Throughout history, for most of the time, world population has been growing, with the rate of that growth also increasing along with the transitions between the foraging era, the farming era, and the industrial era. While declines in the population of specific regions or countries have been more commonplace, declines in the overall world population have been quite rare and haven't lasted very long. The two main causes of such occasional reversals in the overall trend of growth have been pandemics and wars, in that order. While both the data about historical world population and the death tolls of various pandemics and wars are uncertain, what we do know is sufficient to at least give us a sense of the order of magnitude of the worst population crashes in history.
Shifting focus from the past to the future, many of the concerns about existential risk have a natural weaker counterpart in the form of concerns about population crashes which are not large enough to tip the species over the precipice of extinction. Since humans have never actually gone extinct or come anywhere close to it during or after the agricultural era, the only base rate information we have comes from the history of global population declines. It's therefore worth looking over the biggest cases to understand their causes along with their magnitudes.
Worst years In History
To get a sense of both the scale of the world population declines that have occurred in the past along with the causes behind them, a case-by-case examination is appropriate. First: a graph of deaths due to pandemics and wars throughout history. (The x-axis is the year, the y-axis is deaths per year per million people.) This graph was produced based on a non-exhaustive list—but I've tried to include all the events which have a chance of qualifying, so it should still serve as a useful guidepost.
1331 to 1353: The Black Death
The Black Death was the deadliest single event in human history by a wide margin no matter which criterion is used. It killed more people than the Second World War at a time when the global population was only a small fraction of what it was in 1939. The total death toll is estimated to be anywhere between 75 to 200 million people, and if we take geometric means, that comes out to roughly 1.4% of the world population dead per year on average throughout this period. India was spared the worst of the devastation, but the rest of Eurasia was decimated by the plague.
The worst and final outbreak occurred in China in the year 1353: Over 20 million people died in one year, which at the time would have been more than 5% of the global population dead. This will be the largest decline we'll find anywhere in this list, so we can tentatively assume the largest single-year population decline of the past 2500 years, and possibly of the whole agricultural era, was roughly of this magnitude.
541 to 544: The Plague of Justinian
This is perhaps the event on the list where the death toll is most uncertain. The Plague of Justinian is the first known outbreak of bubonic plague, the disease that was also responsible for the Black Death. The plague arrived in Constantinople in 541 and had a devastating impact, by some accounts killing over 5000 people per day for several of the worst months of the outbreak. The Eastern Roman Emperor of the time, Justinian, contracted the disease himself but managed to survive.
Sources suggest anywhere between 15 to 100 million people perished, but what is harder to get a handle on is exactly when this took place (1). The Plague of Justinian gave way to the first plague pandemic, which became endemic in regions around the Mediterranean, with continued outbreaks common for the next two centuries. The death tolls quoted are often for the first plague pandemic overall, and since we're concerned with sharp population declines in a single year, a claim such as "40 million deaths over two centuries" can obscure a lot of relevant variance in how the death toll is distributed over time. Data suggest that around two-fifths of the overall death toll is attributable to the short period from 541 to 544; a period of four years. If we take the geometric mean of the two endpoints I've cited, that gives us a rate of 2% of the global population dead per year for the years from 541 to 544.
Depending on where the true death toll falls in the range of estimates, the Plague of Justinian might have been the second deadliest event in human history.
756 to 763: An Lushan Rebellion
This was a civil war in China marking the end of the golden age of the Tang dynasty. A Chinese general of foreign origin suspected that he was about to be removed from his post by Emperor Xuanzong and launched a preemptive war to secure his position, which engulfed the most densely populated regions of the country. War-related death estimates vary wildly: The most objective measurements come from Chinese census records which report a loss of 36 million people from 753 to 764. It's likely, however, that part or even most of this fall is explained by the breakdown in the capability of the central government to raise taxes from as large of a tax base as they could count on previously.
Going with the upper bound of 36 million would give us a rate of global population decline that's about 2% per year. Matthew White indicates the number of deaths as 13 million, which gives approximately 0.7% per year of population decline. Even if we assume this lower estimate is correct, the An Lushan Rebellion might still be the second deadliest event in human history.
1918 to 1920: The Spanish flu
The Spanish flu was the worst pandemic in recent history and one of the worst pandemics in all of recorded history, with a death toll ranking third after the Black Death and the first plague pandemic. It was exacerbated by the fact that the First World War was still ongoing when the pandemic began to spread, and the military operations and troop movements required by the continuing global war effort accelerated the course of the pandemic worldwide. While WW1 has much more captured the attention of historians, the Spanish flu was more deadly.
An average estimate is that around 35 million died in a space of two to three years from the Spanish flu compared to 15 million from WW1 within the space of four years. Added together, perhaps around 1% of the global population may have perished in the year 1918, slightly more than the rate of natural population growth at the time.
1206 to 1227: Mongol conquests
The initial phase of the Mongol conquests appears to have been the most bloody. During the reign of Genghis Khan, the Mongols invaded northern China and conquered much of Central Asia in wars against Khwarazm. Many cities were razed and their inhabitants were massacred, but the biggest death toll likely came from the need of Mongol armies to requisition supplies from the locals while on campaign, as it was with most wars before the industrial era.
The commonly given death toll for this phase of the Mongol conquests is 40 million people. This amounts to a death rate of 0.5% of the global population per year. It is smaller than the examples above, but it is likely that there were at least some peak years in which the rate of death exceeded 1% per year.
Causes
The deadliest events in history have been outbreaks of bubonic plague. We may be comforted by the fact that we now have modern medicine with all of its tools to assist in such outbreaks, as well as much more ability in reserve to go into extended quarantines to stamp out any outbreaks, but we also have greater capabilities for designing new biological weapons. Perhaps the most worrying scenario is a war combined with a pandemic—either because the war creates the conditions for the pandemic to spread more effectively and hobbles the measures that would otherwise be taken to combat it (as it was with WW1 and the Spanish flu), or because the parties engaged in war design or otherwise employ bioweapons against the enemy which give rise to pandemics.
In the most deadly wars, most people who died were civilians; and most of them died due to lack of food and disruption of normal economic life by trespassing armies, friendly and hostile, rather than being massacred. This applies to the An Lushan Rebellion, the Mongol wars of conquest, and other Chinese dynastic collapses. David Morgan raises another possibility not often discussed in regards to the Mongol conquests, which is the damage they could have done to the infrastructure that supported agriculture (2). If enough peasants are killed, the irrigation systems that kept the land fertile could fall into disrepair, and as a result much of the land which used to be arable could revert to desert until the necessary infrastructure was set up again. In the meantime, the famine would devastate local populations who were previously living at a near-subsistence level already.
Before the industrial era, the biggest killer of soldiers in war was disease. Typhus, dysentery, and cholera were some of the usual culprits since armies on the march tended to have ideal conditions for these diseases to spread. Even during the Napoleonic Wars, more soldiers perished from disease than were wounded or killed in combat. Advances in medicine and armament technology, along with general economic growth, had reversed this situation by WW1: The primary cause of death for soldiers then became enemy soldiers. A similar transition could happen with civilians who die in war. Instead of their deaths being the byproduct of the devastation brought about by war, in the future the main danger to their lives could become being targeted by deliberate measures of extermination by hostile forces, such as the use of nuclear weapons on densely populated cities or being infected with bioweapons designed to inflict maximum damage on the populations of hostile governments.
Forecasts
Using a point forecast of world population at some future date as a proxy for a population crash is problematic since unless the crash is enormous the other sources of variance in this forecast (such as birth rates, whether the logarithm of global population is a unit root process, et cetera) lead to a low signal-to-noise ratio. Instead of taking that approach, I think the good questions are those which try to probe views about the likelihood of large declines in short time periods or after some specific event such as the outbreak of a world war.
For this question, a base rate calculation could be that over 2500 years we had 5 distinct events which led to a population decline of greater than 1% in a single year, so the base rate for such an event starting to occur in any given year is roughly 0.2%. Over 100 years that gives chances of about 20% that we might see such an event, and the falling birth rate trend means it will be harder in the future for population growth to compensate for such declines. Likewise, we have 3 examples of a decline of 2% and only 1 example of a decline of 5%. We can fit the cumulative distribution function of our forecast to these base rates.
This question is more challenging, since the obvious comparisons with WW1 and WW2 don't yield anything useful: Neither of these wars produced any decline in world population. However, the major reason for this in the case of WW2 was the high population growth rate which was sufficient to make up for the losses caused by the war, which were on the order of 0.4% of the global population lost per year. If WW2 had occurred under conditions of low population growth it would have produced at least a mild decline in global population.
The more glaring problem is how we should update from the historical base rates given the compelling inside view that WW3 has the potential to be significantly more destructive than any war of the past. I'm personally wary of making large updates, however. Bioweapons were actually used in the past: Some accounts have it that invaders from the Golden Horde deliberately spread the bubonic plague to the city of Kaffa in the Crimea during a siege in order to get the city to surrender, and Europeans used similar tactics when they came into conflict with the natives in the Americas. The ancient or medieval worlds didn't have nuclear weapons, but the old-fashioned means for massacring civilians in cities that refused to surrender was employed throughout history with quite primitive technology.
The reason massacre of civilians was much rarer than technology would have permitted, even in times past, is that most armies at war had no interest in massacring civilians. Such actions were only taken if they served some concrete purpose. The situation is likely to be similar in a war between two states with nuclear weapons: Even if they enter a conventional war, nuclear weapons may be held in reserve and not used because both sides would make the off-equilibrium threat of retaliating to a first strike with a strike of their own. These threats can go wrong, but they are likely more stable than people give them credit for. To see this scenario is plausible, I think we can look at the Korean War: the US had nuclear weapons but didn't use them, partly out of concern that such an escalation would drag the USSR into the war, which by that time had nuclear weapons of its own. The Korean War is in fact an even stronger example than we need, because the mere presence of the USSR was enough to deter the use of a nuclear weapon by the US even though neither of the actual belligerents in the war against the US, China and North Korea, were in possession of a nuclear weapon.
Overall, I believe there is a two-thirds chance that escalation in the use of WMDs would be avoided in a potential WW3 and a one-third chance that it would not be. If escalation is avoided, then I think historical base rates offer good guidance for the kind of death tolls we might expect. If it isn't avoided, then the belligerent countries could plausibly lose large fractions of their population—but I think the fact that a large fraction of the world population would likely live in neutral countries can put an upper bound on the magnitude of the possible decline.
For this question, we can also rely on base rates. Wikipedia helpfully provides a list of past epidemics along with estimates of their total death toll, and combining this with the information given earlier in this essay about the worst global pandemics in history gives a reasonable picture of a "business as usual" scenario. There are aspects of the inside view which could push us in either direction compared to base rates when it comes to the right tail of the distribution: On one hand, the deadliest pandemics in history were due to outbreaks of bubonic plague which would be relatively easy to handle today. On the other hand, advances in technology also mean that human engineered pathogens pose a risk which was not present in the past to the same extent.
I tried to balance these considerations in my forecast, but generally I tend to be wary of deviating from base rates unless the inside view considerations are clear and mostly point in the same direction. This is reflected in my forecast, and I'm curious to see how my forecast will be different from the community forecast.
Conclusion
There are certainly limits to how much we can learn from the past given the differences between the past and the future, but I do believe base rates give useful information about the future likelihood of population crashes and what they are likely to be caused by. If anyone has compelling arguments that justify departing from base rates in this context, please share them in the comments.
- Mordechai, Lee; Eisenberg, Merle; Newfield, Timothy P.; Izdebski, Adam; Kay, Janet E.; Poinar, Hendrik (2019-12-17). "The Justinianic Plague: An inconsequential pandemic?". Proceedings of the National Academy of Sciences. 116 (51): 25546–25554. doi:10.1073/pnas.1903797116. ISSN 0027-8424. PMC 6926030. PMID 31792176.
- Morgan, D. (2007). The Mongols. Oxford: Blackwell.