1% of preppers (37,000 individuals) would survive. But even then, that 1% alone would keep us well above the minimum viable population of on the order of 100-1,000 survivors (assuming an even enough sex mix).
Even for people who don’t already know key survival skills — how to hunt, gather, and/or grow food — there would be opportunities to learn them. For example, there would likely be information sharing between those who did and those who didn’t, plus a great deal of written information on the topic in libraries and in homes. And this information would cover all of the world’s natural environments — so post-collapse humans wouldn’t be without key knowledge in new environments the way the European explorers were. To demonstrate, I searched the SAS Survival Handbook, Amazon’s best-selling wilderness survival guide, for the key pieces of information Henrich notes would have been critical for the explorers’ survival in the stories above. I found almost all of the relevant bits of information and techniques easily (and for bits that were missing, usually because they were incredibly specific, I found close substitutes).
Certainly, these survival skills would be hard to learn. Some survival skills, like starting a fire with only things found in the natural environment, are notoriously difficult to master. But many survivors wouldn’t have to learn the hardest skills: lighting fires with only sticks or flint, building shelters with only natural materials, making receptacles to hold water from natural resources alone, processing dead animals (making clothes from tanned hides, making fuel from their fat, carving bones into knives, needles, or buttons). They’d have access to knives, lighters, matches, tarps and plastic bags (for shelter) wool socks, plastic water bottles (to purify and carry water), salt (to preserve foods), alcohol (to treat wounds), thread/rope, and cooking oil (to burn). Some survivors would find iodine tablets, fishing rods, and compasses. Indeed, most survival skills look much less difficult with modern technologies.
The grace period: What’s more, at least some of these survivors would have time to learn survival skills during a period in the immediate aftermath of the catastrophe when systems have failed, but there is still a considerable stock of goods available for immediate consumption: food, water, batteries, and vitamin supplements in grocery stores; clothes in malls; cars, planes, ships, and computers that still work; gasoline at the pumps that could be siphoned out even if the power grid was down; medications in pharmacies; generators; books in libraries and in houses, etc. During this period, I expect that at least some survivors (maybe 5–25%) without experience practicing subsistence farming, hunting, and gathering would be able to meet their basic needs while learning those skills.
Without this grace period, I think it’s still fairly likely that hundreds of thousands or millions of people (largely subsistence farmers) would be able to survive the collapse, but I’m less confident. So I decided it was worth doing a BOTEC to understand the likelihood that there would be such a grace period, and how that period would last.
Note: The following BOTEC relies on particularly poor sources, makes a bunch of dubious assumptions (discussed more below), and I’m not confident I’ve thought of all of the most important supplies. It should be considered very rough.
BOTEC: To estimate whether and how long this kind of grace period might last given different population sizes, I investigated the amount of various goods/supplies that would be available in the post-collapse environment. In addition to goods that would facilitate human survival, I also considered goods that might be essential (or at least very helpful) in making use of the goods that are essential to human survival (e.g., humans don’t need petrol to survive, but that might facilitate movement, which might be essential to accessing a distributed food supply).
Note some important assumptions made here, that would meaningfully affect how realistic these estimates are:
I’m assuming food stocks wouldn’t perish before they were eaten (or that food stocks would be deliberately consumed in order of perishability, to ensure stocks last as long as possible). If this assumption doesn’t hold, the grace period would be shorter, perhaps by a lot.
I’m ignoring animal agriculture and cannibalism, in part because without a functioning agriculture system, it’s not clear to me whether enough people would be able to consume living beings. If this assumption doesn’t hold, the grace period could be a fair bit longer.
I didn’t look into whether the food stocks include stocks intended for animal agriculture. If they don’t, it’s possible I’m missing out on a huge amount of grain stock that could be consumed by humans — lengthening the grace period considerably.[6]
See table notes here.[7][8][9][10][11][12][13]
I find that the goods and supplies left after an initial catastrophe would only support a population in the billions for several months to a year — not very long at all. Given this, I expect a large fraction of the survivors who survive this initial catastrophe to die during a period of violent competition following the collapse.
I expect this period of violence to last until the survivors reach some sort of equilibrium, where they perceive there to be sufficient food, water, and shelter. The reason for this view is that it seems very irrational for groups to keep fighting if there are enough resources to go around. Of course, humans aren’t always rational, so I imagine some fighting might persist. But I think it’s very likely that many survivors would correctly conclude it would be more dangerous to fight each other than to accept a relatively low risk of not being able to meet their basic needs because of resource scarcity, if resources weren’t in fact too scarce to support the population.
Carrying capacity for hunter-gatherers and subsistence farmers: Importantly, without the technologies that enable humans to produce enormous amounts of food using a fraction of the land and water as our ancestors, the Earth would only be able to sustain a fraction of the human population it does now. Just how large a population it could support depends on the means of subsistence employed by the survivors. The carrying capacity of the Earth for hunter-gatherers is thought to be around 10 million if the survivors regress to pre-paleolithic levels of technology (if they lose, for example, flakes, handaxes, controlled use of fire, and wooden spears) (Taiz, 2013). The carrying capacity is hypothesized to be in the low billions for settled subsistence farmers — possibly as high as three billion for mostly vegetarian populations with particularly advanced agricultural systems (Ellis, 2013).
Bottom line: Taken together, if a Case 1 scenario leads to collapse (but not to additional catastrophes), I expect that millions to hundreds of millions — possibly even a few billion — of the 4 billion survivors would be able to learn enough survival skills to survive indefinitely even after the grace period. My biggest uncertainty is whether I’m underestimating the difficulty of surviving in a post-collapse world with only basic knowledge of subsistence farming and/or survival skills.
What is the likelihood that humanity would survive in the event of conflict immediately following the catastrophe?
Time spent on this section: 1–2 hours
Types of sources: Academic literature, expert interviews, and speculation
Literature review: Sociologists trying to understand human behavior in the aftermath of shocks like natural and manmade disasters largely agree that violence is not the most common reaction to emergencies (Drabek and McEntire, 2003). Rather than looting, exploitation, and other antisocial behavior, academics have found that groups tend to cooperate in the aftermath of disasters.
But a catastrophe of this magnitude would be unprecedented, and meaningfully different from the disasters that researchers have been able to observe. Unlike smaller scale disasters, where even those most impacted by the disaster could reasonably expect that help was on the way, post-collapse survivors would have no reason to expect this. As I concluded above, the leftover goods and supplies wouldn’t sustain all of the initial 4 billion survivors for very long. Without the hope of support — and facing the threat of starvation and other dangers — I strongly suspect that the limited availability of leftover supplies would almost certainly cause violent conflict at both the group and individual level. Specifically, I expect that subsets of the survivors — likely groups with access to weapons — would use violence (or the threat thereof) to hoard the surviving supplies, killing their competitors or letting them starve to death.
But as discussed, I expect any violence would mostly subside once some equilibrium had been reached — at the point where ~all of the surviving population could reasonably expect to be supported by the available resources (in other words, at the point at which the population dropped to or below whatever the Earth’s carrying capacity was). It’s at that point that regular conflict and violence would actually make the environment more dangerous than coexisting non-violently and sharing the available resources. And while resources would be somewhat more depleted by an initially large group of survivors all consuming them — and by the use of some resources directly in inter-group violence — therefore shortening the grace period, I don’t expect this would change my bottom line (but I’m uncertain about this).
My best guess is that this violent competition would whittle the number of survivors down to the hundreds of millions or fewer — a population for whom the supplies might last up to a decade or more. This is based on my best guess about the population size that I expect could be robustly supported by the supplies left during the grace period and the approximate hunter-gatherer carrying capacity of the Earth.
Bottom line: I don’t expect violence to drive the surviving population so low that humanity would go extinct. Rather, I think it’s extremely likely that violence would mostly disappear once the population had gotten small enough that all of the remaining survivors could expect to survive indefinitely on the available resources — probably hundreds of millions or fewer. My key uncertainty is whether initially large numbers of survivors, many fighting each other, would shorten the grace period enough that survivors wouldn’t have enough time to learn survival skills they’d need to survive after the grace period.
Concrete example: A moderate pandemic
So what, concretely, do I think would happen in the event of a catastrophe like a “moderate” pandemic — one that killed 50% of people, but didn’t cause infrastructure damage or climate change?
My best guess is that civilization wouldn’t actually collapse everywhere. But if it did, I expect that the ~4 billion survivors would shrink to a group of 10–100 million survivors during a period of violent competition for surviving goods in grocery stores/distribution centers, food stocks, and fresh water sources. I expect these supplies would last somewhere between a year and a decade or more — on the shorter side if the period of violence were prolonged, causing a large proportion of the supplies to be used up by larger populations before the population stabilizes at a lower number. After those supplies ran out, I expect that up to 10 million of those survivors would survive if practicing hunting and gathering, or as many as hundreds of millions if practicing advanced subsistence agriculture (after learning any skills they didn’t possess already during the grace period).
It would be reasonable to be skeptical of any or even several of the individual claims I’ve made here. But a toy calculation gives at least a rough indication of how wrong they would have to be for my bottom line to change.
Note: I assume the likelihood that any one group gets wiped out is independent from the probability that another group does. I think this is somewhat reasonable, given that there are many things that affect individual groups or subsets of groups without affecting all groups. For example, a contaminated water source used by a few groups might cause disease rates in that group to go up significantly, perhaps even causing the groups to be wiped out, but groups far away wouldn’t be affected. The higher the true correlation between survivor groups, the more my toy calculations will cause me to underestimate the probability that all of the survivor groups would be wiped out.
With 4 billion survivors, the degree of pessimism you have to have about their ability to survive to end up believing that no groups would survive indefinitely is actually very extreme. The exact beliefs you’d have to have would depend on whether survivors were concentrated into a few big groups, or distributed across many smaller ones. Specifically:
Even if you thought any given group of between 10 survivors and 1 million survivors had a 99% chance of being wiped out, it would still be virtually guaranteed that at least one group would survive.
If you thought every single one of 400 groups of 10 million people had a 99% chance of being wiped out, you would still only think there was a 1.7% chance of extinction. And, to hold that belief, you’d have to think that despite having extremely large populations (larger than any US city), none of those people would be able to work out how to practice very basic farming or fishing, or that none of them would survive conflict within the group, or that a natural disaster would kill all of them. Personally, I find this view implausible.
Similarly, the probability of extinction is pretty high if you think that any of 400 groups of 100 million survivors has a 99% chance of being wiped out. Again, though, to believe extinction risk was that high, you’d have to think that there’s a 99% chance that none of the 100 million people would overcome obstacles posed by the absence of critical systems and violence. (For reference, only 14 countries have a population of 100 million or higher.)
Given this, I believe it’s exceedingly unlikely that this scenario would lead to human extinction.
Case 2: 90% population loss, infrastructure damage, and extreme climate change (e.g. nuclear war that caused nuclear winter)
In a scenario in which a catastrophe causes the deaths of 90% of the population (800 million survivors), major infrastructure damage, and climate change — for example, a severe, global nuclear war that caused a nuclear winter — I believe the question of whether humans would be able to meet their basic needs becomes more difficult.[14] The questions I consider for this scenario are:
What is the likelihood that survivors are able to continue to survive using traditional forms of agriculture, given a catastrophe that causes severe infrastructure damage and climate change?
What is the likelihood that radiation causes extinction?
What is the likelihood that humanity would survive in the event of conflict immediately following the catastrophe?
What is the likelihood that survivors are able to continue to survive using traditional forms of agriculture?
Time spent on this section: 2–3 hours
Types of sources: Academic literature, non-academic reports, and expert interviews
Expert judgment: Several experts, including ALLFED director David Denkenberger, have affirmed this conclusion — they do not expect humanity to dip below the minimum viable population even in relatively extreme sun-blocking scenarios.
Literature review: The nature of all of the catastrophes we know of that would cause extreme global cooling (e.g. nuclear winter, asteroid impacts) would have unevenly distributed impacts — causing extreme global cooling in some parts of the world, but more moderate cooling in others. For example, in the case of a nuclear war between the US and Russia, nuclear winter models suggest that the most severe climate effects would be limited to the Northern Hemisphere, where temperatures would fall by 10–30 degrees C. But in the Southern Hemisphere, and especially at the equator, those effects would be much less severe: between 5–10 degrees Celsius. With heterogeneous impacts like this, it’s likely that agriculture would still be possible in some regions — especially in New Zealand and Australia, and possibly in South America and Central Africa.[15]
To be clear, I’m describing a very grim scenario, in which basically everyone in the Northern Hemisphere — and in many parts of the Southern Hemisphere — would be unable to grow food using standard agricultural techniques. Given this, I expect there would be mass starvation and violent competition and conflict until a new equilibrium was reached, one where the remaining survivors didn’t exceed the Earth’s carrying capacity. While I expect this would be a truly terrible period of widespread suffering, I believe this equilibrium would be reached long before the population got anywhere near the minimum viable population. My best guess is the population would fall to hundreds of thousands to tens of millions, but not much lower.
While I haven’t looked into this much, I feel fairly convinced that hundreds of thousands or millions of people could survive using traditional approaches to agriculture in parts of the world with more moderate climate effects (and basic mitigation strategies, like switching to crop types that are more resilient to temperature and precipitation fluctuations). And as with Case 1, at least some of the survivors in a Case 2 scenario would probably be able to survive the immediate aftermath of a catastrophe that caused civilizational collapse by exploiting food and other supplies in stores and larger stockpiles. This would give survivors some buffer time to learn additional skills required to survive once those supplies run out (e.g. fishing) or develop the techniques necessary to produce food using methods that don’t rely on climate factors like warm temperatures and regular precipitation.
BOTEC: The longer the buffer time, the more likely humanity would be to subsequently survive. But there are a number of different considerations (relative to Case 1) that affect the calculus of just how long such a grace period would be in the context of a catastrophic event like a nuclear war that killed 90% of people and caused a nuclear winter. So I’ve done a similar exercise to the one above where I try to account for some of those differences.
Note: As above, the following BOTEC relies on particularly poor sources, makes a bunch of dubious assumptions (discussed more below), and I’m not confident I’ve thought of all of the most important supplies. It should be considered very rough.
See table note here.[16]
Bottom line: I think it’s extremely likely that these supplies would last somewhere between around a year and a decade or more. I expect it would be closer to the lower end, given that competition and violence could lead to the depletion of supplies more quickly than if the population were reduced to a smaller number by the catastrophe directly.
All this in mind, I think it is very likely that the survivors would be able to learn enough during the grace period to be able to feed and shelter themselves ~indefinitely.
What is the likelihood that radiation causes extinction?
Time spent on this section: 2–3 hours
Types of sources: Academic papers, Wikipedia, and interviews with experts
Literature review: In the aftermath of a nuclear war, radioactive fallout from the nuclear detonations would have long-lasting health impacts. In the most extreme nuclear war scenarios considered by academics (a nuclear war between the US and Russia and their allies, using 10,000 megatons (MT) of nuclear bombs), approximately 30% of the geographic area in the Northern Hemisphere would have enough fallout to be lethal to any adult in the area (Ehrlich et al., 1983). The current US and Russian nuclear arsenals don’t currently have that kind of megatonnage (they currently have closer to 2,500 MT). If we naively assume that radiation scales linearly, we might expect a modern day US-Russia nuclear war to contaminate up to 7.5% of the land area of the Northern Hemisphere. This may not sound like much, but consider that 95% of the world’s population lives on just 10% of its land area — meaning that 7.5% of land area could be home to millions or even billions of people. What’s more, tens to hundreds of millions more might be exposed to enough radiation to be more susceptible to cancer for the rest of their lives.
On top of this, there are currently around 440 civilian nuclear power reactors scattered around the world, and likely tens or hundreds more military reactors. These have fail-safes and automatic shut down measures that are designed to ensure that all of the nuclear material in these reactors would be safely contained in the event of a global catastrophe that meant people stopped attending to them. Concretely, these fail safes make sure that water continues to be circulated around the nuclear fuel to ensure it doesn’t get so hot it causes a meltdown — i.e., an event where the nuclear core partially or completely melts, which might allow the nuclear fuel to breach its multiple layers of containment and leak out into the environment. If fuel did reach the environment, the radioactive fallout could spread across continents, creating exposure levels ranging from immediately fatal (in areas ranging from tens to thousands of square kilometers) to non-lethal but causing potential higher rates of cancer and infertility.
But some of these fail-safes could plausibly fail during a catastrophe that caused infrastructure damage (or afterward, if any components of the fail system degraded). For example, some nuclear reactors rely on backup generators to power the pumps that keep water circulating in the core of the reactor. If those backup generators eventually all broke down, the reactor might melt down.
I currently don’t have a good sense of how likely these failures would be. Newer nuclear reactors rely on more robust safety systems, with parts that wouldn’t break down as easily. And all nuclear reactor safety systems are designed to account for infrastructure damage caused by earthquakes and other physical shocks.
But in a large-scale nuclear war, it seems very plausible that at least some nuclear reactors would melt down. My best guess is that this wouldn’t happen at a large scale, but even if it did, some areas would likely be far enough away from reactors to be spared the radioactive contamination. For example, Australia has just one nuclear reactor. Even if that reactor were to melt down, much of Australia would likely remain uncontaminated (Australia is just under 3 million square miles, and the Chernobyl meltdown is estimated to have contaminated under 60,000 square miles; and only a much smaller fraction of that area was sufficiently contaminated as to be lethal to humans).
Bottom line: While radioactive fallout from nuclear detonations and power plant meltdowns would increase the death toll in the years following the collapse, I expect it wouldn’t be widespread enough to be immediately fatal to everyone, nor would it cause fertility rates or life expectancy to decrease enough to threaten extinction. And at the very least, some areas are sufficiently far away as to be relatively safe from radioactive fallout.
What is the likelihood that humanity would survive in the event of conflict immediately following the catastrophe?
Time spent on this section: 1–2 hours
Types of sources: Academic literature, expert interviews, and speculation
Historical base rate: In Case 2, it seems slightly more plausible to me that violence would lead to human extinction than in Case 1, but still fairly unlikely. I don’t think human extinction could be caused by a conflict fought with conventional weapons; there would just be too many survivors (~800 million) to be killed in conventional warfare (compare this to WWI and WWII, during which ~20 million and ~75 million people were killed, respectively).
Weapons of mass destruction: My best guess is that the only way violence in the wake of a Case 2 civilizational collapse could directly lead to human extinction is if one group of survivors had access to and deployed weapons of mass destruction. This seems unlikely to me, first because it seems hard to imagine a group of survivors incapable of recovering critical infrastructure — and barely capable of meeting even their basic needs — would be able to successfully deploy weapons of mass destruction (though I’m not very confident about this).
Second, it’s hard to imagine a scenario where the use of weapons of mass destruction kills millions of survivors, spread all over the world, without modern technologies like transportation. For example, with potentially many survivor groups, it seems hard to imagine how nuclear detonations would kill ~everyone despite the fact that the groups would likely be spread out all over the world, potentially in small bands that can’t each be individually targeted. Similarly, it’s hard to imagine how a pathogen could spread ~everywhere when survivors would likely have greatly reduced mobility (the latter isn’t obviously impossible, but it at least seems exceedingly difficult to me).
There’s one counterargument I find somewhat persuasive, which is that it seems possible that all of the survivors might be confined to a relatively small area (for example, if only a small fraction of the Earth’s land area is habitable), making them more vulnerable to a single, large attack. If this were the case, it’s easier for me to imagine that the use of weapons of mass destruction could kill all of the remaining survivors. This would presumably mean the aggressors would be killing themselves, which makes it seem even less likely to me. But we’ve seen humans come dangerously close to threatening their own survival before, often because human aggressors aren’t always good at predicting how cascading effects could threaten their survival as well. A random example to make this concrete: If all of the survivors of a nuclear war were confined to Australia, which might be less impacted by a nuclear winter, one group might choose to use nuclear weapons against another group, not realizing that the radioactive fallout or further climate change could make Australia uninhabitable, even for them.
Bottom line: I expect the survivors in Case 2 would not deploy weapons of mass destruction against their competitors, as it would likely pose a pretty big risk to the aggressor as well as the target. But I’m uncertain about this — humans have come close to making similarly self-destructive choices before. Thankfully, even if one group did use weapons of mass destruction against their competitors, I still think it’s very unlikely that their use would cause human extinction. This is because except in a few very specific and very strange scenarios, I expect the survivors would be too geographically distributed and disconnected to be wiped out by a single act of aggression. I therefore expect the result would be a much higher death toll, but not extinction.
Concrete example: A large nuclear war that causes a nuclear winter
So what, concretely, do I think would happen in the event of a catastrophe like a nuclear war that led to the death of 90% of the population, and caused severe infrastructure damage and significant global cooling?
I expect that, in addition to the billions of people killed in the initial catastrophe, hundreds of millions or more would likely die in the famines and violent competition that followed. But my best guess is that hundreds of thousands to hundreds of millions of the survivors of the initial catastrophe would survive this violent period. I think it’s extremely likely these survivors would be able to support themselves using leftover food stocks and supplies, before eventually working out how to feed themselves through traditional agriculture and fishing and/or modified agriculture (using methods that don’t rely on climate factors like warm temperatures and regular precipitation).
All of the catastrophes we know of that would lead to extreme cooling would only do so for 1–10 years, and agriculture would become possible again once the climate began to return to normal. At that point, it seems even more likely that the surviving humans would be able to meet their own basic needs by returning to traditional forms of agriculture.
My key uncertainties are around whether I’m putting too much weight on the idea that humans would figure out how to subsist without traditional agriculture just because it’s technically possible, and whether conflict could lead to extinction through channels I haven’t foreseen.
Another toy calculation suggests that these uncertainties probably aren’t troubling enough to change my bottom line.
Note: I again assume each group’s fate is independent of the fates of other groups. I actually think this is a pretty reasonable assumption in this case. I expect that the survivors of a catastrophe like a severe nuclear war would end up somewhat spread out (at least across the Southern Hemisphere), as doing so would create less competition for resources within a smaller area (I discuss this more later). The farther apart the surviving groups are, the less likely they are to be affected by the same shocks (natural disasters, disease outbreaks, conflict).
Additionally, in the event of a catastrophe like a nuclear war, transportation, communication, and other technologies that facilitate contact between geographically distributed groups would be enormously limited. This would further limit the extent to which each group’s fate ended up relating to another’s. There would be other sources of variation between groups that made their fates less correlated: Some groups might be made up mostly of farmers, while others will be made up of lawyers, some groups will tend toward cooperation, while others toward conflict, plus pure randomness (e.g. some groups might have a high proportion of survivors with genetic immunity to a particular disease).
But there are also factors that point in the other direction — factors that suggest the surviving groups would be at least somewhat correlated. For example, nuclear winter climate conditions, while nonuniform, would nonetheless impact all surviving groups. Similarly, more severe natural disasters might affect large regions, meaning that at least all of the survivor groups at the regional level might end up experiencing very similar challenges to survival simultaneously. Likewise, there might be things about “human nature” that would be shared amongst all survivors. For example, it’s possible that all of the survivors, having witnessed the initial catastrophe, would have similar psychological experiences — like shock, stress, and social distrust, among others — that would make it more difficult to survive and cooperate.
As above, the higher the true correlation between survivor groups, the more my toy calculations will cause me to underestimate the probability that all of the survivor groups would be wiped out.
With 800 million survivors, the degree of pessimism you have to have about their ability to survive to end up believing that no groups would survive indefinitely is actually kind of extreme. The exact beliefs you’d have to have would depend on whether survivors were concentrated into a few big groups, or distributed in many smaller ones. Specifically:
Even if you thought any given group of 100, 1,000, or 10,000 survivors had a 99% chance of being wiped out, it would still be virtually guaranteed that at least one group would survive.
If you thought there was a 99% chance that any one of 800 groups of 100,000 people would be wiped out, there would still only be a 1 in 3,000 chance of extinction.
The probability of extinction is higher (45%) if you believe that larger groups of 10 million would also have a 99% chance of being wiped out. But, again, to hold that view, you’d have to think that out of a group of 10 million people (again, bigger than the largest US city), not even a few hundred of those people would overcome the obstacles of the post-collapse environment (how to fish, how to farm despite global cooling, avoiding being killed by a hurricane or drought). I do not find this view very plausible.
Similarly, the probability of extinction is very high indeed if you think that any given group of 100 million survivors has a 99% chance of being wiped out. Again, to believe extinction risk was that high, you’d have to think that there would be a 99% chance that none of the 100 million people would work out how to survive (for reference, only 14 countries have a population of 100 million or higher).
Given all of this, my subjective judgment is that it’s very unlikely that this scenario would more or less directly lead to human extinction.
Case 3: 99.99% population loss, infrastructure damage, and climate change (e.g. a global war where biological weapons and nuclear weapons were used)
In a scenario in which a catastrophe caused the deaths of 99.99% of the population (800,000 survivors), in addition to major infrastructure damage and climate change (e.g. war using both nuclear and biological weapons), the survivors would face many of the same challenges faced in Case 2, such as food insecurity and the potential for ongoing conflict. Below, I’ll explore the additional obstacles faced by survivors in this scenario caused by the drastically smaller population, as well as the possibility of a persistent pathogen from a biological attack. Specifically, I’ve tried to understand:
Given that 99.99% population loss is obviously extremely significant, what is the likelihood that the survivors are so spread out that each individual group is below the minimum viable population size?
What is the likelihood that humanity would survive without key systems (food, water, power) in the years immediately after a catastrophe (during and after the grace period)?
What is the likelihood that a human population of 800,000 would survive natural shocks like natural disasters, climate fluctuations (drought), and disease?
My answers to each of these questions increases my credence in a Case 3 catastrophe leading to human extinction, but together they only increase it 10–20% over my credence in a Case 2 catastrophe leading to extinction.
What is the likelihood that the survivors are so spread out that each individual group is below the minimum viable population size?
Time spent on this section: 1 hour
Types of sources: Academic literature, popular nonfiction books, and expert interviews
With only 800,000 survivors, it seems much more plausible than in Case 1 and Case 2 scenarios that many of the surviving groups would be below the minimum viable population of 100–1,000 people. For example, it seems at least plausible that there could be 80,000 groups of 10 survivors — especially if there were pressures to stay in small groups or spread out geographically. Some people have argued that, having fallen below the minimum viable population, these groups would run the risk of dying out unless they were able and willing to find each other, forming larger (genetically viable) groups. So a critical question might be: how likely is it that groups of survivors would be able to find each other within a generation? This question breaks down further into:
Would they find each other by chance?
Could they coordinate to find each other if not?
Could they physically reach each other?
Would they want to find each other?
When thinking about whether 800,000 survivors in lots of small groups would stumble across each other by chance, I realized I didn’t have an intuitive sense about how far away groups of survivors would be from each other. To get a very rough sense, I did a simplistic BOTEC, estimating how much land area a group of survivors of a particular size would have to cross in order to find another group of a similar size, in expectation.
BOTEC: In the following toy calculation, I explored two scenarios: one where the survivors were distributed evenly across the world, and another where they were distributed evenly across the Southern Hemisphere (which might be the case in the event of a nuclear war in the Northern Hemisphere).
The BOTEC relies on the following simplifying assumptions:
The survivors are mobile and actively trying to find each other, rather than avoiding it. If I’m wrong about this, my estimates will overstate the likelihood of inter-group contact.
Survivors are evenly distributed across different regions. If I’m wrong about this, my estimates will probably overstate the likelihood of inter-group contact, but will probably understate the size of each group.
I don’t account for obstacles like mountain ranges, nor do I account for the fact that some groups would be separated by oceans. Rather, the following best reflects something like: the ease of reaching groups on the same continent with no major geographic barriers. This assumption means my estimates will overstate the likelihood of inter-group contact.
See table note here.[17]
There were a few things I took away while making this table. First, any groups with populations of 1,000 or above should have enough genetic diversity to repopulate. And for groups of survivors made up of 10 to 100 people, the approximate land area each would have to cross to happen upon another group of survivors is pretty small. If the survivors were in small bands of around 10 people, they’d only have to travel across an area the size of Oxfordshire (in expectation) to happen upon another band. If the survivors were in groups of 100, they’d have to travel across an area the size of the Bay Area.
If the survivors had no way of coordinating, it could plausibly take years for them to happen across each other. But there are reasons to think it would happen even more quickly than that. First, there are likely to be points of disproportionately high traffic — attractor points — like water sources, cities (where survivors will go to scavenge supplies), roads/railroad tracks, etc. Additionally, in the first 5–10 years after the collapse, I think it’s fairly likely that at least some survivors would find ways to retain some motorized transportation, drastically increasing the area that could be covered over time. Unless the initial catastrophe was something that wiped out electronics, many cars would still be drivable, and petrol and diesel could be accessed with relative ease (until they went off after sitting there for five and ten years, respectively). Given this, I think it’s very likely that more than a few groups of survivors would find each other within a generation even if they had no means of communication.
But even that I think is too pessimistic, as I think it’s extremely likely that survivors would be able to employ at least some forms of long-distance communication to contact and connect with other survivors. Even with no advanced technology, survivors could create beacons by lighting fires at high ground. And with some luck, survivors might work out how to communicate using radios — either rebooting existing ones or making their own.[18]
Given this, I actually think it’s extremely likely that survivors would form one or more groups that were well above the minimum viable population of 100–1,000 people (assuming a sufficient balance of the sexes) at this level of population loss — even if survivors were spread out across the whole world and limited to basic forms of transportation (foot, non-motorized wheeled vehicles).
Importantly, though, if these groups were trying to avoid each other — perhaps because of fears of conflict or disease — it becomes a bit more plausible to think that small groups of survivors might stay below the minimum viable population, potentially leading to human extinction. But while I think it’s actually pretty likely that there would be at least some inter-group avoidance and conflict for at least a period of months or years after the initial catastrophe, I think it’s extremely unlikely that this aversion would exist and persist among thousands of different groups of survivors for a full generation. This is because I’d expect the benefits of larger populations (specialization, division of labor, etc.) to eventually outweigh the risks (within a period of years or decades at most) — for example as a human-borne pathogen died out after being unable to spread, or after populations were small enough that resource competition — and by extension, violence — had decreased substantially.
Bottom line: Taking into account both the logistics of coordinating and connecting with other groups of survivors, as well as the potential for decades-long intergroup avoidance, I think it’s extremely unlikely that there wouldn’t be any groups that ended up above the minimum viable population in the event of a catastrophe that left 800,000 survivors.
What is the likelihood that humanity would survive without key systems (food, water, power) in the years after a catastrophe?
Time spent on this section: <1 hour on top of the time spent above
Types of sources: Popular non-fiction books, random unvetted websites, government reports, well-regarded news websites, and expert interviews
As above, I did a BOTEC to understand how long the 800,000 survivors would be able to rely on surviving goods and supplies to meet their basic needs. I found that even the scarcest supplies would last at least 50 years (in theory), with more abundant supplies like grain stocks lasting hundreds of years (again assuming that 55% of stocks are lost during the initial catastrophe and ignoring perishability). I don’t present the full results here, because the limiting factor on how long these resources would keep survivors alive isn’t the number of survivors, but rather the perishability of the supplies. Given this, I think it’s at least safe to conclude that the surviving supplies and goods would likely keep survivors alive for well over a decade, maybe more.
During that period, as above, I expect survivors would learn to subsist without critical systems by hunting, gathering, fishing, and practicing modified and traditional agriculture and animal husbandry.
Bottom line: I expect these supplies would last somewhere on the order of several years or decades, giving survivors plenty of time to learn the survival skills they’d need to meet their basic needs after they eventually ran out/went off. Given this, I think it’s extremely likely that humanity would be able to feed itself basically indefinitely after a catastrophe killing 99.99% of people.
What is the likelihood that humanity would survive natural shocks like natural disasters, climate fluctuations (drought), and disease in the decades/centuries after the collapse?
Time spent on this section: ~1 hour
Types of sources: Our World in Data and expert interviews
While the surviving population in Case 3 is very small (for the decades or centuries until it grows again), one might worry that it could be driven all the way down to extinction as a result of natural shocks. I think it’s very likely that many groups of survivors would be seriously threatened — and some wiped out — by shocks like drought, hurricanes, disease outbreaks, or pests. But in the aftermath even of a Case 3 civilizational collapse, I expect survivors will be somewhat distributed around the world (or hemisphere) for several reasons.
First, it seems likely that the survivors would start off in different regions of the world just by chance. Second, it seems likely that travel would be limited, meaning that at least some groups would be completely disconnected from others. And third, because natural resources and surviving supplies would likely be distributed all over the world, a strategy of spreading out — in order to draw on all of the resources available and to avoid competition — would likely maximize the possible number of survivors. Given these pressures toward geographic distribution, any single shock might affect some or even many groups of survivors in a single region, but it’s hard to imagine any shock that would come close to affecting an entire hemisphere or even continent all at once. We can therefore expect that regional shocks like droughts and disease outbreaks wouldn’t be enough to wipe out groups spread out all over.
Historical base rates: Historically, natural disasters like wildfires, earthquakes, and droughts have been responsible for a pretty small share of global deaths. There aren’t great estimates for the death tolls caused by natural disasters before the mid-nineteenth century, but the most deadly year in the last century was 1931, when over 3.5 million people died as a result of natural disasters — just under 0.2% of the world population at the time. My impression is that, before the 20th century, regional shocks sometimes decimated local populations (and occasionally played a significant role in the collapse of a single civilization), but that the geographic distribution of human societies across the world meant those shocks never threatened the entire species.
There are a few reasons to think that shocks might be more frequent after a collapse. For example, anthropogenic global warming will continue to get worse (even without added emissions), leading to more climate volatility, including more frequent and severe hurricanes. And if shocks became more frequent, it becomes more plausible to think that all of the survivors might be affected by various natural shocks all at once. To my knowledge, they would not be so frequent as to mean we should expect human survivors would fare much worse than our ancestors did. They’d have to be either much more severe or much more frequent — probably both — but this is an area of uncertainty for me.
One scenario that seems like it could more plausibly lead to human extinction is one where, as described above, the survivors of the initial catastrophe are confined to a small area (like a region of Australia). Were this to be the case, it seems more plausible to me that a single shock could threaten human existence.
Bottom line: If groups of survivors were geographically distributed, as I imagine they would be, it’s hard to imagine how even a very severe natural disaster would lead to the deaths of all of the remaining survivors. If ~all of the survivors were in a small area — if there were basically no groups of hundreds or thousands of survivors anywhere else in the world — it seems plausible that all of the survivors could be wiped out directly or indirectly by a natural shock. But as I said above, I think this degree of geographic concentration seems extremely unlikely. So I think it’s extremely likely that humanity would survive natural shocks indefinitely.
Concrete example: A large nuclear war and the use of biological weapons, resulting in the death of 99.99% of the population, and causing infrastructure damage and climate change
So what do I think would happen in the event of a catastrophe like a global war involving both nuclear and biological weapons that led to the death of 99.99% of the population, caused severe infrastructure damage, climate change?
I expect hundreds of thousands of the survivors of the initial catastrophe would be able to subsist using leftover food stocks and supplies, before eventually working out how to feed themselves through traditional agriculture and fishing and/or modified agriculture (using methods that don’t rely on climate factors like warm temperatures and regular precipitation) (similar to the case of a nuclear war-only catastrophe).
Like before, for catastrophes that cause temporary climate effects, survivors would only need to subsist using non-traditional agriculture for a decade at most. After that, survivors could revert to traditional agriculture, which would be much easier.
At this level of population loss, a naive BOTEC makes me think that surviving populations would either stay above the minimum viable population, or find each other (by crossing distances as large as England or Wales). My confidence in this conclusion is further strengthened by the fact that there would be scenarios where survivors would retain or recreate long-distance communication, long-distance transportation, or are confined to a smaller geographic region (e.g. the Southern Hemisphere), making contact even easier.
While it seems likely to me that natural shocks would reduce the number of people that would survive in the decades following the initial catastrophe — likely to somewhere in the tens of thousands — I think it’s fairly unlikely that either of these would actually lead to extinction. This is because neither pathogens nor natural disasters have ever come close to causing human extinction, and have only relatively rarely even caused the collapse of a single human civilization. This seems largely due to the geographic spread of humans across the planet, which makes it hard for any given disaster to affect all groups of humans. I expect the post-collapse conditions to lead to a similar or greater level of geographic distribution.
Given this, I think it’s fairly likely, though far from guaranteed, that a catastrophe that caused 99.99% population loss, infrastructure damage, and climate change (e.g. a megacatastrohe, like a global war where biological weapons and nuclear weapons were used) would more or less directly cause human extinction.
My main uncertainties in Case 3 are around 1) whether I’m underestimating the frequency and severity of shocks the survivors might face in the post-collapse environment, and 2) whether surviving groups might deliberately avoid each other — perhaps because of fears about pathogen spread — allowing their populations to dip/stay below the minimum viable population.
Another toy calculation gives an indication of how wrong I would have to be for this scenario to lead to human extinction.
Note: As above, I assume each group’s fate is independent of the fates of other groups. In Case 3, I think this is a pretty robust assumption. I would expect survivors to be in relatively small, scattered, isolated groups for the reasons discussed above.
As discussed, the higher the true correlation between survivor groups, the more my toy calculations will cause me to underestimate the probability that all of the survivor groups would be wiped out.
With 800,000 survivors, the most plausible scenarios still seem unlikely to lead to extinction, even under very pessimistic assumptions about the vulnerability of each group of survivors.
If you thought any given group of 10 or 100 survivors had a 99% chance of being wiped out, it would still be virtually guaranteed that at least one group would survive.
If you thought there was a 99% chance that any one of 800 groups of 1,000 people would be wiped out, there would still only be a 1 in 3,000 chance of extinction.
If the survivors were in groups of 10 thousand people (80 groups total), and each of those groups had a 99% chance of being wiped out, there would be a 45% chance that humanity would go extinct. I think a 99% failure rate among groups of 10 thousand survivors is too pessimistic. I can imagine scenarios in which a single event or shock like a drought, nuclear meltdown, or natural disaster might have a chance at causing an entire population of 10,000 people to die. But I don’t think these are likely or frequent enough to put the overall probability of the loss of everyone in a group of 10,000 dying at 99%. Personally, I would put the probability that any of these groups is wiped out at 90%, at the highest. At this failure rate, the chance of extinction is about 1 in 5,000. But if you think 99% is reasonable, then it would be reasonable to think there is a 45% chance that this catastrophe would cause human extinction.
If the survivors were in groups of 100,000, and you put the probability that any one of those groups got wiped out at 90–99%, you could conclude that human extinction was pretty likely. Again, I think a 99% chance that a population of 100,000 is wiped out is too high, but reasonable people might disagree with me. I would put the probability that any group of 100,000 survivors might be wiped out in a post-collapse environment at 80% at the highest, which yields a probability of human extinction just below 17%.
Given all of this, my subjective judgment is that it’s fairly unlikely that this scenario would more or less directly lead to human extinction.
Other research directions
A few notes on other approaches to answering some of the questions I discussed that I considered but didn’t pursue, plus a rough guess at how long each approach would take and whether it would require specialized knowledge or skills:
I’d love to see someone host something like a wargame, where people (ideally actual subject-matter experts) adopted particular roles — a power plant operator, the leaders of influential countries, journalists, military generals, business executives, etc. — and participated in simulating a catastrophe. If modeled after a traditional war game, participants would have a scenario described to them, then would take turns making decisions (“moves”), each with the aim of achieving their respective interests/acting on their values. I think such a game might teach us a lot about how individuals and institutions would behave in the aftermath of such a catastrophe. (50+ hours / Special skills required)
Looking at lessons from science fiction books and films. I’m not sure how much there is to gain here, but I think it’s possible that works of sci-fi offer useful insights into what types of conditions survivors would face after a catastrophe of this type. A list of some of the best civilizational collapse science fiction is here. (5–10 hours / No special skills required)
I think there’s literature on society-level factors that inform how people respond to disasters. For example, I think I’ve heard somewhere that places with higher levels of social trust have lower levels of looting, hoarding, and other antisocial disaster behavior. This might be worth looking into more. (2–5 hours / No special skills required)
In my discussion of Case 2, I consider one type of catastrophe that could lead to 90% population death, infrastructure damage, and climate change, where the climate change I’m referring to is nuclear winter (decreased temperature and rainfall) and the infrastructure damage I’m referring to isn’t worldwide. But there are other cases one could choose with different considerations worth exploring. For example, in a scenario with an increase in the frequency of natural disasters (like extreme anthropogenic climate change), each group of survivors would be at a higher risk of extinction from things like hurricanes and droughts. This could have important implications on the bottom line for this case that make it meaningfully different from this one. It could be useful for someone to break down the starting conditions beyond what I’ve done (e.g. rather than thinking about climate change as one parameter, one could think about temperature, precipitation, and natural disasters separately; or, instead of thinking of infrastructure damage as one parameter, one could think about electronics, transportation, etc., separately). (10–30 hours / No special skills required)
I’m not sure how informative this would be, but I wonder if it could be interesting to see if anyone could try to take a survival guide and some basic camping supplies and try out various survival skills like building a fire and gathering an edible meal (without advanced preparation, the internet, or help from others). (3–30 hours / Maybe some adventurousness required)
How much more frequent/severe would natural shocks have to be to threaten surviving populations of different sizes? (2–10hours / Some math required)
I think this is a pretty reasonable assumption in the event of 99.9% population death. I expect that the survivors of a catastrophe this extreme would end up somewhat spread out (because this will mean less competition for the resources within a smaller area). The farther apart surviving groups are, the less likely they are to be affected by the same shocks (natural disasters, outbreaks, conflict). Additionally, in the event of civilizational collapse, transportation, communication, and other technologies that facilitate contact between geographically distributed groups would be enormously limited. This would further limit the extent to which each group’s fate ended up relating to another’s. There would be other sources of variation between groups: Some groups might be made up mostly of farmers, while others will be made up of lawyers; some groups will tend toward cooperation, while others toward conflict, plus pure randomness (e.g. some groups might have a high proportion of survivors with genetic immunity to a particular disease).
On the other hand, there are also factors that point in the other direction — factors that suggest the surviving groups would be at least somewhat correlated. For example, nuclear winter climate conditions, while nonuniform, would impact all surviving groups. Similarly, larger natural disasters might affect large regions, meaning that at least all of the survivor groups at the regional level might end up experiencing very similar challenges to survival simultaneously.
This last bit suggests that we could see economic collapse even if we weren’t seeing massive population losses, but rather just an unwillingness or inability of enough people to work, as might happen in a pandemic with a disease enough people were unwilling to risk contracting (or which made people sick enough not to work, even if it didn’t kill them).
Resource-pooling aims to improve the efficiency and cost effectiveness of systems that need to meet unpredictable demand. For example, demand for firefighters is both unpredictable and high-variance — there are zero or few fires most of the time, but sometimes there are several. To be prepared to respond to several fires that might all happen at once, a precinct would have to employ a lot of firefighters. But this means a lot of firefighters who are idle most of the time. To reduce inefficiency, most precincts pool resources, employing fewer firefighters while building in agreements between precincts that allow one to call on reinforcements from neighboring precincts.
Resource pooling is widespread: For example, banks pool capital reserves, and power plants have agreements allowing them to shift power demands to another power plant if demands are unusually high at one plant. While this makes these systems much more efficient — and also more effective in the short term — it makes them more fragile when viewed over long time frames, during which there are more opportunities for rare but extreme events to cause particularly high demand. Arguably, this kind of resource pooling is at least partly responsible for the severe ventilator shortages during the COVID-19 pandemic.
Because the components in a system with pooled resources are “coupled” (the performance of one is linked to the performance of others), resource pooling means that if 50% of the population were killed, and then even just a few components of a system with pooled resources becomes compromised, the whole system could fail. In other words, the reason systems with pooled resources are more fragile is because it means the nodes in a system become coupled — the performance of one is dependent on the performance of another. This means that if one fails, it’s much more likely that many others fail. To make this more concrete: imagine that a set of government hospital pools medical supplies, sharing between them when demand is particularly high at one hospital. Most of the time, this saves hospitals money, by allowing them to work well at lower safety margins (lower levels of surplus supplies). But when demand is high at all of those hospitals, they’ll all fail to meet demand for that supply. When the failure of one node causes others to fail, this is called a cascading failure.
Hyper-specialization is when systems require several subsystems to all work in order for them to have any output at all, because each ingredient of the output is produced in a separate subsystem. For example, a microchip manufacturer — rather than having several factories with the capacity to build microchips from start to finish — might build each individual subcomponent in different factories, such that failure of one such specialized factory will destroy the company’s entire manufacturing capacity. If a system is hyper-specialized in this way, then losing even a small percentage of its component subsystems (which seems like it could plausibly happen if 50% of the population were killed) would mean it would fail altogether.
In experimental literature that explores the effects of these approaches and their propensity for cascading failure, it’s often found that tightly coupled systems perform poorly over periods long enough to observe rare but extreme shocks (e.g. Elga, n.d.). For example, one study assigned participants to the role of the CEO of a power company (and tasked with the goal of maximizing profits) that supplied the power to ten cities. Some of the participants were allowed to use resource pooling — shifting surplus power supply in one city to meet unexpectedly high demand in another city. The participants were then allowed to choose what safety margins to adopt (the amount of power produced to meet an imperfectly predictable demand). Computer simulations were then run to see how often the participants’ imaginary power supply companies would be able to meet the power demands of the city. Participants whose systems were coupled were 35 times more likely to experience a total system failure, failing to meet the power needs in all ten cities (Elga, n.d.).
If the shock is as extreme as 50% of the inputs disappearing, it seems plausible that the system could fail altogether.
There seems to be a large and fascinating literature on the interconnectedness of systems, and how particular connections between systems make them particularly vulnerable (some examples: Buldyrev et al., 2009, Barigozzi et al., 2010). Unfortunately, a lot of this literature was inaccessible to me (it was highly quantitative, used a lot of network theory I don’t understand, etc.). But I think it could be worth looking into more.
Henrich describes in detail the highly specific knowledge that allowed local peoples to thrive in these environments. In the Arctic, this included the ability to identify the breathing holes that seals make in the ice, and to identify older ice which was low enough in saline to drink when melted, to name just a few. In Australia, it was a complex, non-obvious, multi-step process for processing one of the only available plants in the desert — the nardoo plant, which will slowly deplete the human body of vitamin B1 if not processed correctly. Henrich emphasizes that it was that collective knowledge, discovered and then passed down over generations, that allowed local populations to survive in these environments. Given that most people have not been taught this kind of knowledge, it seems plausible that we’d not have enough knowledge of the environment and agriculture to survive post-collapse.
26.7% of the world population derives its income through agriculture (p.46); this FAO report claims that “About two-thirds of the developing world’s 3 billion rural people live in about 475 million small farm households, working on land plots smaller than 2 hectares.” (p. 1).
I expect that survivors would be able to work out how to consume grain stocks, despite the fact that they may be unprocessed, meaning that they may not be immediately ready for consumption. I believe this would be an overcomable obstacle, but if I’m wrong, it could limit the grace period to the shorter end of the range above (a few months).
Details of how I thought about food rations: I assume rations imposed wouldn’t reduce caloric intake below 2,100 per day — which is the number of calories required to be offered to refugees in refugee food programs, like that offered by the World Food Programme. If all countries that currently consume over 2,100 calories imposed food rationing that limited intake to 2,100 calories, there would be a ~20% caloric surplus on average. Calculations here.
Assumes that 30 days worth of water bottles, consumed at the rate of 1 million per minute under normal circumstances, would be stored and accessible. Also assumes that people drink ~3 water bottles per day.
Assumes that survivors could survive on just one bottle of water per day.
See dataset titled “World Grain Consumption and Stocks, 1960-2012.”
Here, food stocks is defined in the source paper as a group of “92 products” used “to reconstruct 50 years of aggregated food reserves, expressed in caloric equivalent (kcal), at the regional and global scales.” (Laio et al. 2016)
Note that many people don’t actually rely on any medications, so these are only critical to some. Additionally, perishability depends on the specific medication, but medications generally last 1–10 years before expiring.
In some places, power outages would occur within 4–24 hours. In regions where power is supplied by particularly robust power plants (e.g. the Hoover Dam), it’s theoretically possible that the power supply could go on without human support for several years. In either case, generators could be scavenged from roadwork/construction sites once centralized power supplies stopped working.
Note: Extreme climate change can mean lots of things, including increases or decreases in temperature, precipitation, and frequency of natural disasters like hurricanes, among others. The considerations I discuss under Case 2 won’t apply to all of these types of climate changes. See the section on other research directions for possible research projects that could fill this gap.
And while it’d be important for survivors to make sure they could meet their nutritional requirements, this might not be as hard as it seems. For example, consider that one could survive indefinitely on a diet of white potatoes, sweet potatoes, and butter (Nunn, 2009).
Given that there would be major infrastructure damage, it seems likely that a substantial portion of food stocks and other supplies would be destroyed during the initial catastrophe. It’s hard to say how much. For just a very rough sense, I assume 55% of these resources would be destroyed, based on the fact that about 55% of people live in countries with nuclear weapons or in nuclear weapons alliances (so, perhaps we can assume that the supplies destroyed would be proportionate to the population of countries likely to be affected by a global nuclear war).
In the Southern Hemisphere, I include all of the land area in Africa, Australia, and South America (so this is a crude estimate).
Making a makeshift radio might be a surprisingly accessible technology, Lewis Dartnell explains in The Knowledge:
“During the Second World War, soldiers holed up at POW camps built their own makeshift radio receivers for music or news of the war effort. These ingenious constructions reveal the sheer variety of scavenged materials that can be jury-rigged to create a working radio. Aerial wires were slung over trees, or disguised as clotheslines, and sometimes even barbed wire fences were appropriated for the task. A good grounding was achieved by connecting to cold-water pipes in the POW barracks. Inductors were constructed by winding coils around cardboard toilet rolls, the scavenged bare wire insulated by candle wax, or in Japanese POW camps by applying a paste of palm oil and flour. Capacitors for the tuning circuit were improvised out of layers of tinfoil or cigarette-pack lining, alternating with newspaper sheets for insulation; the wide, flat device was then curled like a jelly roll to make a more compact component. The earphone is a trickier component to improvise and so was often salvaged from wrecked vehicles… Perhaps the most ingenious improvisation of all, however, was in creating the all-important rectifier, needed to demodulate the audio signal from the carrier wave. Mineral crystals like iron pyrite or galena were unobtainable on the battlefield, but rusty razor blades and corroded copper pennies were discovered to serve just as well. The blade or coin was fixed to a scrap piece of wood alongside a safety pin bent upright. A sharpened graphite pencil was firmly attached to the point of the safety pin… allowing fine readjustment of the pencil graphite across the metal oxide surface until a working rectifying junction was found.”