Explanatory note: This page grew out of one of my investigations for Open Phil, but then I got fascinated and put a bunch of personal time into elaborating certain parts of it, and it evolved into something that I think is pretty cool, but which would take more work than it’s worth to vet and edit it such that it would be appropriate for Open Phil’s website, so we decided I should just post it here instead as a personal project. Hence, the below doesn’t represent Open Phil’s position on anything, and should be taken merely as my own personal guesses and opinions.
(Probably best to start with my companion blog post.)
One way to look for opportunities to accomplish as much good as possible is to ask “Which developments might have an extremely large impact on human civilization,1 and is there any way we can (in expectation) nudge those developments in a positive direction?”
For example, in the context of philanthropy, the Rockefeller Foundation funded work on an improved agricultural approach that led to the Green Revolution, which some people have credited with kickstarting the development of the “Asian Tigers,” helping several countries transition from “poor” to “middle income,” transforming India from being in the middle of a famine to being a wheat exporter, and saving over a billion people from starvation.2 Of course, the Rockefeller Foundation had no way of knowing their funding would have such incredible impact, but a rare win of that magnitude can make up for a large number of failed (and similarly uncertain) funding efforts. (See Holden Karnofsky’s hits-based giving.)
However, some future developments might have even greater impact than the Green Revolution, and be more comparable in magnitude to the changes often attributed to the industrial revolution. Here, I refer to changes of this magnitude as “transformative,”3 and I refer to developments which might precipitate such transformative changes as potential “transformative developments” for human civilization.4
In the future, I hope to spend more time identifying potentially transformative developments,5 especially those which might also be tractable and neglected. In this report, I hope to lay some groundwork by examining the magnitude of “transformative” change. In particular, I ask:
- The industrial revolution is often considered the most transformative event in recorded history.6 How large, exactly, were the differences in human well-being before and after the industrial revolution? [more]
- Have there been other transitions in recorded history of comparable magnitude, either positive or negative? [more]
- How catastrophic would a development need to be to plausibly result in negative transformative change? [more]
- What do these initial findings suggest about potential future transformative developments? [more]
My initial tentative conclusions from this preliminary investigation can be summarized as follows:
The gains in human well-being observed since the industrial revolution are vastly larger than pre-industrial fluctuations in human well-being. No other transitions in recorded history, either positive or negative, are remotely similar in magnitude. When thinking about which future developments might be most important, we should not forget that the size of their likely impact may differ by orders of magnitude. For example, a universal cure for cancer would bring a huge benefit to human well-being, but its expected impact seems likely to be vastly smaller than (for example) the likely impact of AI systems capable of automating most human labor, or the counterfactual benefit of preventing large-scale nuclear war.
Human well-being before and after the industrial revolution
How large were the differences in human well-being before and after the industrial revolution?
My strategy for providing a preliminary answer to this question is the following:
- List some key aspects of human well-being.
- Find measures for those key aspects of human well-being, for which we have estimates going back to (say) 1000 BCE.
- Chart the data for each measure, from 1000 BCE – 2000 CE.
- Inspect the magnitude of the difference in these measures between the pre-industrial and post-industrial eras.
To compare human well-being “before” and “after” the industrial revolution, I need to clarify which time periods I have in mind. Typically, the industrial revolution is considered to have occurred in Britain from roughly 1760-1830.7 As we’ll see below, while very large differences in human well-being took many decades (after the industrial revolution) to accrue, the “trajectory change” (i.e. the “bend in the curve”) for many global measures of human well-being occurred either toward the end of the industrial revolution or shortly thereafter (say, from 1800-1880). Hence, for my purposes, I’ll refer to the period from 1000 BCE to 1760 as the “pre-industrial” era, and the period after 1800 as the “post-industrial era.” (For ease of analysis across multiple data sources, I limit my analysis up to the year 2000.)
For convenience, I’ll assume the apparently common view8 that the “industrial revolution” (what collection of events is that?) is not just coincident with, but in fact is the primary cause of the major trajectory changes observed during this period, but I should be clear that I think this is quite plausibly wrong, and certainly hard to prove. In any case, my primary focus here is to try to assess the magnitude of the changes that occurred during this period, not to argue for a particular causal story.9
Key aspects of human well-being
How should we define human well-being? Without aiming for philosophical precision,10 here are some aspects of human well-being that seem especially “fundamental” to me:
- Subjective well-being (positive feelings, absence of negative feelings, a sense of satisfaction with one’s life, a sense of meaning and purpose, self-esteem, etc.)
- Physical health
- Economic well-being (income, wealth, etc.)
- Three aspects of empowerment to live the kind of life one wants:11
- Energy capture (it’s easier to build your own house with power tools that pull large amounts of energy from the power grid than it is to do so without access to such large energy resources)
- Technological empowerment (you can’t use power tools to build your house if power tools haven’t been invented yet)
- Political freedom (you can’t build your own house if others prevent you from doing so)
- Social well-being (trust, community participation, social capital, relationship satisfaction, social support, etc.)
No doubt this particular list is controversial. One could easily (and perhaps persuasively) argue for some other list of key aspects of human well-being; I discuss some alternate possibilities in Appendix A.
Long-run measures of human well-being
Unfortunately, my choice of measures for the aspects of human well-being listed above is heavily constrained by the limited availability of long-run data. In particular, I was not able to find any useable long-run measures for subjective well-being or social well-being, since these aspects of well-being have been measured only quite recently.12
After some searching, the long-run measures of human well-being I settled on for this report are:
In the sections below, I explain the rationale behind each measure, chart each measure over time, and discuss potential alternative measures that I chose not to examine here.
Life expectancy at birth
First, let’s look at physical health, as measured by life expectancy at birth. Life expectancy at birth is a pretty good proxy for how physically healthy people are in general, since it sums together in a single measure the assaults on physical health we worry most about — malnutrition, infectious disease, physical violence and war, physical injury due to accidents and natural disasters, etc. — and also captures the success of efforts to preserve people’s physical health, such as food supply, public health measures, efforts to preserve peace both within and between societies, etc. (Note also that improvements to life expectancy at birth are not entirely due to falling rates of childhood mortality: see here.)
What does the long-run history of life expectancy look like? According to Roser (2016a):
Estimates suggest that in a pre-modern, poor world, life expectancy was around 30 years in all regions of the world. In the early 19th century, life expectancy started to increase in the early industrialized countries while it stayed low in the rest of the world… Since 1900 the global average life expectancy has more than doubled and is now approaching 70 years. No country in the world has a lower life expectancy than the the countries with the highest life expectancy in 1800.
According to available data, it seems there was a sharp upward bend in the trajectory for life expectancy in Europe following the industrial revolution, with delayed trajectory changes for countries that industrialized later (such as China and India):
If we chart global average life expectancy from 1000 BCE to 2000 CE, using a slowly rising line to represent the apparent view of historians that average life expectancy probably fluctuated between 25 and 30 (and probably rose slightly over time),13 the long-run history of life expectancy looks like this:
If we had true data covering pre-industrial times, we would no doubt see downward spikes in global life expectancy as a result of major catastrophes such as the Black Death. However, as far as I can tell, in all such cases life expectancy recovered quickly to (roughly) the previous status quo, and no such catastrophes produced a lasting trajectory change in global life expectancy. (See also below.)
If we believe these estimates, then pre-industrial, global average life expectancy may have changed over time by as much as ~5 years. In contrast, the difference between pre-industrial and post-industrial global average life expectancy (at their highest points) is more than 7 times that amount, at ~36 years (through 2000 CE).
Other measures of physical health
Below I list some other measures of physical health and briefly explain why I didn’t use them in this report.
- Adult height is commonly used as a proxy measure for physical health, since poor health in childhood or adolescence can lead to growth stunting. Also, height can be inferred from bones, which are available even from ancient times.15 However, in the time I devoted to this project, I wasn’t able to make a confident-enough estimate of pre-industrial, worldwide average heights over time for me to include adult height in this report.16
- One could also examine long-run trends in the key factors of physical health and mortality that are aggregated together in life expectancy at birth — for example child mortality, maternal mortality, the global burden of disease, hunger and undernourishment, healthcare provision, war, homicide, and others — but these seem less fundamental to me than the measure that sums them all together.17
- Vitality is a measure of “having energy, feeling well-rested and healthy, and being physically active.”18 Unfortunately, this measure has not been used until fairly recently.
GDP per capita
Next, let’s look at the long-run history of economic well-being, as measured by GDP per capita. OECD (2014) explains this measure’s relevance to human well-being:
Economic well-being – people’s command over produced goods and services – can be assessed in an historical perspective through measures of gross domestic product (GDP) per capita… GDP (per capita) is an important indicator for measuring the economic performance of countries, which is a central driver of people’s economic well-being. This is true not only because an increased output of goods and services, which is what GDP measures, tends to translate into an increased ability by residents to buy these goods and services, but also because higher GDP provides the means for spending on non-material components of well-being, such as education and health.
What is the long-run trend of GDP per capita? Below I chart a series of historical GDP per capita estimates, adjusted for differences in cost and living across time and space:
In short, world GDP per capita (in 1990 international dollars) was relatively flat until the final decades of the industrial revolution, when the trajectory for this measure changed dramatically. By DeLong’s estimates, world GDP per capita hovered around 90-200 from ancient times up through 1800, then jumped to 300 by 1850, 679 by 1900, and up to 6,539 by 2000.
If we zoom in on the period from 1 CE to 1800 CE, we can see that (using DeLong’s estimates) GDP per capita actually fluctuated a fair bit during the pre-industrial era:
It is all the more striking, then, that these fluctuations are so small compared to the post-industrial rise in GDP per capita that they show up as a flat line when charting GDP per capita (on a linear scale) in a way that includes the post-industrial era.21
Another important measure of economic well-being is percentage of people living in extreme poverty. Extreme poverty is especially relevant to well-being due to the diminishing marginal impact of money on well-being: a $500 gift boosts well-being more when given to someone living on $2/day ($730/year) than when given to someone living on $50,000/year.
The short history of extreme poverty is this: for nearly all of history, nearly everyone lived in extreme poverty.22 More precise estimates of the proportion of world population living in extreme poverty (under various definitions) pick up the story starting in 1820:
In the chart above, “extreme poverty” (the red line) means living on less than $1/day, and “poverty” (the green line) means living on less than $2/day — both measures adjusted for cost of living (PPP).
If we represent pre-industrial times using a gradually rising line to represent a very slowly rising percentage of people not living in extreme poverty, the long-run trend for this measure looks like this:
Other measures of economic well-being
Other measures of economic well-being I could have used include:
- Other levels of poverty: I could have chosen different thresholds for what counts as “extreme poverty,” for example $5/day or even $20/day. This would result in somewhat different curves, but I suspect the overall story would remain the same.
- Other measures of production, expenditure, and income: In theory, GDP is equivalently a measure of production, expenditure, and income, because your expenditures are my income, my expenditures are your income, and so on. In practice, GDP is never perfectly known, and different ways of trying to measure production, expenditure, and income lead to somewhat different results. Hence, I could have used other measures of average production, expenditure, or income, but I used GDP per capita because it is the most widely available estimate over time and space.
- Economic inequality: Human happiness seems to respond not just to absolute income, wealth, and consumption, but also to relative income, wealth, and consumption.25 However, the details probably vary from culture to culture, and from person to person. One person may primarily compare their income to that of their co-workers, while another compares their income to that of their neighbors, while yet another compares their income to that of others in their city or country. Moreover, people may compare themselves to others on a complex and individually unique mixture of income, wealth, consumption, disposable income, and other variables. Perhaps the easiest inequality measure to use would be global income inequality.26 However, I doubt this is a particularly good measure of human well-being, relative to the ones I use here, because comparisons within a country or city are probably far more salient, and because the post-industrial rise in global economic inequality is generally a result of some nations escaping widespread extreme poverty for the first time in history, rather than a result of some nations or people becoming poorer than they used to be. Unfortunately, long-run global averages of local economic inequalities are not available.27
- Economic mobility, especially various measures of absolute mobility (since relative mobility is a zero-sum game), could be a useful complement to the measures I discuss above.28 Unfortunately, I’m not aware of long-run estimates for such measures.
Empowerment via energy capture
Energy capture must be the foundation for any usable measure of [human empowerment]… Without capturing energy, humans (like plants and other animals) die. Similarly, unless they take up energy from their environments, the societies humans have created break down. To increase their mastery of their physical and intellectual environments and get things done, groups of people have to increase their energy capture.
In other words: to get anything done, you need to capture energy from your environment and use it to do the work of accomplishing your goals. Given (almost) any set of goals, they will be easier to accomplish with more energy devoted to them.
Or as Smil (2017), p. 1, puts it:
Energy is the only universal currency: one of its many forms must be transformed to get anything done… From a fundamental biophysical perspective, …the course of history can be seen as the quest for controlling greater stores and flows of more concentrated and more versatile forms of energy and converting them, in more affordable ways at lower costs and with higher efficiencies, into heat, light, and motion.
Energy capture includes:30
- Food: directly consumed, fed to labor animals, or fed to animals later consumed.
- Fuel: used for cooking, heating, cooling, powering machines, etc., and provided by wood, coal, oil, gas, wind, water, nuclear, etc.
- Raw materials: for clothing, construction, pot-making, metalwork, or any other purpose.
Defined this way, energy capture is clearly related to GDP, though it is a different and broader, and thus partly independent, concept.
Drawing on a very wide range of historical sources and types of evidence, Morris estimated energy capture (in kilocalories per person per day) from 10,000 BCE to 2000 CE — not for the entire world, but for the (moving over time) “cores” of Western and Eastern civilization. Below, I chart his estimates starting from 1000 BCE:
Here, we see a very slow rise in energy capture per capita up until just after 1800, when energy capture per capita skyrockets upward in both the Western and Eastern cores.
I haven’t found global estimates of energy capture per capita over such an extended period of time. If such estimates could be constructed, my guess is that the hockey-stick curve seen here would be attenuated somewhat because energy capture is presumably smaller per capita outside civilizational “cores” (at any given moment in time).
Empowerment via technology
What measure can we use to estimate humankind’s technological empowerment, stretching back thousands of years? After surveying the options, I concluded that the best available measure for present purposes is Morris (2013)‘s measure of “war-making capacity.”
It might seem odd to use war-making capacity as a measure of human well-being, but consider the likely correlation between war-making capacity and other sorts of technological empowerment more obviously promoting of well-being. A society that can fly bombs to the other side of the world in mere hours can also transport humans and goods anywhere in the world in mere hours (whereas previously it would’ve taken weeks, months, or years). And a society that can mass-manufacture chemical and biological weapons carefully engineered for devastating effects has probably also discovered (or will soon discover) how to mass-manufacture synthetic fertilizer and penicillin.
Moreover, relative to other measures of technological empowerment, war-making capacity has several practical advantages as a proxy for technological empowerment:
- Unlike many other measures of knowledge and technology, war-making capacity focuses well on the “empowerment” aspect of knowledge and technology. For example, Maxwell’s equations represent a fundamental advance in human knowledge, but someone who learns them is not thereby much more empowered to achieve their goals, until someone uses Maxwell’s discoveries to develop new (e.g. electric) technologies, including those used in war.
- War-making capacity is also an unusually “concrete” measure of empowerment, in the sense that an actor with greater war-making capacity will prove its higher level of empowerment in a concrete way: by winning conflicts with actors at a lower level of empowerment (e.g. as seen during various waves of colonization).
- We have unusually good historical data (and syntheses of data) concerning war-making capacity, due to “historians’ obsession with recording wars, compulsive military record keeping, artistic patrons’ fondness of being portrayed as warriors, the widespread practice of burying dead men with arms and armor, [and] the archaeological visibility of fortifications…”32
- War-making capacity is a measure with relatively continuous meaning across thousands of years (unlike, say, patents).
Morris’ measure of war-making capacity is a purely relative one, answering questions such as “In 400 BCE, how powerful was the military of the Persian Empire (the Western ‘core’ at that time) compared to that of Egypt-Syria-Iraq (the Western ‘core’ of 700 CE) in 700 CE?” (The details are complicated, and I leave them to a footnote.33 ) Also, as with his measure of energy capture, Morris estimated war-making capacity only for the (moving over time) “cores” of Western and Eastern Civilization. Here are the results:
Granted, this does not show much beyond the fact that differences in war-making capacity prior to the industrial revolution are completely dwarfed by differences in war-making capacity on either side of the industrial revolution.35 But, this is an important historical fact worth remembering, and one that might be forgotten in detailed discussions of the military importance of pre-industrial war-making innovations such as chariots or cavalry, which represent much smaller increments of technological empowerment than planes, tanks, and nuclear weapons do.
Other measures of technological empowerment
Other measures I considered include:
- Peregrine’s Atlas of Cultural Evolution (ACE) is more a measure of “cultural complexity” than a measure of how “empowered” a civilization is. Moreover, it isn’t intended to capture the transformative changes of the last few centuries, which is why Peregrine’s charts all go up to 1500 CE and no later.
- Some sources use variables such as R&D investment or patents (e.g. Coccia 2014), but these data are only available for very recent times.
- Comin et al. (2010)‘s technology dataset includes only three “snapshots” of technological development: in 1000 BCE, 0 CE, and 1500 CE.
- Dong et al. (2016)‘s bibliometric approach is interesting, but unfortunately I was not able to obtain a copy of the dataset, and the earlier publications describing the dataset in more detail are available only in Mandarin.
- Murray (2003)‘s tables for “central events in various sciences” could provide another biblioemtric dataset, but Murray’s approach leaves out “technological” developments, and isn’t much of a measure of how “empowered” a civilization is by its knowledge and technology.
- I could also have used Morris (2013)‘s measure of information technology over time, but this seemed somewhat worse for measuring how “empowered” a civilization is by its knowledge and technology, compared to Morris’ measure of war-making capacity.
Empowerment via political freedom
Since democracies tend to be more responsive to the desires of their citizens than other types of regimes are, and tend to do a better job of preserving human rights, one proxy measure for political freedom is percentage of people living in a democracy. The long-run trend for this measure looks like this:
Since my source for which regimes count as “democracies” is the Polity IV dataset, and Polity IV only provides ratings back to 1800, the chart here shows a drop from 0.69% of people living in a democracy in 1800, to 0% of people living in a democracy for every year prior to 1800. That is of course somewhat inaccurate, but I’m confident that if Polity IV had tried to score regimes for their degree of democracy prior to 1800, the overall trendline would look much the same: i.e. nearly 0% of people lived in a democracy (as Polity IV defines them) until near the end of the industrial revolution, after which the percentage of people living in a democracy skyrocketed upward.
Other measures of political freedom
Below I list some other measures of political freedom and briefly explain why I didn’t use them in this report.
- Slavery: Perhaps the best measure of empowerment via political freedom would be “percent of people not enslaved,” since slavery typically removes or curtails a broad variety of political freedoms all at once. Unfortunately, it would be difficult (and perhaps hopeless) to try to draw lines between the different forms of slavery and bondage experienced in different cultures and centuries throughout history, and in any case we lack the data to make such estimates. See below for details on my aborted attempt to estimate “percent of people not enslaved” over time.
- Measures of some other human rights (besides protection against enslavement) could be considered, but these raise difficult measurement questions, and in most cases we lack estimates for all but the most recent decades.
- Percent of people politically enfranchised: I haven’t seen these data estimated over time, but presumably it could be done, using data on when different regimes granted and withdrew voting rights from different subpopulations (e.g. women).37 If we had these estimates, I suspect the curve for this measure would look much the same as for “percentage of people living in a democracy,” except that it would rise later and more gradually.
Summary of human well-being before and after the industrial revolution
If we chart my chosen measures on the same graph, preserving each unique y axis, we get the following:
In general, the gains and losses in (these measures of) human well-being during the pre-industrial era are miniscule compared to the gains made during the post-industrial era, and there is a sharp upward trajectory change for all these measures shortly after 1800. This provides a qualitative, as well as loosely quantitative picture of the magnitude of change I have in mind when I speak of “transformative” change.
It is worth remembering that the long-run history of human well-being could have looked quite different. It’s easy to imagine a world in which these key aspects of human well-being varied independently of each other, and fluctuated substantially up and down over time, akin to the pattern seen for GDP/cap from years 1-1800. Instead, they all run together, and they all show a single, sharp bend upward shortly after 1800.
Have there been other transitions in recorded history of comparable magnitude?
To many readers, it will not be a surprise that such a transformative change occurred shortly after the industrial revolution. What is perhaps more surprising is to consider how many major historical events seem to have not produced anything close to such a transformative change.
If we drop the y axes (for space reasons), and add other major historical events to the summary chart from the previous section, we get the following:
In short, no other transitions in recorded history are remotely comparable in their magnitude to the transition that followed the industrial revolution.
Perhaps the best contender for a transition of comparable magnitude is the agricultural revolution, but that transition predates “recorded” history, and I don’t discuss it much here because our data for human well-being during that period are so limited.
The possibility of transformative negative change
As shown above, there seems to be no precedent in recorded history for negative changes that would count as “transformative” in my rather extreme sense of the term. Nevertheless, is easy to imagine that some future catastrophe could cause a “reverse industrial revolution,” such that within a century or two (or much less) of its occurrence, human well-being was roughly what it was prior to the industrial revolution, or even worse.
Consider an extreme example: if a nearby supernova explosion destroyed all life on Earth, thereby setting all human well-being variables to zero, this would clearly qualify as an event of transformative negative impact.40
But short of outright human extinction, how bad would a catastrophe need to be, in order to produce a “reverse industrial revolution” or worse — that is, to have a transformative negative impact? To get a feel for this, I briefly studied the magnitude of the deadliest catastrophes in world history.41
How many deaths might result in transformative negative change?
Which are the deadliest catastrophes in recorded history? Death toll estimates vary hugely, but my guess (see my process) is that the deadliest catastrophes in recorded history (which took place over the span of three decades or less) are:42
It’s difficult to know what the effects of these events were, both because of the limited availability of data (especially prior to ~1700) and because the practice of counterfactual history is necessarily highly speculative in most cases.44 Still, we know from the charts above that none of them (including the Black Death45 ) seem to have been “transformative” in the sense discussed above. In fact, none of them seem to have had a lasting impact even on world population:
Here, a catastrophe like World War II — by my estimates, the deadliest event in world history (in absolute numbers) — shows up as a small kink in the upward curve, after which the previous trend continues unabated.46
One might suggest that some catastrophe averted a trajectory change that would have soon happened, if not for the catastrophe. Thus, one might argue, the charts we can produce from the data available to us look mostly flat until shortly after the industrial revolution, but this actually represents a major trajectory change from how they would have looked had the catastrophe not occurred. (I’ll call such a trajectory change an “invisible” trajectory change, since it does not appear as a sharp bend in the most important trend lines.)
I concede this is possible, but it seems very difficult to assess the plausibility of a proposed historical invisible trajectory change, and in general I am skeptical that any such “invisible trajectory changes” have occurred — including as a result of catastrophes that didn’t kill enough people, quickly enough, to make it on my list of deadliest catastrophes, such as the fall of the Western Roman Empire.47
Why? A great many things happened in human history prior to the industrial revolution, some of them seemingly momentous, and yet none of them seem to have produced a trajectory change, positive or negative, that was anywhere near the scale of the trajectory change that followed the industrial revolution. For example, suppose for a moment that the industrial revolution is, in fact, the primary cause of the trajectory changes discussed above. Historians are uncertain about (and disagree about) which factors were jointly needed to produce the industrial revolution,48 but they seem to agree that multiple factors were required, and this explains why the industrial revolution did not occur sooner. For example, in the Ancient Greco-Roman world, Archimedes and a few others practiced modern science (in the sense of experimentally testing predictions deduced from a precise mathematical model of some natural phenomenon),49 Heron invented a working steam engine,50 and an unknown inventor created an analogue computer capable of (among other things) adding and subtracting angular velocities,51 yet all this did not produce an industrial revolution, presumably because some other factors were not in place. For some historical disaster to have produced an invisible trajectory change, it needs to have been the case that all the other factors required for transformative change were in place, such that transformative change would have occurred if not for the disaster. From my (admittedly limited) readings of history, I am doubtful this is the case, albeit not with strong confidence.
From the above discussion, I draw the following tentative conclusions:
- By my estimate, the deadliest event before the industrial revolution (the Black Death) killed ~9.7% of world population, and the deadliest event after the industrial revolution (the 1918 flu pandemic) killed 3.3% of world population. Alternately, we might consider World War I and the 1918 flu pandemic a “double catastrophe”52 — since they occurred so close together in time and place, and the former exacerbated the latter — in which case their combined death toll was ~4.1% of world population.
- As far as I can tell from the limited available data, it doesn’t seem that any of the deadliest events in recorded history had a transformative negative impact (in my sense of “transformative”).
- As such, we don’t any historical precedents from which to estimate how many deaths might be required to produce a transformative negative impact. There is no precedent for a rapid loss of >10% of world population, and no precedent for a rapid loss of >5% of world population since the industrial revolution and the beginning of the positive trends humanity has experienced since then.53
Hence, as Nick Beckstead wrote previously:
Thus, past experience can provide little grounds for confidence that the positive trends [observed since the industrial revolution] would continue in the face of [a catastrophe of] unprecedented severity.
In this way, our situation seems analogous to the situation of someone who is caring for a sapling, has very limited experience with saplings, has no mechanistic understanding of how saplings work, and wants to ensure that nothing stops the sapling from becoming a great redwood. It would be hard for them to be confident that the sapling’s eventual long-term growth would be unaffected by unprecedented shocks — such as cutting off 40% of its branches or letting it go without water for 20% longer than it ever had before — even taken as given that such shocks wouldn’t directly/immediately result in its death. For similar reasons, it seems hard to be confident that humanity’s eventual long-term progress would be unaffected by a catastrophe that resulted in [an unprecedented number of deaths].
The question is, at what threshold of “unprecedented number of deaths” should we start to worry about transformative negative impact?
My instinct is to start with what we know from the deadliest events occurring after the industrial revolution. Perhaps the ~10% population losses produced by Genghis Khan (in the 13th century) and the Black Death (in the 14th century) do not tell us much about the plausible impact of a 10% loss of population today, because they occurred during the long, “flat” period of history dominated (in the long run, in most places) by a generally Malthusian equilibrium.54 Perhaps it is easier to derail and even “reverse” the sort of fast-moving, positive trajectory we’ve observed since the industrial revolution than it is to achieve a (worldwide) “escape” from a long, slow, generally Malthusian dynamic.55
On the other hand, even the double catastrophe of World War I and the immediately subsequent 1918 flu pandemic does not seem to have “come close” to having a transformative impact. How much deadlier than that would a catastrophe need to be, to have a serious chance of transformative impact?
Somewhat arbitrarily, I’d guess that I should start to seriously worry about the possibility of transformative impact from a loss of life (within three decades) that is ~4x as large as the double catastrophe mentioned above — i.e. a loss of 16.4% of world population. To avoid fake precision, I’ll round this to 15%. Today, this would be equal to a loss of ~1.1 billion people.56
Of course, transformative impact might result from a much smaller number of deaths, depending on how those deaths are distributed, and on other factors. For example, the emergence of global authoritarian rule wouldn’t need to kill a billion people in the space of a few decades to nevertheless bend the curves of human well-being enough to qualify as a “transformative” negative change.
What do these initial findings suggest about potential future transformative developments?
Overall, my initial findings lead me to the following tentative conclusions:
- When thinking about potential negative transformative developments, we have little evidence to draw from, since there has not been a single negative change of “transformative” magnitude in recorded history. Based on my brief reasoning here, I would begin to “seriously worry” about the possibility of transformative negative impact from an event that killed ≥15% of world population. Depending on other factors, I might also seriously worry about transformative negative impact from some events which result in a much smaller number of deaths.
- The gains in human well-being observed since the industrial revolution are vastly larger than pre-industrial fluctuations in human well-being. No other transitions in recorded history are remotely similar in magnitude.
- Hence, when thinking about which future developments might be most important, we should not forget that the sizes of their likely impact might differ by orders of magnitude.
I will illustrate point (3) with an example. A universal cure for cancer would be a huge benefit to human well-being, but the expected benefit to U.S. life expectancy from such a cure is only 2.83 years,57 and the average expected benefit in the rest of the world is even lower.58 Presumably the impacts on economic well-being would be of roughly comparable magnitude, and the impacts on the other key aspects of well-being discussed above would be even smaller, and perhaps negligible.
In contrast, consider the magnitude of the likely impacts on human well-being from the development of AI systems capable of automating most human labor, or the counterfactual benefits of preventing large-scale nuclear war. We have no precise estimates of the likely impacts of these potential future transformative developments (if they occurred), but they are very probably vastly greater in magnitude than the likely impact of a cure for cancer.
Of course, one must consider not only a cause’s importance but also its tractability and neglectedness. Perhaps all potential future transformative developments are so intractable and/or crowded that they are not worth investing in despite their extreme importance. Or perhaps they aren’t quite so intractable or crowded. I’d like to check.
Appendix A. Other important aspects of human well-being
Above, I chose the following as “key aspects” of human well-being: subjective well-being, physical health, economic well-being, empowerment via energy capture, empowerment via technology, empowerment via political freedom, and social well-being.
No doubt some readers will think my list misses some fairly fundamental aspects of human well-being. Below, I list some additional aspects of human well-being I considered including, and I say a bit about why I didn’t include them.
- Education, e.g. as measured by rates of literacy, primary education, or higher education, or by test scores, or by measures of educational mobility.59 In this report, I primarily treat education as an important input to what I consider to be “more fundamental” aspects of well-being — in particular, economic well-being, technological empowerment, and empowerment via political freedom — rather than as a key aspect of well-being itself. (See footnote for an argument.60 ) That said, a case could certainly be made that education is an important aspect of well-being even apart from its impact on economic well-being, technological empowerment, and political freedom.
- Empowerment via personal capabilities: Education can be seen as a special case of what we might call “empowerment via personal capabilities,” which is distinct from empowerment via energy capture, technology, and political freedom, and also distinct from the features of “being in good health” captured by the measures of physical health discussed above. Measures in this category could include measures of conscientiousness, work ethic, cognitive capabilities, various manual skills, and so on. Some of these do seem pretty important to me, but I suspect they are indirectly measured fairly well by measures of economic well-being and physical health (because those with greater capabilities tend to earn more and live longer), and we have better long-run data for those measures anyway.
- Work-related well-being, e.g. as measured by job satisfaction, leisure time, and career advancement.61 Personally, I think of work-related well-being as less fundamental than the aspects of well-being I chose to cover in this report. Work-related well-being is also partly assessed by measures of economic well-being.
Appendix B. Some methodological details
Where my death toll estimates come from
Here, I summarize how I constructed the death toll estimates used in my section on deadliest catastrophes.
Initially, I had hoped to simply pull death toll estimates from Wikipedia. Unfortunately, I found Wikipedia to be an unreliable source for death toll estimates. For example, when I checked Wikipedia on March 15, 2017:
- Wikipedia’s List of wars and anthropogenic disasters by death toll provided “lowest” and “highest” estimates for World War II of 65 million and 85 million, respectively. However, I quickly found that in Matthew White’s collection of historians’ estimates of the death toll of World War II, most estimates were lower than 55 million, with the median estimate at 50 million.
- The first sentence of Wikipedia’s Black Death article claimed a death toll of “75 to 200 million people,” citing three news articles62 which themselves provide no sources for their estimates.63 Searches for academic literature64 turned up many academic sources using the 200 million figure, but nearly all of them (that I checked) made clear that the 200 million figure sums deaths across multiple (usually, “three”) different pandemics caused by Yersinia pestis, occurring centuries apart.65 Moreover, for the academic sources providing a citation for the 200 million figure that I was able to track down,66 the reference trail in all cases led eventually to a single source: an article from the May 1988 issue of National Geographic by photojournalist Nicole Duplaix, titled “Fleas: The Leathal Leapers” (pp. 672-694). That article, too, is clear that the 200 million figure refers not to deaths for the “Black Death” of the 1340s-50s, but to deaths summed across multiple pandemics.67 Moreover, this number is highly suspect, given that Duplaix provides no source or analysis for it.
Thus, I began to estimate historical death tolls myself, drawing from credible-looking scholarly primary sources wherever possible. Fortunately, I soon noticed that the estimates I came up with after a bit of research tended to closely match the estimates provided in Matthew White’s book The Great Big Book of Horrible Things: The Definitive Chronicle of History’s 100 Worst Atrocities (2011).68 Hence, I decided to use White’s estimates as a starting point, and then, for each estimate, I checked his reasoning and the apparent credibility of his key sources.
Unfortunately, White’s book only covers anthropogenic disasters, a category which excludes pandemics, which account for several of the deadliest catastrophes in history. Thus, I constructed pandemic death toll estimates myself.
1918 flu pandemic
For the 1918 flu pandemic, the most credible-looking estimate of the death toll I’ve seen is Johnson & Mueller (2002), which seems to be the primary source for widely-cited estimates of “50-100 million deaths” due to these statements:
Further research has seen the consistent upward revision of the estimated global mortality of the [1918 flu] pandemic, which a 1920s calculation put in the vicinity of 21.5 million. A 1991 paper revised the mortality as being in the range 24.7–39.3 million. This paper suggests that it was of the order of 50 million. However, it must be acknowledged that even this vast figure may be substantially lower than the real toll, perhaps as much as 100 percent understated.
There are vast areas of the world for which we have no or little information, and often what information we do have is of dubious quality and contradictory. Sometimes the data cover only certain cities or populations; often the indigenous mortality has never been considered. Sometimes the figures given are only those that were recorded as influenza deaths; at other times, they are influenza and pneumonia deaths. Consequently, the real pandemic mortality may fall in the range of 50 to 100 million, but it would seem unlikely that a truly accurate figure can ever be calculated.
In other words, their estimate from published records is 50 million deaths (actually 48.8 million deaths; see Table 5), but they speculate that the true number may be as high as 100 million deaths.
To construct my estimate, I struck a balance between empirical estimates and speculation, relying more on the former than the latter, and (somewhat arbitrarily) guessed that the 1918 flu pandemic killed ~60 million people. (Purposely, this is substantially less than the geometric mean of 48.8 million and 100 million, which is 69.9 million.)
Plague of Justinian
Wikipedia’s article on the Plague of Justinian claims a death toll of 25-50 million “in two centuries of recurrence,” but my focus is on the first (and deadliest) wave of the plague, from 541-544.69
After checking many scholarly sources on the Plague of Justinian,70 I was not able to find a credible scholarly estimate of the death toll from 541-544. Thus, I contacted a scholar who wrote a recent review article and book on this topic, Dionysios Stathakopoulos of King’s College London, and asked how he would recommend I produce a (very rough) estimate of the death toll. Dr. Stathakopoulos suggested I could use the rough mortality rate he found plausible for Constantinople in Stathakopoulos (2004) (20%),71 and multiply that by the rough estimate of the population of the empire at the time that he came to in Stathakopoulos (2008) (28 million).72 This suggests a (very rough) death toll estimate of ~5,600,000.
Plague in the 14th century may have arisen as early as the 1330s, and it recurred in numerous waves across multiple centuries, but for this report I’ll use “Black Death” to refer to its first (and deadliest) proven wave in the 14th century, from 1345-1353.73
I consulted several scholarly sources on the Black Death,74 and my impressions are that:
- The most common view seems to be that about 1/3 of Europe perished in the Black Death, starting from a population of 75-80 million. However, the range of credible-looking estimates is 25%-60%.
- The only credible-looking estimates of the death toll in the Islamic world I’ve found are from Dols (1977). For Egypt, Dols estimates a loss of 25%-33% from a starting population of 4.2-8.0 million, and for Syria he estimates a 33% loss from a starting population of 1.2 million.75 The Black Death also reached some other areas of the Middle East and North Africa, for example Barca in Libya and Tunis in Tunisia, but I haven’t seen any attempts to estimate the death tolls outside Egypt and Syria.
- Some sources claim large death tolls in Asia,76 but as far as I can tell, this is mere speculation, and not substantiated by any hard evidence to date.77
To generate a very rough death toll estimate for this report, I took the geometric mean of my “lowest plausible estimate” and my “highest plausible estimate.” My lowest plausible estimate is:
- For Europe, perhaps 25% of people died, from a starting population of 75 million.
- For the Middle East and North Africa, perhaps Egypt lost 25% from a starting population of 4.2 million, Syria lost 33% from a starting population of 1.2 million, and the rest of the Middle East and North Africa lost 33% from a population of (let’s say) 2 million from the most populated areas.
- In Asia, perhaps the Black Death killed almost no one.
My highest plausible estimate is:
- For Europe, perhaps 60% of people died, from a starting population of 80 million.
- For the Middle East and North Africa, perhaps Egypt lost 33% from a starting population of 8 million, Syria lost 33% from a starting population of 1.2 million, and the rest of the Middle East and North Africa lost 33% from a population of (let’s say) 6 million from the most populated areas.
- In Asia, perhaps the Black Death killed (let’s say) 15 million, even though we have no direct evidence of this.
Thus, my lowest plausible estimate adds up to 20,856,000, and my highest plausible estimate adds up to 68,016,000. The geometric mean of these two numbers is 37,666,533.
My incomplete attempt to generate worldwide slavery estimates
Above, I mention that percent of people not enslaved would be perhaps the best measure of empowerment via political freedom, if only we had the data. Below, I recount my unfinished attempt to generate highly speculative estimates of the percent of people not enslaved over time.
While studying the history of slavery, my impression of how well the curve for “percent of people not enslaved” would mirror those charted above flipped back and forth multiple times.
In brief: Iceland was the first state to abolish slavery, in 1117. But progress on abolition continued to be extremely slow until about 1775, during the industrial revolution. Then, the number of states abolishing slavery rose quickly up to 1981, when Mauritania became the last country on Earth to abolish slavery. Unfortunately, slavery continues (illegally) in several countries, but the proportion of people enslaved during the past few decades is probably lower than at any other time in recorded history.79
At first glance, Pinker’s chart seemed to mirror the combined chart above surprisingly well. However, I quickly grew suspicious of this initial judgment. First, the inflection point in Pinker’s chart appears early in the industrial revolution rather than after 1800. Second, my guess was that by the time a political jurisdiction abolished slavery, it was probably already a fairly unpopular practice in that jurisdiction anyway — suggesting that if we could chart “proportion of world population enslaved” over time, the trajectory change would appear long before the industrial revolution. One of the first histories of slavery I read seemed to confirm this (Payne 2004, ch. 13):
Serfdom as an alternative to slavery began to appear in Italy in the late years of the Roman Empire (a.d. 400– 500)… In France, serfdom had almost entirely replaced slavery by around 1200. In England, slavery rapidly disappeared after the Norman Conquest of 1066. Up until that time, slavery was common, with approximately 10 percent of the population being slaves. The Norman invaders, who considered France their home, freed slaves on the English estates they took over and made them serfs “to avoid the necessity of close supervision and management of slaves.” The economic pressures against slavery meant that this practice had largely died out in Europe by the late Middle Ages. It had also faded away in other places as well. In Japan, for example, it had become rare by the year 1200.
Thus, I concluded that slavery’s primary trajectory change probably came long before the industrial revolution.
However, as I continued to study the history of slavery, I learned several things that made me wonder once again whether slavery’s primary trajectory change actually did coincide quite well with the other trajectory changes charted above.
- Actually, slavery was still quite popular in most jurisdictions at the time it was abolished; northern Europe seems to have been the exception here, not the rule.80
- Even as slavery declined (in favor of serfdom) in northern Europe during the medieval period, it flourished and perhaps even grew in the Islamic world and perhaps elsewhere. Hence “proportion of world population enslaved” might not have declined much during the medieval period.81
- Arguably, many forms of serfdom ought to be counted as not very different than slavery from a “political freedom” perspective; if so, then perhaps political freedom in this sense did not increase much during the medieval period, even in Europe.
- The global antislavery movement more-or-less began with the British abolitionists, circa 1800-1840 — suspiciously, right at the end of the industrial revolution, in the home country of the industrial revolution. Britain was thereafter the primary global exporter of antislavery, sailing around the world and pressuring countries all across Africa and Asia to abolish the slave trade and slavery in general. Most slavery abolition around the world seems not to have been “home-grown” (as in Britain and the USA), but in fact was exported by Britain and (later) some other European states to their colonies and the rest of the world.82
Unfortunately, I could not afford the time to resolve the question. In the absence of an initial estimate for “percent of people not enslaved” over time, I instead list the major sources I consulted, and explain my unfinished spreadsheet of regional slavery estimates.
General sources on the history of slavery
The relatively general sources on the history of slavery I consulted were:83
- Patterson, Slavery and Social Death (1982)
- Davis, Slavery and Human Progress (1984)
- Rodriguez, Chronology of World Slavery (1999)
- Campbell (ed.), Structure of Slavery in Indian Ocean Africa and Asia (2003)
- Payne, A History of Force (2004), ch. 13
- Campbell (ed.), Abolition and its Aftermath in Indian Ocean Africa and Asia (2005)
- Hinks & McKivigan (eds.), Encyclopedia of Antislavery and Abolition (2006), in particular the entries categorized under “Antislavery and Abolition in Specific Countries and Regions” in the “Guide to Related Topics,” plus the entries on “Classical Greek Antislavery,” “Classical Rome and Antislavery,” and “Slavery and Abolition in the Twentieth Century”
- Davis, Inhuman Bondage (2006)
- Grayling, Toward the Light of Liberty (2007)
- Campbell et al. (eds.), Women and Slavery, Volume One (2007)
- Walvin, A Short History of Slavery (2007)
- Dal Lago & Katsari (eds.), Slave Systems: Ancient and Modern (2008)
- Drescher, Abolition: A History of Slavery and Antislavery (2009)
- Bradley & Cartledge (eds.), The Cambridge World History of Slavery, Volume 1: The Ancient Mediterranean World (2011)
- Bodley, Cultural Anthropology: Tribes, States, and the Global System, 5th edition (2011), ch. 5, for its discussion of slavery in tribal societies (drawing from Gray 1999)
- Heuman & Burnard (eds.), The Routledge History of Slavery (2011)
- Hunt, “Slavery” (2015)
Some especially informative “overview” passages from some of these sources are provided in a footnote.84
How I built my slavery spreadsheet
My plan for constructing an initial (and highly speculative) estimate of “percent of people not enslaved” over time was the following:
- Collect every scholarly estimate I could find for the proportion of people enslaved in a given region during a given time.
- Combine those estimates with a database of estimates for the total population of different regions at different times.
- For the region × time interval cells for which I didn’t have a “percent enslaved” estimate (the vast majority of all cells), simply guess that the rate of slavery then and there may have been similar to that of the most similar time and place for which I do have a “percent enslaved” estimate. When making these guesses, prioritize making guesses for the most populous areas (at a given time).
- Once I have gathered estimates or guesses for regions accounting for (say) 40% of global population at each time interval, make broader, quicker guesses for the rates of slavery in the remaining regions.
- Use these estimates and guesses to calculate an initial guess for “percent enslaved” at each time interval.
- Show the resulting curve and spreadsheet to experts in the world history of slavery, and learn which guesses they most strongly disagree with, and adjust accordingly.
Naturally, the resulting curve would be extremely speculative, but it might qualify as “the best guess I could make, if I had to guess.”
I collected many slavery and population estimates in a spreadsheet here, but I did not come close to executing the full plan.