by Brian Tomasik
First written: 11 Feb. 2016; last update: 11 May 2017


While it's important to assess the impact on wild-animal suffering of particular environmental choices, there's also value in zooming out and trying to estimate the net impact that all of humanity's activities together exert on suffering in nature. One reason is that this informs policy questions about family planning and life extension. Another reason is that this information can help us judge whether a given personal environmental impact is a large or small portion of a person's total environmental impact.

Based on defaunation studies, it appears that human activity over the last 40 years alone has reduced both vertebrate and invertebrate populations on balance, although these studies may overstate the extent of decline(?). A plausible estimate is that the average person on Earth prevents ~1.4 * 107 insect-years by his/her environmental impact each year.

Given the possible methodological issues with these numbers, further exploration into this topic is warranted, especially to ensure that the sign of net impact is correct. For example, this analysis argues that it's unclear whether humanity's existence has actually decreased sentience-weighted global animal populations on balance.

Note: I haven't yet read all the details behind the studies discussed here, so my critiques of them should be regarded as only provisional right now.

Net impact on vertebrates

Living Planet Index

The World Wide Fund for Nature (WWF) reported in its 2014 "Living Planet Report" "that animal populations are roughly half the size they were [...] 40 years ago." The following figure shows the decline in the Living Planet Index (LPI):

This page says:

The Global Living Planet Index is the aggregate of three equally-weighted indices of vertebrate populations from terrestrial, freshwater and marine systems. All three system LPIs also show a decline – terrestrial and marine both at 39% whereas freshwater systems are faring worse with a decline of 76% since 1970.

These numbers should be taken with some caution. Just as the consumer price index can overstate inflation because it uses a fixed basket of goods, so too I assume(??) that the LPI overstates the decline of vertebrates because even if the measured species are declining, some may be taking their place. If the species put into the index tend to be at their peak at the beginning of the measured period, then regression to the mean suggests there should always be something of a downward trend on average. The new species that are rising in numbers may not be counted in the LPI, which means the LPI might overstate the extent of decline in total animal numerosity? This supposition seems reinforced by the fact that the LPI page lists "Invasive species/genes" as being ~5.1% of the "Primary threats to populations in the LPI." In other words, it appears that the LPI ignores increases in populations of the invading species?

This bias in the LPI might not be strong on short time scales, but at least in the very long run, it should apply. For instance, if we looked at a fixed basket of species from 500 million years ago, an abundance index would show a severe decline because perhaps 99.9% of species that existed in the past are now extinct. That doesn't mean the abundance of all animals is lower now than 500 million years ago.

Finally, it may be that conservationists bias their studies of animal populations toward species that are threatened or endangered. These data points may constitute a disproportionate share of all the data points used to calculate the LPI. If so, the LPI will be biased to show more population declines than increases.

Dirzo et al. (2014)

Dirzo et al. (2014) reported a slightly less dramatic number than the 2014 WWF report did: "vertebrate data indicate a mean decline of 28% in number of individuals across species in the past four decades (fig. S1, A and B) (14, 21, 22)". However, one reason this is lower is that this estimate cites the older 2012 version of the "Living Planet Report", which showed only a 30% decrease in the LPI. I assume that the difference from the 2012 to 2014 reports was a change in methodology.a

Dirzo et al. (2014) explain that larger-bodied vertebrates are more vulnerable to extinction (p. 403).b Assuming they also have bigger population declines, then indices of vertebrate abundance may decline more than total vertebrate populations decline? For example, suppose there are just two animals: mice and elephants. Suppose that elephants decline 80%, while mice don't decline at all. A naive index of abundance would decline by (0% + 80%)/2 = 40%. But if mice are, say, 1000 times more numerous than elephants, then the actual decline is from a population size proportional to 1000 + 1 to a population size proportional to 1000 + 0.2 -- i.e., hardly any decline at all.

This page says: "small-bodied species normally compete with large-bodied vertebrates for food and other resources. As an area becomes defaunated, dominant small-bodied species take over, crowding out other similar species and leading to an overall reduced species diversity.[21]" So it's possible that defaunation in some cases actually increases total animal abundance, by increasing populations of smaller-bodied species a lot, as larger-bodied animals become less numerous.

Comparison with farm animals

Humanity's net impact on vertebrate suffering overall becomes murkier when we consider farm animals -- at least when we focus only on mammals and birds.

Suppose there are, say, 500 billion wild mammals and birds in total. This is a rough estimate that might be off by an order of magnitude. And suppose humanity has decreased wild-vertebrate populations by, conservatively, ~10%. Then humanity prevents ~50 billion wild mammals and birds from existing.

There are about 24 billion farm animals at any given time worldwide. However, many farm animals plausibly suffer more than most wild animals, maybe a few times more. Suppose a farm animal suffers ~2 times more than a wild animal per unit time.c Then humans create an amount of farmed-animal suffering equivalent to 2 * (24 billion) = 48 billion wild animals, which about cancels out the reduction in wild-animal numbers.

Plus, farmed mammals and birds are, on average, bigger than wild mammals and birds. As a result, the raw biomass of farm mammals exceeds that of wild land mammals. So if one assigns moral importance to animals in proportion to their mass, then suffering of mammals and birds on farms probably outweighs the suffering of wild mammals and birds that humans prevent by encroaching on habitats.

However, adding reptiles, amphibians, and fish into the calculation would increase the importance of wild animals. And if we also count invertebrates, then human reductions in wild-animal suffering probably outweigh the harm of factory farming, even if one only counts invertebrates in proportion to their biomass. Human reductions in invertebrate numbers are discussed next.

Net impact on insects

Dirzo et al. (2014)

Dirzo et al. (2014) reports that "Globally, a compiled index of all invertebrate population declines over the past 40 years shows an overall 45% decline". UCL News summarized the finding with this headline: "Invertebrate numbers nearly halve as human population doubles".

I think(?) this estimate has the same problem of potentially overstating declines as was discussed in the case of vertebrates.

Vogel (2017)

Vogel (2017) reports on observations by the Krefeld Entomological Society that "The mass of insects collected by monitoring traps in the Orbroicher Bruch nature reserve in northwest Germany dropped by 78% in 24 years", from 1989 to 2013. Moreover: "Through more direct comparisons, the group—which had preserved thousands of samples over 3 decades—found dramatic declines across more than a dozen other sites." However, it should be remembered that "No one knows how broadly representative the data are of trends elsewhere."

My reading of Vogel (2017) is that (all of?) these insects were caught using Malaise traps, which "collect mainly species that fly a meter or so above the ground." Is it possible that populations of smaller, soil-dwelling invertebrates like mites and springtails have remained stable or increased? Vogel (2017) mentions that one possible cause of insect decline is that "Increased fertilizer on remaining grazing lands favors grasses over the diverse wildflowers that many insects prefer." This change would presumably reduce numbers of flying insects that feed on flowers, but unless total plant productivity is changed, it wouldn't necessarily reduce populations of detritivorous invertebrates?

Vogel (2017) reports on more data from the UK:

Since 1968, scientists at Rothamsted Research, an agricultural research center in Harpenden, U.K., have operated a system of suction traps—12-meter-long suction tubes pointing skyward. Set up in fields to monitor agricultural pests, the traps capture all manner of insects that happen to fly over them; they are "effectively upside-down Hoovers running 24/7, continually sampling the air for migrating insects," says James Bell, who heads the Rothamsted Insect Survey.

Between 1970 and 2002, the biomass caught in the traps in southern England did not decline significantly. Catches in southern Scotland, however, declined by more than two-thirds during the same period. Bell notes that overall numbers in Scotland were much higher at the start of the study. "It might be that much of the [insect] abundance in southern England had already been lost" by 1970, he says, after the dramatic postwar changes in agriculture and land use.

Once again, these traps only seem to be sampling flying insects?

Vogel (2017) quotes Scott Black: "So lots of insects will do great, but the insects that we love may not." I assume that many of the beloved insects are flying ones. I'd be interested in seeing more study of soil invertebrates (mites, springtails, nematodes, etc.) in addition to charismatic flying insects.

Per-capita insect impact

To be cautious about the problems with the above numbers, let's suppose that the actual decline in global insect populations has only been ~10%. Given ~1018 insects on the planet, a 10% decline in insect populations worldwide averts 1018 * 0.1 = 1017 insect-years per year.d Since there are 7 * 109 humans, the per-capita reduction in insect abundance is about (1017 insect-years per year) / (7 * 109 human-years per year) = 1.4 * 107 insect-years per human-year. Since this calculation is based on the invertebrate decline over only the last 40 years, not since humanity first appeared, it may be an underestimate. The calculation implicitly assumes that a single year of the human population's existence causes a single year of invertebrate population decline.

Comparison with farm animals

Most farmed animals are poultry. If we add up the livestock counts in this table from "Chickens" to "Turkeys", humans farm about ~20 billion birds (and rodents) per year. This section estimates that a chicken has ~2500 times more neurons than an insect. Since some bugs can be quite small, let's round that up to 104 times more neurons. Suppose we weigh animals linearly in neuron count. And suppose we think a chicken-year of suffering is, generously, 5 times as bad as an insect-year of suffering, not counting brain-size differences. Then humans create an amount of chicken suffering equivalent to the suffering of this many wild insects: (20 billion) * 5 * 104 = 1015. But humans prevent ~1017 wild insects. So the reduction in wild-insect suffering seems to dominate, even given assumptions favorable toward a high amount of chicken suffering.

What if we also count cattle, sheep, etc., which have bigger brains than chickens? If we add up the livestock counts in this table from "Asses" to "Sheep", we get ~4.5 billion. This section suggests that cattle matter ~105 times as much as bugs using pure neuron-count weighting. Generously assume that the same holds true for sheep, goats, etc. (even though they're smaller than cattle). Again using a generous 5-to-1 ratio for farm vs. wild suffering per year, the farming of non-poultry livestock creates an amount of suffering equivalent to this population of wild insects: (4.5 billion) * 5 * 105 = 2.3 * 1015, which is pretty close to the number for chickens in the previous paragraph. Even adding this to the chicken number, the total is quite a bit lower than 1017 for wild insects.

Comparison by body mass

However, if we weigh animals by body mass, not neuron count, then it becomes less clear if reductions in invertebrate abundance can outweigh animal farming. Consider just cattle, which number about 1.4 billion.

In the US, a typical cattle live weight at slaughter is around 700-900 pounds. In developing countries, weights are generally lower. And the average weight of a cow or steer over its lifetime should be something like half the slaughter weight. So say the world-average mass of cattle is, umm, ~300 pounds, or 140 kg.

In contrast, a typical bug has a mass of ~3 mg. So a single cow or steer is about (1.4 * 105 g) / (3 * 10-3 g) = 5 * 107 times more massive.

Assuming cattle and wild insects suffer about the same amount per year ignoring body-size differences, the suffering that humans impose on cattle is equivalent to this many wild insects: (1.4 billion) * (5 * 107) = 7 * 1016. That's basically equal to the (conservative) 1017 estimate for how much humans reduce wild-insect numbers. And this calculation doesn't include sheep, pigs, chickens, etc. So someone who weighs animals strictly by body mass may be more concerned about humanity's overall impact on animal suffering. (However, I personally think insects deserve more moral weight than their proportional number of grams of body mass, so I disagree with this conclusion.)

Why I'm cautious about taking defaunation studies at face value

Above I expressed some concerns with the LPI's methodology. A further small concern is that conservationists generally have incentive to exaggerate (what they consider to be) environmental problems in order to make people more motivated to take action.

However, the biggest reason I'm skeptical about the magnitude of fauna declines expressed in the LPI and in Dirzo et al. (2014) is that declines in net primary productivity (NPP) have not been as severe -- see this piece for details.

Other approaches to the question

Forest loss

This page reports: "Forests cover 31 percent of the world’s land surface, just over 4 billion hectares. [...] This is down from the pre-industrial area of 5.9 billion hectares." This means that before industrialization, forests covered (31%) * (5.9/4) = 46% of land area.

Let the average density of bugs per hectare on non-forest land be D. Suppose that on forested land, this density is, say, 1.5 * D, given that forests are generally more productive. Let N be the total hectares of land on Earth. Then by reducing forest cover from 46% to 31%, humans would have reduced global bug populations from N * (0.46 * 1.5 * D + 0.54 * D) to N * (0.31 * 1.5 * D + 0.69 * D). That's a decrease from 1.23 * N * D to 1.155 * N * D, which is a reduction of 6%, or slightly more than half of the ~10% estimated earlier in this piece. Given that this calculation only considers forest loss (with some pretty conservative input parameters), I wouldn't be surprised if humanity's entire impact was to reduce bug populations by at least 10%, although some human activities increase bug populations.

Net primary productivity

This study estimates that anthropogenic land-cover changes have reduced worldwide net primary productivity (which we might take as a proxy for secondary productivity in the form of herbivores and other heterotrophs) by ~5%.

Global bird decline

Gaston et al. (2003) estimate that by 1990, humans had reduced the global population of breeding birds by "between a fifth and a quarter of pre-agricultural bird numbers" (p. 1293). Moreover (p. 1298):

The estimated losses of individuals are likely, however, to be conservative, as the activities of indigenous peoples are likely to have affected even otherwise apparently ‘pristine’ habitats (Grayson 2001), extensive areas of several natural land-use types may have a substantially lowered carrying capacity (e.g. through small-scale habitat fragmentation and degradation, the replacement of natural forest area with plantations: Pimm (2001) and see references in § 1), whereas the mainly migratory breeding populations of relatively pristine high latitude regions (Newton & Dale 1996a,b) may be affected by the degradation of wintering sites at lower latitudes (Faaborg 2003).

Considerations regarding technological growth

The numbers in this piece help inform questions about whether we should, ceteris paribus, prefer a bigger or smaller human population. However, another consideration relevant to that question is the net impact of technology. In general, a bigger human population means that technology advances faster, as do philosophy, science, and other domains of human activity.

If we simplistically assume that there's a fixed number of human-years that will exist prior to the emergence of artificial general intelligence (AGI) because more human-years means proportionally faster AGI development, then changing the size of the human population at any given time wouldn't change the total number of human-years that will ever exist, since it would just mean that more human activity happens faster rather than slower.

Of course, in practice, developed countries tend to produce technology at a disproportionate rate. So maybe saving lives in Africa has less impact in terms of speeding up AGI, even relative to the environmental impact of those lives. (For instance, to make up sample numbers, maybe one extra person in Africa speeds up AGI by 1/20 as much as one person in the US, while the environmental impact of a person in Africa is 1/2 as much as for a person in the US?) On the other hand, if Africa eventually becomes developed and starts contributing more to AGI, then speeding up the date by which that happens could matter nontrivially.

Also, if AGI mostly destroys the biosphere after it emerges, then speeding up AGI would reduce the total number of years during which nature exists. On the other hand, AGI will also multiply suffering in other ways, so faster emergence of AGI is plausibly net bad. A bigger human population also has side effects on how hard international cooperation is, how robust civilization will be, and so on.

In the I = PAT formula for the impact of humanity on the environment, all three terms have nontrivial side effects on the far future:

  • "P" (population) can indirectly speed up economic and technological development, as discussed above.
  • "A" (affluence), i.e., GDP per capita, has effects on people's moral views, rates of violence, etc.
  • "T" (technology) improvements are direct speed-ups in technological progress, many of which may cause AGI to happen sooner than it otherwise would have.

That said, it seems as though many of the kinds of human economic impact on the environment that are most malleable also tend to have the fewest side far-future effects -- e.g., using land less efficiently has important impacts on wild-animal suffering but may not dramatically change economic-growth trajectories. (Less efficient land use might increase technological progress by reducing hindrances to growth, or it might slow long-term technological progress by degrading natural capital. It's not at all clear.)

Even before AGI emerges, faster technology may have good and bad impacts on wild-animal suffering. While historically human technology has tended to allow for more extensive habitat loss, in the most advanced economies, technology often reduces environmental impact (e.g., alternative energy, less use of paper, more efficient transportation algorithms, in vitro meat). Since most people want to preserve the environment, and since technology generally allows people to achieve their goals more effectively, we might expect that in the long run, technology will reduce per-capita environmental impact (at least until the dominant intelligence on Earth becomes digital), although since technology also allows for a greater total quantity of resource consumption, the opposite might also be true. It would help to do a deeper analysis of the net impact on wild-animal suffering of speeding up technology, focused on specific emerging technologies.

How the simulation hypothesis affects evaluation of speeding up AGI

The simulation hypothesis suggests that there may be one or many copies of us in simulated realities, instead of or in addition to copies of us in a "real" (non-simulated) Earth. Such simulations may be run for reasons of science, entertainment, or intrinsic value.

If the simulation hypothesis holds, then your short-term impacts on wild-animal suffering may matter more in total than if it doesn't hold. That's because if there are lots of simulations of you, there are more copies of you, so any action you (i.e., the collection of your copies) take would be replicated many times over. Of course, if you're in a simulation, animals in nature may not be simulated in full detail. But if the simulation is to be accurate enough, the simulators might indeed be simulating wild-animal minds to an intermediate or high level of resolution, so reducing wild-animal suffering in a simulation would still be important in expectation.

This section considers a few scenarios for how the simulation hypothesis might affect the ways in which speeding up AGI would increase or decrease total suffering.

Scenario 1: Simulation only includes Earth, fixed number of sims

Simulating the salient parts of Earth is computationally expensive, but simulating the salient parts of a superintelligent colonization wave is much more computationally expensive. So it could be that our simulation focuses on Earth prior to large-scale space colonization. If our simulation cuts off whenever Earth-originating intelligence becomes smart enough to begin a major space-colonization effort, what would this imply for the relationship between AGI speed and wild-animal suffering?

The following figures illustrate two scenarios: one with faster AGI development due to a larger human population and one with slower AGI development due to a smaller human population. The y axis represents the total amount of suffering on Earth. In the beginning, this suffering is mainly in the form of wild animals, while toward the end, this suffering comes mainly from increasingly complex and numerous artificial minds or subprocesses.

The figure shows that a bigger human population implies less total suffering (area under the curve) because a world with a bigger human population

  • has less wild-animal suffering, because more people are displacing wild animals at any given time
  • produces AGI sooner, which ends the simulation more quickly.

Scenario 2: Simulation only includes Earth, fixed computing resources

Rather than simulating a fixed number of copies of Earth, the simulators may have a fixed budget of computing resources. Thus, if a simulation finishes more quickly, maybe more copies or close variations of it can be run. In this case, ending the simulation sooner doesn't necessarily reduce suffering, because a new simulation will be spawned off as soon as the old one ends.

If we assume that the computational cost of a simulation is proportional to the number of Earth-years for which it lasts, then we might have the following graphs for larger vs. smaller human populations, where it's now possible to run two copies of the version of Earth that has a bigger population because each such simulation ends sooner:

Scenario 3: Simulation includes galactic colonization

Suppose the simulators don't stop the simulation when space colonization begins but rather allow it to continue some time thereafter. The amount of suffering in the post-AGI world grows immense because more and more planets are being harvested for increasingly efficient intelligent computation. Assuming both simulations end after a fixed number of years, we'd have the following comparison:

Of course, it requires more computing power to run the high-population simulation in this case, so perhaps the simulators can't run as many of these given limitations on computing resources.

If the computing cost of each simulation were directly proportional the complexity-weighted number of mind-years (and hence roughly the amount of suffering) that it contained, then if our simulators had a fixed computing budget, we wouldn't be able to change the total amount of suffering much at all, except to improve the quality of life of whatever organisms existed.


Edward Miller pointed me to the "Living Planet Report". This Facebook post inspired my sections comparing wild- vs. farmed-animal suffering. Ozy Brennan pointed out an error, which I then fixed.


  1. This page says:

    The method has recently been adapted with a new weighting procedure to give a better representation of global vertebrate diversity and to correct for a bias towards well studied species from Europe and North America. The result is a steeper decline than in other versions of the LPI as a result of placing more weight on highly diverse regions and species groups which, on average, are declining faster.

    I can't tell if it's talking about the 2014 LPI or a previous version.  (back)

  2. This page echoes the point:

    Large mammals are often more vulnerable to extinction than smaller animals because they require larger home ranges and thus are more prone to suffer the effects of deforestation. Large species such as elephants, rhinoceroses, large primates, tapirs and peccaries are the first animals to disappear in fragmented rainforests.[10]


  3. It's plausible that farm animals suffer more than 2 times as much as wild animals during life, but perhaps many wild-animal deaths are at least as bad as farm-animal deaths? Another consideration is that farmed chickens have relatively short lifespans, so the proportion of their lives comprised of painful slaughter is pretty high. On the other hand, early mortality rates for wild birds are also pretty high.  (back)
  4. Pedantically, if the current bug population is 1018, then the original bug population was 1018/0.9 = 1.11 * 1018, giving a decline of 1.1 * 1017 insect-years per year. But I think the original 1018 estimate was made decades ago, and it has so much error in it anyway that I mainly just care about the order of magnitude.  (back)