Alexey Turchin, 

Anonymous coauthor 1 

Anonymous coauthor 2


 TL;DR: Because of survival bias we underestimate the location of climate tipping points and thus our world is more fragile to small anthropogenic actions. Therefore, human extinction because of runaway global warming is more probable.


Abstract: Humanity may underestimate the rate of natural global catastrophes because of the survival bias (“anthropic shadow”). But the resulting reduction of the Earth’s future habitability duration is not very large in most plausible cases (1-2 orders of magnitude) and thus it looks like we still have at least millions of years. 

However, anthropic shadow implies anthropic fragility: we are more likely to live in a world where a sterilizing catastrophe is long overdue and could be triggered by unexpectedly small human actions. In the same way, an over-inflated toy balloon, which will soon burst, is very fragile.

Anthropic fragility can manifest itself in the higher chances of runaway global warming. It has often been suggested that the Earth's atmosphere remained life-supporting for billions of years by sheer chance. Therefore, the survival bias can be strong. It is also known that Earth-like water worlds could experience transitions into deadly moisture greenhouse (mean T = 65C). This means that relatively small anthropogenic actions could put the climate above an unpredictable tipping point, which could lead to the moisture greenhouse. Thus, it is necessary to carry out urgent geoengineering studies and prepare to prevent an unexpected climate catastrophe.

There are three main counterarguments against the existence of the anthropic shadow: self-indication assumption (SIA), past observers and the Gaia hypothesis; we show that they fail. It was proposed that SIA exactly compensates the anthropic shadow as an observer unlikely to find herself in a world with a strong anthropic shadow; however, there is a baseline level of the anthropic shadow for all habitable planets, similar to the rate of evolutionary transitions like abiogenesis. There are no “past observers” as qualified observers appeared only 50 years ago. Gaia hypothesis assumes existence of self-stabilizing feedback in climate, but new types of events like quick CO2 growth could override its coping ability.

We present a list of other catastrophes that may have been underestimated because of the anthropic shadow, including collider catastrophes, nuclear war and even an alien invasion. 

We also hypothesised that human intelligence is more likely to emerge in an unstable world that is nearing its end and thus we get a new form of Doomsday argument.

pdf with Appendix:



· The decrease of the expected habitability time of the Earth’s because of the survival 

bias (anthropic shadow) is not very large and thus it looks like that we still have at least millions of years. 

· But anthropic shadow implies anthropic fragility: we are more likely to live in a world where a catastrophe is long overdue and could be triggered by unexpectedly small human actions.

· Catastrophic runaway global warming may be an example of such anthropic fragility, and we cannot predict where the tipping point is.

· We should research urgent geoengineering to counter the unpredictable anthropic fragility of climate.

· We also suggested a hypothesis that human intelligence is more likely to arise in an unstable world that is nearing its end.


1. Introduction

One formulation of the anthropic principle (Carter, 1974) is that observers can exist only in those worlds in which there are no conditions that prevent the appearance of the observers. Such conditions may take the form of fine-tuning of the initial parameters of the universe – or the absence of life-ending catastrophes. Circovic et al called the lack of life-ending catastrophe in the past “anthropic shadow”, as it causes underestimation of the background rate of catastrophes.

The goal of this article is to research how anthropic shadow affects the fragility of our world, in the sense of Bostrom’s vulnerable world hypothesis (2018), especially relative to the risk of runaway global warming

Our central argument and illustration are presented in section 2. We explore the idea of anthropic fragility and how it could work in different types of catastrophes, first of all, for global warming in section 3. We shortly discuss what types of geoengineering may be needed to counter unexpected rapid global warming. We will create an overview of all possible types of catastrophes where anthropic shadow and anthropic fragility can manifest themselves in table 1 in section 4. Then we will go into counterarguments that were suggested against the anthropic shadow and will demonstrate that they can’t completely disprove it, but they limit its power. We will explore what kind of evidence of anthropic shadow in the past of Earth we have in section 5. In the end, we will take more a general overview of the problem and of its connection with x-risks. We will suggest a new form of the Doomsday argument: that intelligent life is likelier to appear in the unstable world which is close to its end. 

Previous literature review

Bostrom and Tegmark wrote: 

One might think that since life here on Earth has survived for nearly 4 Gy (Gigayears), such catastrophic events must be extremely rare. Unfortunately, such an argument is flawed, giving us a false sense of security. It fails to take into account the observation selection effect that precludes any observer from observing anything other than that their own species has survived up to the point where they make the observation. Even if the frequency of cosmic catastrophes were very high, we should still expect to find ourselves on a planet that had not yet been destroyed. The fact that we are still alive does not even seem to rule out the hypothesis that the average cosmic neighborhood is typically sterilized by vacuum decay, say, every 10,000 years, and that our own planet has just been extremely lucky up until now. If this hypothesis were true, future prospects would be bleak”. (Tegmark & Bostrom, 2005)

However, they also demonstrated based on the relatively late appearance of Earth in the history of the Universe that the space catastrophes capable of sterilizing Earth should be relatively rare, occurring no more often than once in 1 billion years with 99.9 per cent confidence. This is an example of SIA counterargument discussed in Section 4, which unfortunately does not work for Earth habitability

Circovic, Sandberg and Bostrom wrote a subsequent article, “Anthropic Shadow: Observation Selection Effects and Human Extinction Risks” in 2010 (Ćirković, Sandberg, & Bostrom, 2010), in which they created a Bayesian update equation to calculate the actual probability of a natural global catastrophe which takes into account observation selection effects. In that article, they listed five different types of natural catastrophes which observed probabilities could be affected by observation selection: asteroid/comet impacts, solar superflares, supervolcanic eruption, close flyby of rogue black holes and nearby supernovae explosions or gamma-ray bursts.

However, Ord et al (Snyder-Beattie et al., 2019) suggested that the natural catastrophes background rate for Homo Sapiens is not affected by the observation selection effects, as there could be earlier observers (this will be discussed in section 4.2). 

Manheim applied anthropic bias analysis to the frequency of natural pandemics in the past (Manheim, 2018).

Meanwhile, A. Scherbakov noted that the history of the Earth’s atmosphere is strangely correlated with the solar luminosity and the history of life, which could be best explained by anthropic fine-tuning, in the article “Anthropic principle in cosmology and geology” (Shcherbakov, 1999). In particular, he wrote that the atmospheric temperature was closely preserved in the range of 10–40 °C, and on four occasions the Earth came close to a “snowball” steady-state, and on four occasions came close to turning into a water vapor greenhouse where the temperature could reach of hundreds of degrees centigrade. However, these life-ending outcomes were prevented by last-minute events such as volcanic eruptions or covering of volcanoes in the ocean by water, which regulates the CO2 level following an eruption. Such “miracles” are best explained by observation selection effects. 

Waltham looked at the Milankovitch cycles and using the modelling of a group of planets found that the Solar system has unusually low orbit perturbations: “…the probability of all three occurring by chance is less than 10−5. It therefore appears that there has been anthropic selection for slow Milankovitch cycles. This implies possible selection for a stable climate, which, if true, undermines the Gaia hypothesis and also suggests that planets with Earth-like levels of biodiversity are likely to be very rare” (Waltham, 2011).

In the article “On the absence of solar evolution‐driven warming through the Phanerozoic” Waltham argues in favor of anthropic explanations of climate stability: “The Gaia hypothesis, anthropic selection or some other unconventional mechanism may therefore have to be invoked to explain the absence of long‐term warming through the Phanerozoic” (Waltham, 2014).

In 2020, two important articles on the topic were published. The article by Tyrrell “Chance played a role in determining whether Earth stayed habitable” (Tyrrell, 2020) demonstrated via computer modelling of the history of many Earth-like planets that chance played a significant role in preserving the stability of the atmosphere. 

Another article, “The Timing of Evolutionary Transitions Suggests Intelligent Life Is Rare” (Snyder-Beattie et al., 2020) is analyzing the timing of the important steps of the evolution of life and suggests that the observed frequency of such steps has strong anthropic effects, and median frequency is likely much lower, which implies longer expected time of development of intelligent life; from this follows Rare Earth. Robin Hanson in the “Grabby aliens” article discussed similar ideas (Hanson et al., 2021).

Wordsworth (Wordsworth, 2021) showed that runaway global cooling (snowball Earth) is still probable and could start if the global temperature falls to 7 C, and it was just a few degrees more than that only 20 000 ago at the peak of the Ice Age. He regards it as a possible explanation of the Fermi paradox, as in the snowball Earth glaciers will reach the equator and will destroy all complex life.

Survival bias is clearly ignored in some estimations of the risks of runaway global warming: 

“The paleoclimate record can be used to check our prediction that surface temperatures might increase dramatically were they to exceed ~305 K (Fig. 4b). The closest analog to the climate regime modeled here is the mid-Cretaceous Period, ~100 million years ago, which is considered to be the warmest Earth has ever been during the Phanerozoic eon. Life clearly existed and flourished during this time, so we know that whatever surface temperature prevailed then was not high enough to trigger a moist greenhouse.” (Ramirez et al., 2013)


2. The central argument and a toy example

2.1. Toy example

Imagine that every 100 million years on average a sterilizing catastrophe is happening on any Earth-like planet. It is a random process with a half-life of 100 of millions years. It is completely unobservable if it didn’t happen. 

We suggest here as a toy example that such a catastrophe is a nuclear explosion of a natural nuclear reactor in the Earth core, which some hypothesized to exist (Herndon, 1993). The explosion of the reactor will not destroy the planet but it will produce strong shockwaves and volcanism which will result in complete surface replacement in Earth, like the one which has happened in Venus. As Earth has existed for 4.5 billion years, the chances of the Earth to survive until now is 1 in 245 =3.5x1013. That is, anthropic shadow on Earth with the power around 1013 corresponds to the future life expectancy of 100 million years. 

If the reactor explodes with the half-life of 10 million years, the power of anthropic shadow is 1 in 10135, which is truly immense. But for evolutionary transitions like abiogenesis, such probabilities are not uncommon and we will show later (section 6) that the power of anthropic shadow is likely to be similar to the power of evolutionary transitions. Totani estimated the chances of abiogenesis and calculated that only 1 of 10100 planets will generate self-replicating RNA via randomness (Totani, 2020).

Even if the reactor explodes every once in one billion years, we have one billion years of future life expectancy, but anthropic power is 1 in 16. If there will be no anthropic shadow then expected future survival will be equal to the past survival and will be 4.5 billion years.

It seems that human civilization will be fine, as we still will have 10 million years (with 50 per cent probability) in the worst case, which is enough to become a multi-planetary civilization.

However, any small additional fuel will trigger the explosion. Russian scientists suggested sending a probe to the Earth’s core based on a small nuclear reactor that melts all the way down (Ozhovan et al., 2005). If the natural reactor is “trigger-happy”, such experiment could be enough to destabilize it. Such an Earth-core probe will be like a pin that is poking an overinflated balloon – dangerous game! 

 Remind you, that it is a toy example, so the real probe unlikely will reach the core, but it is still could cause large a degasation event (Cirkovic & Cathcart, 2003)

Moreover, the problem is that the nuclear reactor inside Earth could have accumulated the fuel all that time. We suggested before that its explosion is a truly random effect, but for some processes like inflating balloons or overstretched springs, the probability and the power of explosions grow non-linearly in time. This means a significantly shorter future life expectancy and higher fragility. Here we looked at exponentially distributed risks. We look at normally distributed catastrophes in Appendix 2 and we assume there that for an overinflated-balloon-type of catastrophe the probability is normally distributed around some mean value.  In that case, for example, 14 sigmas anthropic shadow event has 1 in 1045 chances. It produces the future habitability of Earth equal to 0.04 per cent of the past age, but it is still 1.6 million years half-life which provides enough time for a civilization to leave Earth.

If there are two universes, and in one of them all Earth-like planets have half-life 100 million years because of the core reactor explosions, and in another universe, it is 200 million years, – at the end, there will be 222.5=500 million times more habitable planets in the second universe. Therefore, we are more likely to find ourselves in the universe with a weaker anthropic shadow. The second universe provides the baseline level of the anthropic shadow if only two types of universes exist. This is the SIA counterargument in a nutshell which will be discussed later in section 4. 

This thought experiment demonstrated to us the following properties of the anthropic shadow: 

1. Even a minor anthropic shadow means a significant cut of future life expectancy

2. The sensitivity of the future life expectancy to the power of anthropic shadow is relatively small. The growth of anthropic shadow for 10134 times lowers the life expectancy by only 100 times.

3. Any anthropic shadow means significant growth of fragility. 

4. The power of the anthropic shadow is limited to some baseline which applies to all Earth-like planets. 

5. Non-linear probability distributions of catastrophes mean much shorter future life expectancy time and higher fragility


Humanity will either become an interplanetary civilization in the next millennia, or it will never do it, given the expected quick growth of nanotech and AI. Therefore, a million years timescale of natural catastrophes is not itself a significant risk for our space future. 

2.2. The central argument

The toy example above helps us to formulate the central argument about the anthropic fragility of the habitability of the Earth, which we will explore later in detail. The argument runs according to the following lines:


  1. Habitable planets with intelligent life in the Universe are rare because many types of catastrophes could kill life.
  2. The fact that we are alive means that we were very lucky and have escaped many past catastrophes.
  3. But therefore, we can’t estimate the frequency of such sterilizing catastrophes based on the observations. This is the anthropic shadow.
  4. If some catastrophe is long overdue, this lowers our future life expectancy for around 1-2 order of magnitude, based on the low sensitivity of anthropic shadow to initial parameters. This seems to be not problematic as it still gives us millions of years of life expectancy.
  5. However, the fragility of our environment also has grown, and thus relatively small or unnatural anthropogenic actions could cause an unexpected catastrophe. 
  6. The primary risk here is a sudden climate catastrophe caused by crossing an unexpected tipping point which will start a positive feedback loop and will increase Earth’s temperature to 65 C moisture greenhouse level or even higher. Collider accidents and nuclear war are other examples of anthropic fragility.
  7. To counter this unobservable fragility, we need to be more careful and to have quick reaction instruments, like urgent geoengineering.
  8. However, as intelligence is a universal adaptation, it is more likely to evolve in the world with changing climate, and such changes themselves are the sign that the climate catastrophe is near. This increases the chances of catastrophe soon. More on that in section 6.


3. Anthropic fragility and global catastrophes affected by it

3.1. Anthropic fragility: underestimation of the fragility of our environment because of the anthropic shadow

Now it’s time to look deeply at what is anthropic fragility. Not all types of survival bias result in anthropic fragility. If a plane with holes from an enemy fire has returned to the base, there is no anthropic fragility. Thus, anthropic fragility requires that the situation of risk is continuous. But even this is not enough: asteroid impact risk is continuous, but there is no anthropic fragility, as there are no meaningful ways how humans could affect the probability of large impacts, except asteroid deflection. Poking an overinflated balloon is an example of the fragility, and injuries from falling for an older man is another ­– but all these examples are not anthropic: there is no observation selection effect which results in the underestimation of risks. 

If a plane with many holes is still flying back to its base, it is closer to anthropic fragility, as the plane is damaged and any sharp turn could put too much force on its structures and will fail apart. 

Anthropic fragility is the following situation:

  1. Anthropic shadow. There is a strong anthropic shadow, so we live in a world where some kind of otherwise very probable sterilizing catastrophe has not yet happened by pure chance.
  2. Parameter. The probability of the catastrophe is connected with a slow accumulation of some parameter, similar to the pressure in an inflating balloon.
  3. Tipping point. The catastrophe will inevitably happen if the parameter will achieve some threshold level (or at least, catastrophe’s probability significantly increases if the parameter grows even slightly).
  4. Human actions, like experiments or emissions, are changing the value of the parameter, so it could reach the threshold. Human actions may affect some other foundations of stability that don’t look fragile at first glance. These indirect human actions must be unique in the sense that they never happened before in exactly the same way.
  5. Unknown to humans. Humans do not know the real activation level of the parameter because they can’t infer its value from the past rate of catastrophes, which they never observed because of the survivorship bias. They may even not know that such a catastrophe is possible at all.


Anthropic fragility is defined both as physical and epistemic situations: the physical situation is that a catastrophe is long overdue, and the epistemic part is that we can’t know it from the past observation and therefore underestimate the safe level of the changes of the parameter. Even if we agree with the idea of anthropic fragility, we still can’t know which parameter is causing fragility and what are its safe levels of manipulation. In the damaged plane example, the pilots may know that the plane is fragile, but do not know which actions are risky: climbing, turning or landing.

There are two types of anthropic fragility: one is when we increase the pressure on something which is already under pressure, that is, like adding more air in the inflated balloon, and another is when we perform unique, never happened before actions on a system which is in a metastable condition, like poking the balloon with a pin. We could call them a “parameter increase” and “unique actions”. In the case of global warming, adding more CO2 and increasing temperature is the “parameter increase”; and the unprecedented speed of warming and unique methane effects are an example of unique human actions. In reality, both types of fragility are connected, because unique actions cause some parameter increase or the parameter increases in some unique way. 

For example, if our false vacuum is very close to the transition to the next metastable state, Large Hadron Collider (LHC) experiments may be more dangerous than they appear (Hut & Rees, 1983) because they create some unique type events, as was suggested by Kent, more on that below. 

3.2. Runaway global warming as a most dangerous form of the anthropic fragility


3.2.1. Moist greenhouse, tipping points and sea-floor methane

A recent article suggested, based on computer simulation, that water-world planets (which Earth is similar to) have a second semi-stable temperature regime, moist greenhouse, with a mean temperature of 57°C while Earth now has a mean temperature of 15°C (Popp et al., 2016). All temperature regimes between these two will collapse to either the current climate or to a moist greenhouse; thus, there should be some tipping point between these two temperatures after which positive feedback loops dramatically accelerate. 

We assume here that the moist greenhouse will cause human extinction, the same way as ocean evaporation could do it, so there is no practical difference between the two from anthropic and existential risks views. However, there is a small possibility to survive moist greenhouse in some coldest places on Earth like very high mountains, like Himalayas, and in the Antarctic. Toby Ord estimated in the Precipice (Ord, 2020) that there is a 0.1 per cent probability of the existential risk this century because of climate change, mainly due to the start of the moist greenhouse. 

The idea of tipping points in climate has been often discussed, but because of anthropic shadow, the position of the actual location of the tipping point could be underestimated. In a climate context, “tipping point” means not only a temperature but some combination of temperature, greenhouse gases concentrations, albedo, solar luminosity and sea-floor methane release rate. The limited size of the previous “close-call” tipping points, like PETM warming, could be explained by the observation selection effect: if the Earth had turned into a moist greenhouse 55 million years ago, there would be no observers now.

The question of the speed and the possibility of methane-driven climate change is a topic of scientific debate (Shakhova et al., 2010) which are too complex to be completely presented here. But anthropic shadow effect of climate fragility increases our uncertainty about such feedback loops. If the transition takes only a few years, we will not have time to cope with it via geoengineering, except perhaps via a nuclear explosion in a supervolcano to create artificial volcanic winter. However, the anthropic shadow prevents us from observing quick and large magnitude changes of the climate in the past, if they are possible, so the absence of such events in our history is not evidence for further slow future climate changes.

The anthropic fragility becomes especially important relative to the catastrophes which are long overdue. An overinflated toy balloon is in a metastable state, where even a small punch could lead to its explosion. Higher levels of radiation from the Sun has been compensated by historically low levels of CO2, which, however, helped glaciation and methane accumulation in the permafrost and ocean floor, which later could cause a very large “clathrate gun” rapid warming (Kennett et al., 2003). This fragility can’t be observed in the historical record because of anthropic shadow. We should be extremely careful with climate change. 

The main difference between now and previous periods of warming is the large accumulation of methane hydrates which by some estimates are 10 times higher now than during PETM, as current ice age conditions helped such accumulation on the seafloor and in the permafrost (Ananthaswamy, 2015, 2015; Dean et al., 2018). The topic of risks from methane eruption (Ananthaswamy, 2015; Dean et al., 2018) from the Arctic is controversial: some claim that it is the main risk of warming and other present models that methane leaking will not be enough to start runaway global warming. Anthropic shadow could make us underestimate the power of previous methane discharges. Also, there is a difference between anthropogenic and natural effects on methane: as methane is a short-lived gas in the atmosphere, its concentration depends on the speed of its leaking from reservoirs, which itself depends on the speed of the temperature change. Anthropogenic global warming is relatively quick, because of the unprecedented speed of CO2 emissions and thus will produce more methane concentrations than the same CO2-driven warming if it were slower. Recent research showed that the speed of change was a characteristic of past mass extinctions (Song et al., 2021).

Many argue that climate change is not an existential risk and that its danger is exaggerated (Lomborg, 2020). But what makes it a real existential risk is the fat tail of uncertainty of its magnitude powered by anthropic bias.


3.2.2. Runaway Greenhouse

Besides moisture greenhouse, which is hypothetically survivable, there is an even worse scenario, when whole oceans evaporate and equilibrium temperature reaches 1400K, which is described in the article “The Runaway Greenhouse: implications for future climate change, geoengineering and planetary atmospheres” (Goldblatt & Watson, 2012): 

The ultimate climate emergency is a “runaway greenhouse”: a hot and water vapor rich atmosphere limits the emission of thermal radiation to space, causing runaway warming. Warming ceases only once the surface reaches ∼1400 K and emits radiation in the near-infrared, where water is not a good greenhouse gas. This would evaporate the entire ocean and exterminate all planetary life. Venus experienced a runaway greenhouse in the past, and we expect that Earth will in around 2 billion years as solar luminosity increases.

Goldblatt & Watson concluded that CO2 emissions alone can’t cause this, but if other warming sources will add up, this becomes possible:


The question here is simply how much could human action increase the strength of the greenhouse effect? Kasting & Ackerman (1986) found that, with carbon dioxide as the only non-condensible greenhouse gas, over 10,000 ppmv would be needed to induce a moist greenhouse. This is likely higher than could be achieved by burning all the “conventional” fossil fuel reserves—though the actual amount of fossil fuel available is poorly constrained, especially when one includes “exotic” sources such as tar sands (which are already being exploited). Greenhouse gases other than carbon dioxide, cloud or albedo changes could all contribute further warming. Likewise, the exhibition of multiple equilibria in the relevant temperature range (Renn´o, 1997; Pujol & North, 2002) complicates matters. 


They conclude: We cannot therefore completely rule out the possibility that human actions might cause a transition, if not to full runaway, then at least to a much warmer climate state than the present one. 

High climate sensitivity might provide a warning.” (Goldblatt & Watson, 2012) and then argue that we may need geoengineering to stop runaway warming. Growing climate sensitivity could be a warning sign. However, they validate the sensitivity estimates on past climate without taking into account a possible anthropic shadow: “Such high sensitivity is inconsistent with our knowledge of paleoclimate and the model cases which provide the extremes do not seem likely (due to poor representation of contemporary climate)”.

Therefore, the question is: are existing climate models are capable to predict risks of runaway warming? If yes, we should not worry about the anthropic fragility of climate. But the models are limited: 

Ideally, we would want numerical climate models to robustly resolve the transition to a much hotter atmosphere. However, most such models have been developed for fairly small perturbations from the existing climate and their wider applicability may be limited by the obvious unavailability of data to tune the model to, and by simplifications made to reduce computational cost… A new generation of model may well be needed (Collins et al., 2006; Goldblatt et al., 2009)). 


3.2.3. CO2 concentrations which could start runaway warming

We could measure anthropic fragility as an amount of deviation from the current level of some parameter, which will cause a global catastrophe and which estimation is distorted by the anthropic shadow. For example, how much CO2 could be added to the atmosphere before a tipping point is reached?

The Earth in the past had CO2 levels much higher than today, so a naïve view is that even having 10 times more CO2 than now will not cause runaway global warming. But this view does not take into account the anthropic effects, that is, we cannot use the past data about the probability of runaway global warming, as we could observe ourselves only on a planet where runaway global warming never happened. Also, previous higher levels of CO2 were at least partially compensated by lower Sun’s luminosity, lower deposits of methane hydrates and other geophysical factors then, such as the different configurations of oceans and different parameters of Earth orbit. Such factors in the past worked as protection against runaway global warming, but they are not in place now.

There are different assessments of the critical levels of CO2 after which runaway warming will happen. As was cited above 10 000 ppm is needed for a moisture greenhouse. Another estimation of the dangerous level is 30 000 ppm CO2 which is unachievable by anthropogenic mineral fuel consumption (Goldblatt et al., 2013), and another estimation is only 12 times the preindustrial level, that is, 3360 ppm (Ramirez et al., 2014). The current CO2 level seems to be the highest in the last 5 million years ago and could become higher than it was in the whole Miocene (20 million years long) in the next 100 years (Dean et al., 2018).

According to IPCC’s, in the worst-case SPP5-8.5 scenario the CO2 levels could reach 1000 ppm at the end of the 21 century (Allan et al., 2021), which seems to be not enough to trigger runaway global warming, according to cited above scenarios.  

However, the real problem of the anthropic shadow is that we can’t know for sure that this CO2 level will not cause runaway warming. There are several sources of uncertainty: 


  • Large uncertainty comes from the possibility of methane emissions from the permafrost and other sources. Methane warming potential and release rate depend on the speed of warming, as methane is short-living gas: if the release rate is low, methane concentration will be also low. Human anthropogenic warming is different from the natural increase of CO2 in the past as anthropogenic CO2 levels are increasing quicker, and thus the role of methane will be higher.
  • Other pollutants, like N2O, also have high warming potential. 
  • Our efforts to cut emissions can backfire, because of aerosol-removal effects, as industrial aerosols block part of sunlight in the upper atmosphere (Hansen & et al, 1992).
  • More generally speaking, our models are calibrated on the historical relation between CO2 and global temperature, but because of the anthropic shadow, this relation could be unreliable. We survived only in those worlds where extreme deviations never happened.
  • Chaotic nature of climate. In the same way as weather, climate also could be fundamentally unpredictable. There could be strong deviations from the mean and bifurcation between different semi-stable attractors. 
  • Unpredictable combination of random events may force Earth out from the zone of climate stability. One such event was Younger Dryas when a large ice lake leaked.  
  • Our models may be incomplete in long run and may not take into account some factors.


We are not trying to say that the IPCC model is wrong. It is the best what we have. What we try to show is that there is a small probability of a tail risk, which, however, has most of the expected negative value.

Also, the nature of the tipping point is that it is the beginning of a self-accelerating runaway process, but not the point where the process will reach its maximum level. Thus, the CO2-concentration when the runaway warming starts is not the same as the one at which the CO2 level will be high enough to cause tens degrees of warming and human extinction. Methane and water vapor may be the main drivers of warming after reaching the tipping point, so the lack of CO2 sources will not stop the warming. This is similar to poking of a small hole by a needle in a toy balloon, which will inevitably result in its explosion. The needle creates a hole but does not participate in the process of destruction after that.

Fragility can be illustrated by an example of overstretched spring, in which fragility and life expectancy are measured in the same units: per cents of additional inflation. Therefore, the anthropic fragility should be measured not in the CO2 levels, but in the temperature increase, as CO2 levels are non-linearly proportional to warming and depends on other things that could affect temperature. There were periods when the Earth temperature was 15C higher than now and the hypothetical moisture greenhouse temperature is 45C higher than current temperatures, so the naïve view is that we are safe even if maximum expected global warming will happen with around 5-8C temperature increase. Thus, the tipping point for moisture greenhouse is somewhere between 15C and 45 C increase, according to the naïve view.

However, the idea of anthropic fragility means that we should lower our estimates for around an order of magnitude, which, given all uncertainties, means that the tipping point could lie not in tens but in single digits of temperature increase (that is, between 1.5C and 4.5C, if we just divide on 10 the above estimate). This temperature increase could be reached in the 21st century, and maybe even in the next decade. See the discussion below why PETM warming 55 million years ago is not a safe data point, as at that time was different conditions than now.

How long it takes from the tipping point to full-blown moisture greenhouse is unclear, and it could be from weeks, if water vapor feedback is activated, to centuries, if methane and ocean thermal inertia will play a major role (Karnaukhov, 2001). It is also unclear, if the warming will stop on the moisture greenhouse level, or will overshoot and will go to the Venusian runaway scenario.

3.2.4. Urgent geoengineering may be needed to counter climate models’ uncertainty

As humans do not—and cannot because of the anthropic shadow effects—know the location of the climate tipping points (Lenton, 2011) based on the past historical record, it may be prudent for society to perform more research about the climate tipping points and to research methods of urgent geoengineering, which could give us more time if runaway warming starts. One model gives 4 years estimate until moisture greenhouse from the moment of the beginning of radiative forcing (Seeley & Wordsworth, 2021).

he only such measure which could be used on short notice is the use of nuclear weapons to start nuclear winter by causing large fires in taiga or by the initiation of volcanic eruptions. Such measure could put us back below the tipping point, or give a few more years to prepare better protection measures. Higher altitude explosions over taiga will have less nuclear fallout and could be done using existing delivery systems. The slower method is the use of existing airplanes to perform sulfate stratospheric injection (Halstead, 2018). “Normal” geoengineering aimed at the gradual removal of CO2 does not give us enough flexibility to react to unexpected climate changes. We also should do everything else possible to prevent anthropogenic global warming, first of all, cutting emissions. 

3.3. Collider experiments as another type of anthropic fragility

Another type of fragility is associated with the risk of conducting physical experiments that create completely new conditions on Earth, i.e. experiments at the hadron colliders. It has been suggested that they could cause three types of catastrophes: a false vacuum collapse that would end the entire observable universe, a mini-black hole that would slowly but acceleratingly eat Earth's matter, and a special type of quark matter called strangelets, which supposedly capable of turning ordinary matter into strange matter (Kent, 2004)

One popular argument for collider safety is based on the observed safety of high-energy collisions of cosmic rays with Earth’s atmosphere, but it could also suffer from anthropic effects. The argument is that if experiments at colliders could create mini-black holes, then cosmic rays would be capable of it. However, since we have survived for billions of years, this has never happened. Therefore, high-energy collisions cannot create anything dangerous. But this argument clearly suffers from the fact that it does not take into account the survival bias.

Dar et al. (1999) presented an "anthropically invulnerable" argument for the safety of the collider: if high-energy collisions are dangerous, we could observe random supernovae from sudden collapses of other stellar objects and planets. Note that Dar's argument works only for the formation of strangelets and small black holes, which will only cause a local catastrophe on Earth, but not for the collapse of a false vacuum that can destroy the entire universe.

In the case of collider catastrophes, anthropic fragility plays a role, since human experiments turn out to be different from natural phenomena and, thus, can cause a catastrophe not observed on other planets. A. Kent wrote that there is a subtle difference between collider experiments with collisions of cosmic rays with the Earth's atmosphere. Namely, the products of collider collisions have zero velocity relative to the Earth, because they are produced by two opposite particle beams, and the products of cosmic ray collisions continue to move at near-light velocity relative to the Earth. This may change the nature of the interaction of products with terrestrial matter (Kent, 2004) giving mini-black holes more time to accumulate mass and start growing.

The main question here is: are the experiments at the collider unique or not in the sense that these are events that have never happened in the history of the Earth? If they are not unique and similar events have occurred on Earth and other planets without causing the end of the world, there is no risk of anthropic fragility.

Bostrom and Tegmark explored the probability of the natural false vacuum decay and similar cosmological catastrophes and found it to be low given our late existence. Their article (Tegmark & Bostrom, 2005) does not directly apply to the collider experiments, as it puts a limit only on the natural cosmological catastrophes, and if the colliders create really unique conditions, no natural process is a reference class; of course, it also means that there are no alien civilizations in our past light cone which had performed such dangerous experiments before us and had created a wave of destruction through the whole universe. Their counter-arguments are discussed in more detail in section 4.1. 

In 2008, it was suggested that the series of collider failures that occurred before the launch of the LHC could be explained by the anthropic shadow, since we can only survive on worlds where the collider does not work; see also (Ord et al., 2010) about anthropic considerations in the collider risk estimations. However, LHC started its operations after that and continued for around 10 years. However, the LHC began its work after that and lasted for about 10 years. If the LHC had a high probability of causing a false vacuum decay, then our own existence as the authors of this article seems too late, based on the same line of argument used by Bostrom and Tegmark: if LHC causes vacuum breakdown, most scientists should find themselves just after the beginning of its operation (or before). This is also an example of a more general counter-argument against the anthropic shadow called "early observers", which is discussed below in section 4.2.

One way or another, the anthropic shadow can make us underestimate the fragility of our vacuum, and any unique (which has never been in nature) experiment can become a pin that can burst the “over-inflated ball” of an over-due catastrophe.  

3.4. Different types of known global catastrophes and their anthropic shadows

Waltham (Waltham, 2019) explored seven possible anthropic fine-tunings of the Solar System, including the mass of the Sun, the mass of the Moon, the orbits of the major planets, the Earth's ocean, the Earth's magnetism, and the Earth's plate tectonics. The evidence for the fine tunings he found is not very strong when taken individually, but collectively they point to some anthropic pressure in the evolution of the solar system. However, most of the fine-tuning methods he explores are positive conditions; here we are interested in negative conditions, that is, what types of events should not occur for the emergence of intelligent life. We have included in Table 1 several other global risks which have been hypothesized to be affected by the anthropic shadow.


Table 1. Types of possible catastrophes and corresponding anthropic effects

Type of natural catastrophe

Description and comments

Evidence for anthropic pressure

Could observed catastrophes rate be affected by anthropic shadow?

Could it have anthropic fragility?

Type of period for periodic events

Collider catastrophe

(Ord et al., 2010; Sandberg, 2008)

Colliders hypothetically could create new types of matter which could destroy Earth: mini black holes, strangelets or false vacuum transition (Kent, 2004)

Sandberg suggested that the LHC's series of failures in the early 2010s is best explained by the "quantum immortality effect" because if it starts working, it will destroy the world.


Yes, in the form of uniqueness of human experiments


2. False vacuum decay (Hut & Rees, 1983)

Hypothetical type of catastrophe

String theory suggests 10500 possible vacuums. 

(Douglas, 2003);  Many of them are unstable (Adler et al., 1995)


Hadron collider may trigger such a catastrophe 

Not known; probably random event

3. “Alien invasion” (Gertz, 2016), (Hanson et al., 2021), (Turchin & Denkenberger, 2019)

The shock wave of alien colonization, at a speed close to the speed of light, consumes all matter in the universe. (Armstrong & Sandberg, 2013).


We have only been able to survive in that part of the universe where aliens have not yet appeared or are hiding.


Aliens may be closer than Fermi's Paradox suggests.


Messaging to Extra Terrestrial Intelligence could attract hostile aliens if they are nearby 

(Baum et al., 2011) or we could find their dangerous signals (Turchin, 2018c).

If natural intergalactic panspermia (spread of life) is possible, civilizations of our age are closer in space and could arrive on Earth relatively soon. (Turchin, 2020).

4. Gamma-ray burst (GRB)

 (Ćirković & Vukotić, 2016)

Directed GRB could sterilize planets 1000 light years away.

The sun is located in a relatively distant part of the galaxy, and terrestrial life arose relatively late in the history of the universe, so gamma-ray bursts are rare in our region.

No, because we are seeing a lot of distant gamma-ray bursts. However, some nearby stars may be on the verge of forming a burst that was "delayed" by the anthropic shadow.


Random event

5. Nearby supernova (Ćirković & Vukotić, 2016)

Supernova is dangerous if it is closer than 8 light years (Ćirković & Vukotić, 2016).

We are not in that part of the Galaxy where there are many supernova explosions (in the core).

No, as we observe many remote supernovas.


Random, very rare event based on observational constraints

6. Solar super-flare

 (Lingam & Loeb, 2017)

Sun-like stars have super-flares which could be dangerous to life on Earth.

The Sun seems to be a surprisingly quiet star, as other stars of its type are more active 

(Reinhold et al., 2020). Also, Sun may have had larger flares in the past but now is quiet.

If the Sun were to become more active, higher levels or radiation and electromagnetic storms might prevent the rise of a technological civilization.


Not known

7. Asteroid/comet impact (Chapman & Morrison, 1994)


Over the past few hundred million years, there have been no collisions that could have wiped out all vertebrates. (Rampino & Caldeira, 2015)

There is a periodicity of about 30 million years of mass extinctions that could be caused by an influx of comets from the Oort cloud, and we are now living near the end of such a period.


In addition, several stars have recently passed close to the Sun and may have disturbed the Oort cloud. (Bailer-Jones, 2015).


Random cometary streams could be semi-periodic if they were associated with perturbations of the Oort cloud by solar oscillations around the galactic plane.

8. Supervolcanic eruptions and flood-basalt events

(Rampino, 2008).

Large-scale eruptions, i.e. Siberian traps; even larger eruptions are possible, resulting in surface replacement

There have been no comparable basalt flood events since the Great Dying 242 million years ago, except perhaps for Deccan traps.


Could be triggered by experiments with deep drilling and core penetration but currently unlikely. (Cirkovic & Cathcart, 2003), 

Tension-spring mode

9. Runaway global warming

(Popp et al., 2016).


The peculiar stability of the Earth's climate, despite the change in the luminosity of the Sun.


May be caused by increased CO2 and methane emissions from permafrost.

Inflated balloon mode

10. Ocean anoxic event and H2S poisoning of atmosphere

(Ward, 2007).


Black sea could produce enough H2S to poison atmosphere and the level of H2S in it is growing

(Kump et al., 2005)


Phosphorus is the main problem. It is used as a fertilizer throughout the world and ends up in the oceans. This could trigger an anoxic event in the ocean, which has been associated with mass extinction events, probably due to the formation of H2S (Handoh, 2013). Global warming increases the likelihood of anoxic events.


11.Stellar encounters with giant molecular clouds 

(Kokaia & Davies, 2019).


This can lead to a closer supernova, an increase in the frequency of asteroid impacts, and climate change.

Such encounters affect the stability of the Oort cloud and the planet's climate through dust accretion.

Anthropic effects define the galactic habitable zone where such encounters are rare.


Random, 1.6 event in Gyr for Sun.

12.Nuclear war

(Sandberg et al., 2018).

Some have suggested that the fact that World War III did not occur in the 20th century is best explained by anthropic selection.

A nuclear war is unlikely to kill everyone, so the anthropic effects should not be strong.

A nuclear war would most likely destroy civilization and there would be fewer scientists left to discuss anthropics, so we would most likely be discussing it in a world where there was no World War III.

The risk of an accidental nuclear war and the ease of provocation can be underestimated because of survival bias.


13.Next Ice age

Human civilization was able to appear only during a period of warm and stable climate which was good for agriculture and large empires (Gowdy, 2020). It has been relatively warm and stable for the last 11 700 years after the Young Dryas. It was predicted that the next Ice age will happen in a few millennia from now. The previous interglacial period’s duration was around 13 ky and happened 130 ky ago. 


Human civilization arose during a stable interglacial period, and such periods, as a rule, make up only one tenth of the total time of the modern glaciation.

Humanity's efforts to combat global warming could unexpectedly backfire and lead to an early onset of the next cooling period. This includes CO2 reduction and geoengineering.

Periodic event with a random component



4. Counterarguments against the anthropic shadow 

4.1. Self-Indication Assumption counterargument: an observer is less likely to be in a world with high anthropic shadow

The counterargument states that if there are two possible worlds, and one of them has an anthropic shadow and the other does not, then there will be more observers in a world without an anthropic shadow, and so I am more likely to be in a world without such a shadow.

In a nutshell: The anthropic shadow is exactly compensated by my smaller chance to be in such world.

Thus, according to this counterargument, we are likely to live in a world where there is no survival bias and no problem with underestimating the risks of future disasters. 

But there are three objections to this counterargument which are limiting its power: minimal level of anthropic shadow for all worlds; the numerical dominance of the semi-fine-tuned worlds and a possible correlation between catastrophes and habitability which will be discussed below.

But firstly, a counterargument’s example: imagine that there are two groups of potentially habitable planets, and in each group initially there were 100 planets. In the first group, the probability of the past catastrophes is 0, and in the second group, only 1 planet has survived and gave rise to a civilization. Thus, in the second group, the anthropic pressure is 100 to 1. However, if an observer does not know in which world she is located, she has 100 times greater chances to be in the first group of worlds, as there will be 100 habitable planets. 

This reasoning is based on so-called Self-Indication Assumption (SIA) which favors the worlds with larger number of observers (Bostrom, 2013). The SIA counterargument was suggested by Stuart Armstrong in a comment to my blog post about anthropic shadow.

SIA have two interpretations: first, it is simply about the distribution of observers among different universes in the multiverse, but it requires that all these universes actually exist. In another, stronger interpretation, the SIA is seen as an argument that of two possible universes, the more populated one is more likely to actually exist like in the Presumptions philosopher thought experiment (Bostrom & Ćirković, 2003). In case of modal realism, these two versions merge. Here we will use the first, simpler one, which assumes the existence of many universes with different properties. For more on SIA and observer densities see in my article “Presumptuous philosopher proves panspermia” (Turchin, 2020). 

In other words, it looks like that SIA exactly compensates the anthropic shadow, the same way as it exactly compensates the estimation of future catastrophes following from the Doomsday argument (Bostrom & Ćirković, 2003).

For example, if in the one part of the multiverse the false vacuum decay is likely and in the another it is not possible, the observer is much more likely to find herself in the part of multiverse where false vacuum decay never happens.

We can generalize this principle: 


The anthropic shadow’s power is limited by the possibility of existence of another universe that produces the same observer with less anthropic shadow  


However, this principle cannot exclude the anthropic shadow idea completely, as some anthropic shadows may be irreducible.

For example, the anthropic shadow has limited the size of the asteroids which have impacted the Earth in the last billion years, but the existence of asteroids and impacts may be a necessary condition for the evolution of life on Earth-like planets, as a large impact likely created the Moon (Barr, 2016), comets may have brought water to Earth (Hartogh et al., 2011), and the impact-driven extinction of dinosaurs was probably necessary for mammals to rise to dominance (Alvarez et al., 1980). Thus, all possible habitable planets are likely under the risk of large impacts. Because of this, intelligent life may have evolved only on a small share of lucky planets; thus, the anthropic shadow for impacts would be strong and future expected rate of impacts would be higher.

Therefore, the first objection to the SIA-counterargument is that there is a minimal level of anthropic shadow for all habitable planets.

Another objection to SIA counterargument is the idea that the worlds with anthropic shadow could still produce more planets with civilizations than the worlds without such shadow as they are more numerous. To illustrate this, we will return to our example above. Before, we assumed that the initial number of the planets was equal. Now let us assume that initially there are 10 000 planets with 0.01 survival rate, and 10 planets with 1 survival rate. At the end, there will be 100 habitable planets in the first group, and still 10 in the second. In that case, we are still more likely to find ourselves on a planet that had a strong anthropic shadow. Let us call the first group semi-fine-tuned for intelligent life, and the second one – fine-tuned

In that case, the anthropic shadow is real if the share of semi-fine-tuned planets in the multiverse is larger than the share of the fine-tuned ones. Obviously, there are no data about such shares, but mathematical simulation of possible planets by Tyrrell gives similar results: most currently habitable planets will have anthropic shadow (Tyrrell, 2020). 

The observational consequence of this is that most observers may appear not because of the perfect fine-tuning of initial conditions, but because of some very random events which overcome non-perfect fine-tuning. In a fictional example, in a perfectly fine-tuned universe there would be no dangerous asteroids/comets in stellar systems. But in our solar system, we have many asteroids, and this could be compensated by some additional conditions, such as Jupiter protecting Earth from comets, or by pure chance. If humanity is protected by pure chance, this protection may not work in the future, which means a higher chance of asteroid/comet impacts in the future.

In other words, the domination of imperfectly fine-tuned worlds means that negative fine-tuning—where the random absence of the catastrophic events is the main mechanism of civilizational survival—may be the dominating form of anthropic selection, and thus the anthropic shadow would be much stronger.

The third reason to reject the SIA-counteragent is that there could be a correlation between higher risk of catastrophes and habitability. Universes which allow interstellar panspermia will have billion times more planets with life, but in such universes asteroid impacts are more frequent and stars are closer to each other, which implies higher rate of natural catastrophes (Turchin, 2020). Another idea of this type is that intelligence is more likely to appear in an unstable world which will discussed in section 6.

However, SIA is a strong argument against extreme forms of anthropic bias, like the collapse of false vacuum every 10 000 years, but doesn’t work against weaker forms of anthropic bias, like higher rate of impacts or a possibility of runaway global warming. It is not easy to say without concrete data where is the line of balance between SIA and anthropic bias. 

We think that SIA allows some form of anthropic bias, maybe as weak as 1 order of magnitude, but as we showed in section 2, the magnitude of anthropic bias has a relatively small impact on the future life expectancy, which is around 0.1 of previous time, if some form of anthropic shadow is present. However, in section 5, we will explore an argument that minimal level of anthropic shadow is comparable with probability of other big evolutionary filters and is many orders of magnitude. 

4.2. “Early observers” counterargument

4.2.1. Early observerhood

In the article “An upper bound for the background rate of human extinction” Snyder-Beattie et al. (Snyder-Beattie et al., 2019) introduced the idea of early observerhood which could appear long before now: “To model observation selection bias, let us assume that after Homo sapiens first arises another step must be reached. This could represent the origin of language, writing, science, or any relevant factor that would transition early humans into the reference class of those capable of making observations (we call this step ‘observerhood’).” They assume that such a factor may appear something like 20 000 years ago, while Homo Sapiens has existed for 200 000 years. 

After long calculations, they conclude: “In summary, observer selection effects are unlikely to introduce major bias to our track record of survival as long as we allow for the possibility of early observers”. The main reason for this is that if observers have been here for a long time, it is a strong argument against the high background extinction rate, as otherwise, one would likely find oneself near the beginning of the existence of observers. For example, if observers have been on Earth for 20 000 years, this is a strong counterargument against the background extinction rate of 1 in 1000 years, as only 1 in 220 timelines will reach such stage. 

As one should think about oneself as an observer randomly selected from all observers, one should find oneself where most observers are concentrated, that is, in the first millennium in this example. But humanity does not find itself at this time, according to Snyder-Beattie’s logic, so 1 in 1000 extinction rate per year is most likely wrong. 

In a nutshell, it is a Doomsday argument in reverse, as we use observers’ long survival as an argument against the high rate of catastrophes (Bostrom, 2001; Turchin, 2018a).


4.2.2. Qualified observers

However, it looks like that Snyder-Beattie et al. used a too relaxed definition of a “qualified observers” that constitute the relevant reference class. We argue that the qualified observer should be able to understand the idea of the “anthropic shadow” and be able to think about related topics. Neither hunter-gatherers nor medieval writers could do this. People with the necessary mathematical training and similar ideas began to appear only in the 18th and 19th centuries, i.e. Laplace with his sunrise problem (Marquis De Laplace, 1814), but the number of such observers increased only at the end of the 20th century when anthropic reasoning and Doomsday argument appeared. 

The reason we use a more rigorous definition of observerhood is the idea that “I am randomly selected among functionally indistinguishable observers”, that is, from all those who think about a certain topic (Yudkowsky & Soares, 2017). We should ignore in our calculations all observers who are not thinking about this topic, no matter if they are bright minds, can speak, can feel or whatever. 

Therefore, there are no “early observers” (with one exception discussed below). Currently living population of the observers is the first one, and because of that, they are especially unsure about their past and future extinction rates, like Adam and Eve in Bostrom’s article (Bostrom, 2001). Moreover, it now looks like an argument in favor of higher extinction risks: as qualified observers exist for only around 50 years, future life expectancy could be also short.

Despite our criticism of the definition of observerhood, Snyder-Beattie’s counterargument applies only to the natural risks that could have occurred in the last 20 000 years, where the difference between a current person’s location and earlier observers is significant. It can’t be applied to long-term risks, such as the rate of asteroid impacts or stability of the atmosphere, with reoccurrence rate of dangerous situations around tens of millions of years. Snyder-Beattie’s counterargument also doesn’t work for climate’s anthropic fragility, as anthropogenic global warming has started only in last couple of hundred years. 


4.2.3. Time until nuclear war

However, there is one situation where Snyder-Beattie’s counterargument works. It is anthropic estimate of the probability of nuclear war. There was no global nuclear war for more than 75 years, and anthropic reasoning exists from the 1970s, as Carter suggested anthropic principle in 1973, that is 48 ago from now (as of 2022). 

If nuclear war has median timing of around 10 years, we are unlikely to find ourselves so late. Every-100-years-nuclear war hypothesis seems to be more likely in such a situation than every 10 years hypothesis. Every-1000-years nuclear war again does not look probable because now the question arises: why we are so yearly? In other words, if we have 3 hypotheses about the typical frequency of nuclear war: 10, 100, or 1000 years from the creation nuclear weapons and assume that there are no other x-risks, then finding ourselves 75 years after the creation of nuclear weapons supports the 100-years hypothesis.

Even if nuclear war is not killing all the people, it will significantly reduce the number of qualified observers, as the biggest university centers will be affected and Internet will not appear. Therefore, nuclear war could be regarded as equal to the extinction of qualified observers. 

Bostrom’s argument against the frequent false vacuum decay is based on the relatively late time of Earth formation in the history of our galaxy (Tegmark & Bostrom, 2005). It is also an example of the “early observerhood” counterargument. Its logic can’t be applied to the catastrophes on Earth, as we don’t have evidence that life on Earth can evolve quicker than it did; moreover, we have opposite observations about the frequency of evolutionary transitions on Earth.

4.3. Gaia counterargument

The hypothesis of Gaia (Lovelock & Lovelock, 2000) suggests that climate can self-regulate via negative feedback loops and that earth life is an important part of this self-regulation. An example of such feedback loop may be the growth of the Earth’s albedo because of the larger amount of clouds if temperature grows too much. Life plays important role in the self-regulation of climate as it is able to capture CO2.

The appearance of Gaia also could be explained by anthropic selection (Tyrrell, 2020; Watson, 2004), as only planets capable of self-regulating climate have preserved their habitability. Gaia cannot be explained via the Darwinian selection on Earth as was suggested by (Doolittle, 2019), as Gaia exists only in one example. The planets with life that did not evolve a homeostatic mechanism of climate regulation would not exist for long and therefore would not be capable of producing observers. 

Gaia seems to protect against anthropic fragility of climate, as most of the anthropic selection for stability has already happened in the past and has produced a self-stabilizing mechanism, now independent of anthropics. However, it may be that Gaia is just an observational selection illusion: there is no self-regulation, but there is just a series of random events which helped our survival.

Generalizing the Gaia counterargument, one may say: all anthropic selection has happened in the past and this has selected a system that is very stable and has homeostatic mechanisms supporting its stability to a wide range of perturbations. The theory may claim that homeostatic systems are rare, but eventually, they become more numerous than the worlds which just randomly escape all possible dangers. E.g. Jupiter is like Gaia for asteroid collisions, as it cleans space from hazardous comets.

Tyrrell demonstrated that only some planets get Gaia-like mechanism of homeostasis, and such mechanisms still have limits, above which these planets go into runaway warming (Tyrrell, 2020). 

The main objection to this counterargument is that any self-regulation has its limits. The Sun will eventually overheat the Earth, which is typically estimated to happen around 1 billion years from now if one does not account for anthropic effects and feedback climate loops, which could imply earlier runaway global warming. 

Biological organisms also age, become fragile and die and Gaia can age too. Thus, Gaia also may be unable to cope with some unexpected blows when “planetary boundaries” are exceeded (Baum & Handoh, 2014) and may stop protecting us from climate change. 

Unique anthropogenic actions could unexpectedly end Gaia, as its protection works only for natural variations. Human-caused deforestation and other interventions in nature could affect Gaia coping ability and increase the probability of greenhouse catastrophe as was discussed in “Role of the Biosphere in the Formation of the Earth’s Climate: The Greenhouse Catastrophe” by (Karnaukhov, 2001).

4.4. PETM argument against runaway global warming

The PETM episode of global warming, when 55 million years ago global temperatures jumped 8 C (or even 15C according to recent research), probably because of methane eruption, could be presented as a counterargument to the danger of runaway global warming, as temperatures returned to normal.

However, the main idea of this article is that one cannot use evidence from the past as arguments for our future survival because of anthropic bias. Maybe the PETM had a 99 per cent chance to turn into runaway global warming, but one cannot observe this, so we observe only the timeline (or Everett branch) where this did not happen. Also, the situation during PETM was different when now: different amounts of methane deposits, different speeds of warming and different disposition of continents. 

During the last interglacial period 130 ky ago, the temperatures were also warmer than now by 2C and this did not trigger methane-driven runaway global warming. But as we discussed above, the methane-driven warming will be significant only if the speed of the initial CO2-driven warming is high, because of the short lifetime of methane in the atmosphere. So, the condition then was different than now, and thus it is not proof that we are safe now. 


5. Estimating the power of anthropic shadow on Earth

5.1. Three types of power of anthropic shadow

The power of anthropic shadow – that is, the chances of our survival until now – is not very important, as very different anthropic shadows give only 1-2 orders of magnitude reduction of future life expectancies, as we show in section 2. Such difference doesn’t strongly affect our decision-making. As we assume that fragility increase is proportional to the life expectancy decrease, it means that fragility variation is not very large.

Based on the power of anthropic shadow, we could distinguish 3 significantly different situations:

  • anthropic shadow (ASH) doesn’t exist at all.
  • ASH is weak, from one to few orders of magnitude. Future life expectancy decrease is around one order of magnitude and is not important for us in case of most natural catastrophes. Anthropic fragility is present but manageable. This type of ASH is favored by SIA counterargument. For example, 1 in 1010 chances of past catastrophes give 100 million years of future life expectancy instead of 1 billion years. 10 times increase in fragility means that, say, not 45C is needed to reach moisture greenhouse, but only 4.5C, which is still manageable by emission control. 
  • ASH is strong, many orders of magnitude. The decline of life expectancy of the biosphere is significant. The world is so fragile to human actions that the fatal damage is likely already happened. This type of ASH is favored by evolutionary transitions argument discussed below. For example, 1 in 10100 ASH gives only around 12 million years of life expectancy and two order of magnitude anthropic fragility. This means that only around 0.5C of anthropogenic temperature increases is enough for the start of runaway global warming, and we already past that point. 

5.2. The similar size of the positive and negative fine-tuning

A recent article “The Timing of Evolutionary Transitions Suggests Intelligent Life Is Rare” by Sandberg et al (2020) shows that given a known future life expectancy for Earth's climate of 1 billion years, all evolutionary transitions could have a very small probability, or, in other words, very long mean times, maybe many orders of magnitude longer than the age of the universe. This means a very strong anthropic selection effect, since we are observing a world in which all such events occurred in time. 

We suggest that strong anthropic selection for evolutionary transitions means that the that an equally strong anthropic effect existed in avoiding natural disasters. Although this is not easy to prove, we can illustrate this with the concept of "budget: if a person has an expensive house, he is likely also has an expensive car, since the expensive house suggests that the person has a large budget. 

Similarly, powerful anthropic effects in evolutionary transitions may suggest the existence of a large "anthropic budget", that is, the actual existence of a large number of different worlds from which our world was chosen at random. The reason for this is that the Earth's longer survival time allows more time for evolutionary transitions.

Therefore, if it is easier to “buy” more time for the stable existence of a planet than to accelerate evolutionary transitions, then anthropic selection will choose planets with a longer existence without catastrophes. This will result in situation in which gaining time becomes as difficult as accelerating the evolutionary transition. Thus, there is a trade-off between avoiding catastrophes by pure chance and the speed of evolutionary transitions. For example, if the Earth were habitable not for 5, but for 10 billion years, this would allow more time for transitions, but would require much more incredibly long survival without catastrophes. The details of this trade-off are beyond the scope of this article and merit further study.

The similarity between anthropic shadow and evolution transitions probabilities is not exact, and it is not even of the same order of magnitude. It could be 1 in 10200 chances of intelligent life per planet and 1 in 10180 that a planet survives intact all risks of sterilizing catastrophes, like large asteroid impacts and runaway warmings. The second probability is trillions of trillions times higher, but still, they are similar in some sense. 

In addition, the evaluation of fine-tuning by Sandberg et al. (2020) refer to global base rate, i.e. it is a probability distribution for all terrestrial planets in the Universe, not just for the Earth. If there is a region of the universe where evolutionary transitions are much easier, we will be there. Therefore, the estimation of the risk’s frequencies is also a base rate applicable to all planets, and thus SIA counterargument (section 4.1) can’t kill it. This means that everywhere in the universe there is a strong base rate of sterilizing catastrophes.

5.3. Anthropic bias and the hard steps in the evolution

As we showed above, the power of anthropic shadow is comparable with the difficulty of evolutionary transitions, and one more estimation of this difficulty is presented in the article by Kipping.

The article “An objective Bayesian analysis of life’s early start and our late arrival” (Kipping, 2020) compared the probability of abiogenesis, which took around 100-300 million years, and the arrival of intelligence, which required 3.8 billion years after the appearance of life and concluded that intelligence appearing was a more difficult and less probable event. But the abiogenesis itself is estimated to be a very improbable event: a recent estimate of the minimum length of self-replicating RNA is around 100 bases (Totani, 2020). Totani thinks that only one of 10^100 stars will generate a viable RNA strand. Thus, only one of 10^80 of Hubble volumes of the universe would have life. But intelligence would be even less probable, based on Kipping’s logic. 

If intelligence is a rare step, it could be explained if the typical habitability of planets is shorter than the one of Earth and thus Earth is long overdue to lose its habitability. Thus, the world’s end is nigh, though we still may have around 100 million years. But anthropic fragility means this situation is worse as any significant change will cause a catastrophe.

However, this estimate may be inaccurate due to natural interstellar panspermia (Ginsburg et al., 2018), since in the case of panspermia, life may take much longer to develop. Life may have evolved on other planets before coming to Earth. For anthropic reasons, universes with panspermia will create more observers, as an entire galaxy with billions of potentially habitable planets(Hsu et al., 2019) could be fertilized with life (Turchin, 2020).


5.4. How could one know about the existence of the anthropic shadow? 

Anthropic shadow may be hinted by several types of evidence, but its main feature is that it is mostly unobservable. These types of evidence are:


  • Frequency of near-misses. One indication is the frequency of near-misses, or situations of improbable survival in the past. There is currently unpublished work “Nuclear war near misses and anthropic shadows” on the topic by Sandberg et al (Sandberg et al., 2018). For climate, it is several past Snow Ball situations, and one potential runaway greenhouse event 55 mln years ago (PETM), which didn’t turn permanent.
  • Absence of the fat tail. Another indication of anthropic shadow is the absence of a fat tail in the distribution of smaller catastrophes. For example, we observed asteroid impacts of bodies only below some size; there was no 100 km bodies impacts in last couple of billion years on Earth. We also never observed that the Earth becomes too hot, like 60 C, as it seems to be irreversible tipping point for runaway moisture greenhouse.
  • Surprising coincidences. A further evidence is some surprising coincidences which helped survival and evolution of life on Earth, as was analyzed in Sandberg et al (Snyder-Beattie et al., 2020). One of such evidence is the ability of Earth to have habitable temperature despite the growth of Sun luminosity.
  • Observation that we are located near the end of several period catastrophic cycles. This is a situation where one finds oneself closer to the end of stability period, or even some signs of the end of stable period could be observed. For example, asteroid impacts’ waves seem to be semi-periodic events every 30 or so million years, probably cause by disturbance of Oort cloud from passing of Galactic plane.
  • Increasing instability. Smaller catastrophes could be a sign that a larger one is near, e.g. increase of smaller asteroid impacts. 
  • Mathematical modelling of the similar to the Earth planets with different initial conditions (Tyrrell, 2020). This could provide a priori probability of anthropic shadow.
  • General considerations about Rare Earth and Fermi paradox.

We will continue discussion about the different types of evidence for anthropics shadow in Appendix 3.

6. Discussion: anthropic shadow and its connection with the evolution of intelligence

6.1. Hypothesis: intelligence is more likely to appear in an unstable world that is close to its end

Above we discussed a weak version of the anthropic shadow. But in the same way, as the strong anthropic principle claims that the universe needs to create observers (Barrow & Tipler, 1986), we could formulate a strong anthropic shadow

Intelligence tends to appear only in a world that is close to its end.

The main reason for the emergence of intelligence near the end of the world is that intelligence is a general adaptation that outperforms specialized adaptations in a rapidly changing world, which in the case of Earth is a world with an unstable climate. During Homo sapiens’ evolution, humans’ predecessors changed several main ways of feeding in just a few million years (Henry, 2018). Our hominid ancestors started out as arboreal primates, then evolved into savannah scavengers, then into foragers, then—maybe—“aquatic apes” then fire-cooking hunters, then plants-eating early agricultural inhabitants. communities, and then again to the high-calorie eaters of modern cities. Periodic glaciations and deglaciations in the last few million years have affected human living conditions and create the need to adapt to new feeding opportunities.

The second reason that the end is near is that the evolution of intelligent observers requires unusually long periods of relative stability: thus, a very large catastrophe may be overdue. This may seem to contradict what we said above, but here we are referring to much larger catastrophes that can kill all humans or irreversibly destroy civilization, such as moisture greenhouse.

Intelligent observers (scientists) are much more fragile than the entire biosphere and even the species Homo sapiens. A large economic collapse could reduce their numbers, e.g. as it happened after the collapse of the USSR. This means, as Circovic et al. wrote, that anthropic shadow must be stronger in recent times and suppress even smaller catastrophes (Ćirković et al., 2010). This includes smaller asteroid impacts, supervolcanic eruptions and nuclear war.

The third reason is that the impact of human civilization on nature is growing exponentially. If there is some hidden vulnerability (Bostrom, 2018) that can provoke a global catastrophe, humanity will stumble upon it sooner or later, especially since the anthropic shadow does not let us know where such a vulnerability is. The anthropic shadow also results in underestimation of the fragility of our environment to small anthropogenic changes.

Waltham came to similar conclusions: 


Given this link between climate change and species diversity, it is plausible that planets with high climate variability may be less likely to produce intelligent observers than planets with more stable conditions. However, it is also arguable that the ultimate emergence of intelligent species is actually encouraged by adverse conditions because these help to clear ecological niches (cf. the adaptive radiation of mammals following demise of the dinosaurs) and because evolutionary innovations may be particularly advantageous during testing times [cf. the emergence of Homo sapiens during the relatively unstable Neogene … and the emergence of multicellular life around the time of the Neoproterozoic glaciation] (Waltham, 2011).

In addition, human actions are unique and different from events that have occurred naturally in the past, and the homeostatic mechanisms that exist in nature may not be adapted to the new type of change. This includes the speed of change in the case of global warming and possibly a unique combination of several anthropogenic factors.

Another argument for a strong anthropic shadow is the Doomsday argument, which suggests a high level of future unknown risks based on our early location in human history (Bostrom, 2012; Turchin, 2018), and the general abundance of global catastrophic risks in our world (Bostrom, 2002). 

In a sense, the anthropic shadow is similar to the idea of so-called quantum immortality where a person observes oneself surviving many rounds of Russian roulette because of the many world interpretation of quantum mechanics (Turchin, 2018) if this idea is applied to humanity’s past. Even if humanity is in a simulation, most simulations likely model a civilization near its end, and will terminate after modeling the “singularity” (Greene, 2018; Turchin, Yampolskiy, Denkenberger, & Batin, 2019).

Similar idea was suggested by global warming research pioneer Budyko who discovered that cephalization of animals grew in the periods of unstable climate of Ice ages when forest and savanna replaced each other with periodicity of tens of thousands of years and universal adaptive ability – intelligence – was in demand. 


6.2. Relation of the anthropic fragility to Fermi paradox and x-risks

A strong anthropic shadow means that we are alone in the universe, so there is less risk of encountering hostile aliens. And vice versa, the observation that we alone, hints at a higher frequency of natural catastrophes that we have never seen due to the anthropic shadow.

Therefore, there is an increase of catastrophic risks in both branches of possibility: if we are alone, then we are exposed to higher natural risks and to anthropic fragility. If we are not alone, then the risk of hostile aliens is higher (Turchin, 2018c; Turchin & Denkenberger, 2019). 

Anthropic fragility puts natural risks into the short-term perspective of the next 100 years. Therefore, they become comparable with the most serious technological risks: synthetic biology, AI and nuclear war.


The anthropic shadow may seem like a purely theoretical effect, as far greater global risks pose threats to human civilization, such as unaligned AI, nuclear war, and synthetic biology (Turchin & Denkenberger, 2018). 

However, the detailed analysis presented here shows that anthropic shadow means that there are natural catastrophes that may be long overdue, have accumulated destructive energy, and could be triggered by small human actions. Such human actions will be unique events capable to trigger a change in semi-stable conditions. The most dangerous of these catastrophes is the runaway global warming that could occur if some currently unknown threshold of warming levels is reached.


Adler, R. J., Casey, B., & Jacob, O. C. (1995). Vacuum catastrophe: An elementary exposition of the cosmological constant problem. American Journal of Physics, 63, 620–626.

Allan, R. P., Hawkins, E., Bellouin, N., & Collins, B. (2021). IPCC, 2021: Summary for Policymakers.

Alvarez, L. W., Alvarez, W., Asaro, F., & Michel, H. V. (1980). Extraterrestrial cause for the Cretaceous-Tertiary extinction. Science, 208(4448), 1095–1108.

Ananthaswamy, A. (2015). The methane apocalypse. New Scientist, 226(3022), 38–41.

Armstrong, S., & Sandberg, A. (2013). Eternity in six hours: Intergalactic spreading of intelligent life and sharpening the Fermi paradox. Acta Astronautica, 89, 1–13.

Bailer-Jones, C. a. L. (2015). Close encounters of the stellar kind. Astronomy & Astrophysics, 575, A35.

Barr, A. C. (2016). On the origin of Earth’s Moon. Journal of Geophysical Research: Planets, 121(9), 1573–1601.

Barrow, J. D., & Tipler, F. J. (1986). The Anthropic Cosmological Principle (Clarendon. Oxford.

Baum, S. D., & Handoh, I. C. (2014). Integrating the planetary boundaries and global catastrophic risk paradigms. Ecological Economics, 107, 13–21.

Baum, S. D., Haqq-Misra, J. D., & Domagal-Goldman, S. D. (2011). Would contact with extraterrestrials benefit or harm humanity? A scenario analysis. Acta Astronautica, 68(11–12), 2114–2129.

Bostrom, N. (2001). The Doomsday Argument Adam & Eve, UN++, and Quantum Joe. Synthese, 127(3), 359–387.

Bostrom, N. (2002). Existential risks: Analyzing Human Extinction Scenarios and Related Hazards. Journal of Evolution and Technology, Vol. 9, No. 1 (2002).

Bostrom, N. (2012). A Primer on the Doomsday Argument. Anthropic-Principle.Com. /?q=anthropic_principle/doomsday_argument

Bostrom, N. (2013). Anthropic bias: Observation selection effects in science and philosophy. Routledge.

Bostrom, N. (2018). The Vulnerable World Hypothesis. 38.

Bostrom, N., & Ćirković, M. M. (2003). The doomsday argument and the self-indication assumption: Reply to Olum. The Philosophical Quarterly, 53(210), 83–91.

Bowen, G. J., Maibauer, B. J., Kraus, M. J., Röhl, U., Westerhold, T., Steimke, A., Gingerich, P. D., Wing, S. L., & Clyde, W. C. (2015). Two massive, rapid releases of carbon during the onset of the Palaeocene–Eocene thermal maximum. Nature Geoscience, 8(1), 44.

Bridges, W. (2012). Gains from Getting Near Misses Reported. Presentation at 8th Global Congress on Process Safety, Houston TX April, 1–4.

Carpenter, P. A., & Bishop, P. C. (2009). A review of previous mass extinctions and historic catastrophic events. Futures, 41(10), 676–682.

Carter, B. (1974). Large number coincidences and the anthropic principle in cosmology. Symposium-International Astronomical Union, 63, 291–298.

Chapman, C. R., & Morrison, D. (1994). Impacts on the Earth by asteroids and comets: Assessing the hazard. Nature, 367(6458), 33–40.

Cirkovic, M. M., & Cathcart, R. (2003). Geo-engineering Gone Awry: A New Partial Solution of Fermi’s Paradox. ArXiv Preprint Physics/0308058.

Ćirković, M. M., Sandberg, A., & Bostrom, N. (2010). Anthropic shadow: Observation selection effects and human extinction risks. Risk Analysis, Vol. 30, No. 10, 2010.

Ćirković, M. M., & Vukotić, B. (2016). Long-term prospects: Mitigation of supernova and gamma-ray burst threat to intelligent beings. Acta Astronautica, 129, 438–446.

Cui, Y., Schubert, B. A., & Jahren, A. H. (2020). A 23 m.y. Record of low atmospheric CO2. Geology.

Dar, A., De Rújula, A., & Heinz, U. (1999). Will relativistic heavy-ion colliders destroy our planet? Physics Letters B, 470(1–4), 142–148.

Dean, J. F., Middelburg, J. J., Röckmann, T., Aerts, R., Blauw, L. G., Egger, M., Jetten, M. S., Jong, A. E., Meisel, O. H., & Rasigraf, O. (2018). Methane feedbacks to the global climate system in a warmer world. Reviews of Geophysics.

Doolittle, W. F. (2019). Making Evolutionary Sense of Gaia. Trends in Ecology & Evolution, 34(10), 889–894.

Douglas, M. R. (2003). The statistics of string/M theory vacua. Journal of High Energy Physics, 2003(05), 046–046.

Firestone, R. B., West, A., Kennett, J. P., Becker, L., Bunch, T. E., Revay, Z. S., Schultz, P. H., Belgya, T., Kennett, D. J., & Erlandson, J. M. (2007). Evidence for an extraterrestrial impact 12,900 years ago that contributed to the megafaunal extinctions and the Younger Dryas cooling. Proceedings of the National Academy of Sciences, 104(41), 16016–16021.

Flis, A. (2020, October 26). Arctic Sea Ice is not freezing In October for the first time since measurements began, now having an unknown effect on weather development towards Winter. Severe Weather Europe.

Gertz, J. (2016). Reviewing METI: A critical analysis of the arguments. ArXiv Preprint ArXiv:1605.05663.

Ginsburg, I., Lingam, M., & Loeb, A. (2018). Galactic panspermia. The Astrophysical Journal Letters, 868(1), L12.

Goff, J., Dominey-Howes, D., Chagué-Goff, C., & Courtney, C. (2010). Analysis of the Mahuika comet impact tsunami hypothesis. Marine Geology, 271(3–4), 292–296.

Goldblatt, C., Robinson, T. D., Zahnle, K. J., & Crisp, D. (2013). Low simulated radiation limit for runaway greenhouse climates. Nature Geoscience, 6(8), 661–667.

Goldblatt, C., & Watson, A. J. (2012). The runaway greenhouse: Implications for future climate change, geoengineering and planetary atmospheres. Phil. Trans. R. Soc. A, 370(1974), 4197–4216.

Gott III, J. R. (1993). Implications of the Copernican principle for our future prospects. Nature, 363, 315–319.

Gowdy, J. (2020). Our hunter-gatherer future: Climate change, agriculture and uncivilization. Futures, 115, 102488.

Grace, K. (2010). SIA doomsday: The filter is ahead | Meteuphoric. Meteuphoric.

Greene, P. (2018). The Termination Risks of Simulation Science. Erkenntnis, 1–21.

Halstead, J. (2018). Stratospheric Aerosol Injection Research and Existential Risk. Futures.

Handoh, I. C. (2013). Phosphorus and Chemical Pollution as Global Catastrophic Risks. GCRI Blog.

Hansen, J., & et al. (1992). ‪Climate Forcing by Anthropogenic Aerosols. Science.

Hanson, R. (1998). The great filter-are we almost past it. Preprint Available at Http://Hanson. Gmu. Edu/Greatfilter. Html.

Hanson, R., Martin, D., McCarter, C., & Paulson, J. (2021). A Simple Model of Grabby Aliens. ArXiv Preprint ArXiv:2102.01522.

Hartogh, P., Lis, D. C., Bockelée-Morvan, D., de Val-Borro, M., Biver, N., Küppers, M., Emprechtinger, M., Bergin, E. A., Crovisier, J., & Rengel, M. (2011). Ocean-like water in the Jupiter-family comet 103P/Hartley 2. Nature, 478(7368), 218–220.

Henry, D. O. (2018). From foraging to agriculture: The Levant at the end of the Ice Age. University of Pennsylvania Press.

Herndon, J. M. (1993). Feasibility of a nuclear fission reactor at the center of the Earth as the energy source for the geomagnetic field. Journal of Geomagnetism and Geoelectricity, 45(5), 423–437.

Hsu, D. C., Ford, E. B., Ragozzine, D., & Ashby, K. (2019). Occurrence Rates of Planets Orbiting FGK Stars: Combining Kepler DR25, Gaia DR2, and Bayesian Inference. The Astronomical Journal, 158(3), 109.

Hut, P., & Rees, M. J. (1983). How stable is our vacuum? Nature, 302(5908), 508.

Karnaukhov, A. V. (2001). Role of the biosphere in the formation of the Earth’s Climate: The Greenhouse Catastrophe. BIOPHYSICS-PERGAMON THEN MAIK NAUKA-C/C OF BIOFIZIKA, 46(6), 1078–1088.

Kelsey, H. M., Nelson, A. R., Hemphill-Haley, E., & Witter, R. C. (2005). Tsunami history of an Oregon coastal lake reveals a 4600 yr record of great earthquakes on the Cascadia subduction zone. Geological Society of America Bulletin, 117(7–8), 1009–1032.

Kennett, J. P., Cannariato, K. G., Hendy, I. L., & Behl, R. J. (2003). Methane Hydrates in Quaternary Climate Change: The Clathrate Gun Hypothesis. American Geophysical Union.

Kent, A. (2004). A critical look at risk assessments for global catastrophes. Risk Analysis, 24(1), 157–168.

Kipping, D. (2020). An objective Bayesian analysis of life’s early start and our late arrival. Proceedings of the National Academy of Sciences.

Kokaia, G., & Davies, M. B. (2019). Stellar encounters with giant molecular clouds.

Kump, L. R., Pavlov, A., & Arthur, M. A. (2005). Massive release of hydrogen sulfide to the surface ocean and atmosphere during intervals of oceanic anoxia. Geology, 33(5), 397–400.

Lenton, T. M. (2011). Early warning of climate tipping points. Nature Climate Change, 1(4), 201.

Lingam, M., & Loeb, A. (2017). Risks for life on habitable planets from superflares of their host stars. The Astrophysical Journal, 848(1), 41.

Lomborg, B. (2020). Welfare in the 21st century: Increasing development, reducing inequality, the impact of climate change, and the cost of climate policies. Technological Forecasting and Social Change, 156, 119981.

Lovelock, J., & Lovelock, J. E. (2000). Gaia: A new look at life on earth. Oxford Paperbacks.

Manheim, D. (2018). Questioning Estimates of Natural Pandemic Risk. Health Security, 16(6), 381–390.

Marquis De Laplace, P. (1814). A philosophical essay on probabilities. Cosimo, Inc., 2007.

Napier, B., Asher, D., Bailey, M., & Steel, D. (2015). Centaurs as a hazard to civilization. Astronomy & Geophysics, 56(6), 6–24.

Napier, W. M., & Clube, S. V. M. (1979). A theory of terrestrial catastrophism. Nature, 282(5738), 455.

Ord, T. (2020). The precipice: Existential risk and the future of humanity. Hachette Books.

Ord, T., Hillerbrand, R., & Sandberg, A. (2010). Probing the improbable: Methodological challenges for risks with low probabilities and high stakes. Journal of Risk Research, 13(2), 191–205.

Ozhovan, M. I., Gibb, F., Poluektov, P. P., & Emets, E. P. (2005). Probing of the interior layers of the Earth with self-sinking capsules. Atomic Energy, 99(2), 556–562.

Popp, M., Schmidt, H., & Marotzke, J. (2016). Transition to a Moist Greenhouse with CO2 and solar forcing. Nature Communications, 7.

Ramirez, R. M., Kopparapu, R. K., Lindner, V., & Kasting, J. F. (2014). Can increased atmospheric CO2 levels trigger a runaway greenhouse? Astrobiology, 14(8), 714–731.

Ramirez, R. M., Lindner, V., & Kasting, J. F. (2013). How close is Earth to a runaway greenhouse? ArXiv Preprint ArXiv:1306.5730.

Rampino, M. R. (1998). The galactic theory of mass extinctions: An update. In Dynamics of Comets and Asteroids and Their Role in Earth History (pp. 49–58). Springer.

Rampino, M. R. (2008). Super-volcanism and other geophysical processes of catastrophic import. Global Catastrophic Risks, 1, 203.

Rampino, M. R. (2015). Disc dark matter in the Galaxy and potential cycles of extraterrestrial impacts, mass extinctions and geological events. Monthly Notices of the Royal Astronomical Society, 448(2), 1816–1820.

Rampino, M. R., & Caldeira, K. (2015). Periodic impact cratering and extinction events over the last 260 million years. Monthly Notices of the Royal Astronomical Society, 454(4), 3480.

Reinhold, T., Shapiro, A. I., Solanki, S. K., Montet, B. T., Krivova, N. A., Cameron, R. H., & Amazo-Gómez, E. M. (2020). The Sun is less active than other solar-like stars. Science, 368(6490), 518–521.

Sandberg, A. (2008). Bayes, Moravec and the LHC: Quantum Suicide, Subjective Probability and Conspiracies. Andart.

Sandberg, A., Drexler, E., & Ord, T. (2017). Dissolving the Fermi Paradox. Future of Humanity Institute.

Sandberg, A., Snyder-Beattie, A., & Armstrong, S. (2018). Nuclear war near misses and anthropic shadows. Manuscript in Preparation.

Seeley, J., & Wordsworth, R. (2021). Episodic deluges in simulated hothouse climates. Nature, 599(7883), 74–79.

Segschneider, J., Beitsch, A., Timmreck, C., Brovkin, V., Ilyina, T., Jungclaus, J. H., Lorenz, S. J., Six, K. D., & Zanchettin, D. (2013). Impact of an extremely large magnitude volcanic eruption on the global climate and carbon cycle estimated from ensemble Earth System Model simulations. Biogeosciences, 10, 669–687.

Shakhova, N., Semiletov, I., Salyuk, A., Yusupov, V., Kosmach, D., & Gustafsson, Ö. (2010). Extensive methane venting to the atmosphere from sediments of the East Siberian Arctic Shelf. Science, 327(5970), 1246–1250.

Shcherbakov, А. С. (1999). Антропный принцип в космологии и геологии. Вестник Московского Университета. Серия, 7, 58–70.

Snyder-Beattie, A. E., Ord, T., & Bonsall, M. B. (2019). An upper bound for the background rate of human extinction. Scientific Reports, 9(1), 11054.

Snyder-Beattie, A. E., Sandberg, A., Drexler, K. E., & Bonsall, M. B. (2020). The Timing of Evolutionary Transitions Suggests Intelligent Life Is Rare. Astrobiology.

Song, H., Kemp, D. B., Tian, L., Chu, D., Song, H., & Dai, X. (2021). Thresholds of temperature change for mass extinctions. Nature Communications, 12(1).

Strom, R. G., & Schaber, G. G. (1992). Pulsed Resurfacing Events on Venus, Earth and Mars. Bulletin of the American Astronomical Society, 24, 946.

Strom, R. G., Schaber, G. G., & Dawson, D. D. (1994). The global resurfacing of Venus. Journal of Geophysical Research: Planets, 99(E5), 10899–10926.

Tegmark, M., & Bostrom, N. (2005). How unlikely is a doomsday catastrophe? ArXiv:Astro-Ph/0512204.

Thrasymachus, K. (2012). UFAI cannot be the Great Filter.

Totani, T. (2020). Emergence of life in an inflationary universe. Scientific Reports, 10(1), 1671.

Trigo-Rodríguez, J. M., & Williams, I. P. (2017). Dynamic sources of contemporary hazard from meteoroids and small asteroids. In Assessment and Mitigation of Asteroid Impact Hazards (pp. 11–32). Springer.

Turchin, A. (2018a). A Meta-Doomsday Argument: Uncertainty About the Validity of the Probabilistic Prediction of the End of the World.

Turchin, A. (2018b). Forever and Again: Necessary Conditions for the “Quantum Immortality” and its Practical Implications.

Turchin, A. (2018c). The Risks Connected with Possibility of Finding Alien AI Code During SETI. Journal of  British Interplanetary Society, 70.

Turchin, A. (2020). Presumptious philosopher proves panspermia.

Turchin, A., & Denkenberger, D. (2018). Global Catastrophic and Existential Risks Communication Scale. Futures, 102, 27–38.

Turchin, A., & Denkenberger, D. (2019). Classfication of ETI riks. Inder Review in JBIS.

Turchin, A., Yampolskiy, R., Denkenberger, D., & Batin, M. (2019). Simulation Typology and Termination Risks.

Turner, M. G., Romme, W. H., & Tinker, D. B. (2003). Surprises and lessons from the 1988 Yellowstone fires. Frontiers in Ecology and the Environment, 1(7), 351–358.

Tyrrell, T. (2020). Chance played a role in determining whether Earth stayed habitable. Communications Earth & Environment, 1(1), 1–10.

USGS. (2020). Volcanoes Can Affect Climate. USGS.

Waltham, D. (2011). Testing anthropic selection: A climate change example. Astrobiology, 11(2), 105–114.

Waltham, D. (2014). On the absence of solar evolution-driven warming through the Phanerozoic. Terra Nova, 26(4), 282–286.

Waltham, D. (2019). Is Earth special? Earth-Science Reviews.

Ward, P. D. (2007). Under a green sky. Smithsonian Books/Collins.

Ward, P. D., & Brownlee, D. (2003). Rare Earth: Why Complex Life is Uncommon in the Universe. Copernicus.

Ward, P. D., & Brownlee, D. (2004). The life and death of planet Earth: How the new science of astrobiology charts the ultimate fate of our world. Macmillan.

Watson, A. J. (2004). Gaia and observer self-selection. Scientists Debate Gaia: The next Century, 201–210.

Wordsworth, R. (2021). How likely are Snowball episodes near the inner edge of the habitable zone? ArXiv:2104.06216 [Astro-Ph].

Yudkowsky, E., & Soares, N. (2017). Functional Decision Theory: A New Theory of Instrumental Rationality. ArXiv:1710.05060 [Cs].


New comment
25 comments, sorted by Click to highlight new comments since: Today at 4:35 PM

Thank you, this is fascinating stuff!

One quick question on the policy implication as I (as a climate philanthropist) often ask myself whether or not to fund geoengineering advocacy.

While your argument -- if correct -- pushes strongly in the direction of worlds where we would want to use geoengineering by making an emergency situation much more likely in expectation, even the lowest estimate you cite  in term of the transition (four years) still seems like plenty of time if the global community decided geo-engineering was  a global priority (like, say, in a way that vaccine development was for COVID-19). 

I.e., does it really change something in term of optimal action now re geo-engineering?

(I do think your argument has strong force re optimal climate policy more generally, I am just less sure it affects geo-engineering prioritization within the portfolio). 

I think, yes. We need a completely new science of "urgent geoengineering" - that is something like creating artificial nuclear winter by controlled fires in forests which will give us a few years of time to develop better methods or to reverse the dangerous trend.

I tried 6 years ago to create a more detailed plan (it may obsolete, but that is what I have) here it is a chart

and it is its explanation 

Thanks, will have a look!

I would appreciate your thoughts on the shape of the relationship between existential risk due to climate change and global warming. For example, would it be reasonable to assume it is linear, i.e. that the x-risk linked to an increase of 2 ºC relative to today's temperature is 2 times as large as the x-risk linked to an increase of 1 ºC?

We don't know where is the tipping point, so uninformed prior gives equal chances for any T between 0 and, say, 20 C additional temperature increase. In that case 2C is 2 times more likely. 

But the idea of anthorpic shadow tells us that tipining point is likely to be 10 per cent of the whole interval. And for 40C before moisture greenhouse it is 4C. But, interestingly, anthropic shadow tells us that smaller intervals are increasingly unlikely. So 1C  increase is orders of magnitude less likely to cause a catastrophe than 4 C increase.

I will illustrate this as following example:

Imagine you are buying a used car which had run 300K miles. It is a unique survivor for its age.

If the car had 1 in 1000 chance to survive until its age, then doubling period of the probability of death (aka half-life) for it is 30K miles (10 doublings); if it had 1 in 1 000 000 chances to survive until current age, it has 20 doubling or 15K miles. The 1000 times growth of anthropic shadow lower car's life expectancy only 2 times.

Future survival declines very slowly:

Anthropic shadow power 1 in 1000 = survival 10 per cent more

Anthropic shadow power 1 in 1000 000 = survival 5 per cent more.

Thanks for the reply!

Your calculations apply to an exponential distribution. Do we have reasons to choose an exponential prior over a uniform/loguniform prior for the location of the existential tipping point? I guess one possible disadvantage of the exponential prior is the lack of a maximum (which should arguably be assumed given our knowledge about moisture greenhouse), but this could be solved by using a truncated exponential.

I use exponential prior to illustrate the example with a car. For other catastrophes, I take the tail of normal distribution, there the probability declines very quickly, even hyperexponentially. The math there is more complicated. But it does not affect the main result: if we have anthropic shadow, the expected survival time is around 0.1 of the past time in the wide range of initial parameters. 

And in the situation of anthropic shadow we have very limited information about the type of distribution.  Exponential and normal seems to be two most plausible types for catastrophes. There is also semi-periodic ones, but they could be described as a sum of periodic plus normal. 

But obviously there is more to dig here.

Interesting post! 

I think the extent to which the following claim is true is quite important (emphasis mine):

However, the idea of anthropic fragility means that we should lower our estimates for around an order of magnitude, which, given all uncertainties, means that the tipping point could lie not in tens but in single digits of temperature increase (that is, between 1.5C and 4.5C, if we just divide on 10 the above estimate).

From what I understand, the above is based on the Appendix.

6. Over simplification of all said above is the following rule of thumb: Anthropic shadow lowers future life expectancy for no more than 1 order of magnitude in most plausible cases.

Even if this is true, why would the toy example of an old man apply to humanity?

Normally distributed anthropic shadow

I do not think the lifespan of humanity follows a normal distribution:

  • I think the life expectancy of humanity is at least of the order of 10^15 years:
    • Star formation will continue for 10^15 to 10^17 years (see here).
    • Toby Ord guesses in The Precipice that the total existential risk is 0.5.
    • So there is arguably a chance of 0.5 of humanity surviving at least 10^15 years, which means its life expectancy is at least 10^15 years.
  • However, the median lifespan of humanity seems to be very short:
    • Based on the estimate of the total x-risk of 0.5, there is a chance of 0.5 of humanity not surviving the Time of Perils.
    • Is such time ends in e.g. about 1 k years, the median lifespan of humanity is also about 1 k years.
  • The mean and median are similar for a normal distribution, but for humanity lifespan, I would say the mean is at least 10^(15 - 3) = 10^12 times as high as the median. So normality does not seem to apply.

Anthropic shadow applies not to humanity, but to underlying conditions on which we can survive. 

For example, the waves of asteroid bombardment are every 30 million years, but not exactly 30 mln. 

The next wave is normally distributed around 30 with mean deviation, say, 1 mln years. If 33 mln years have gone without it, it means that we are 3 sigmas after the mean. 

I see, thanks!

Image as a toy example a tense spring which is described by Hooke's law. Fs = kx.

Imagine also that we can observe only those springs that are tensed far beyond their normal breaking point = it is a model of anthropic shadow.

From logarithmic nature of the relation between remaining life expectancy and  the power (probability of past survival) of anthropic shadow follows that for almost any anthropic shadow the remaining life expectancy is between 5-20 per cent of past survival time, lets call it dA.

For a tensed spring it means that its additional length beyond the breaking point is around 5-20 percent of total length (with several linearity assumptions which  will not list here for simplicity).

Now, the fragility of the spring could also be measured in its additional length increase which will cause its breakdown, dL. 

dL can't be more than the dA, as dA is already improbable event, and dA is less than 5-20 per cent of total length of the spring. Therefore, dLis less than 0.05-0.2 of L.

This seems to work only for the situation when the relation of the main parameter L is linear with the force of tension. But even for non-linear parameters they could be approximated by linear ones near the point.

TL;DR: Fragility is delta L (increase) of the main parameter of the system which causing its catastrophe. Fragility is proportional to anthropic shadow in a system similar to tensed spring or overinflated ballon, but in most cases of anthropic shadow it is independent of initial parameters an is around 10 per cent change. In the case of climate, it is not very clear what is main parameter, but likely it is means temperature. 

Thanks for clarifying!

Let me see if I have understood your argument:

  • If the probability of having avoided existential catastrophe due to climate change until now is smaller than p_max = 0.1 %, the "half-warming" is smaller than HW_max = 0.100 (= -1/log2(p_max)). 
  • So, given the lack of memory of the exponential distribution, the mean additional warming until existential catastrophe due to climate change is smaller than 14.5 % (= HW_max/log(2) = 1/log(1/p_max)) of the maximum historical warming until now. 
  • Based on this article, temperature was 18 ºC (=(90-58)/1.8) higher than now 250 Myear ago. This means the existential additional warming is smaller than 2.61 ºC (=18*14.5 %). This is in agreement with your conclusion that "the tipping point could lie not in tens but in single digits of temperature increase (that is, between 1.5C and 4.5C, if we just divide on 10 the above estimate)".

However, why should the anthropic shadow be smaller than 0.1 %? 

  • As the anthropic shadow tends to 1, the existential warming tends to infinity.
  • Given that we are still here, I think the probability of 18 ºC of warming not having led to an existential catastrophe 250 Myear ago should be larger than 50 % (instead of smaller than 0.1 %). In this case, the existential additional warming relative to today's temperature would be larger than 26.0 ºC (= 18/log(1/0.5)) for an exponential prior.

From climate point of view, we need to estimate not only the warming, but also the speed of warming, as higher speed gives high concentration of methane (and this differential equation has exponential solution). Anthropogenic global warming is special as it has very high speed of CO2 emission never happened before. We also have highest ever  accumulation of methane hydrates. We could be past tipping point but do not know it yet, as exponential growth is slow in the beginning. 

From SIA counteragrument follows that anthropic shadow can't be very strong: we are unlikely to observe the world with a very strong anthropic shadow. However, some anthropic effects on climate likely to exist as we observe the  preservation of habitability  of the Earth despite changes а Sun luminosity. This gives us some range of values there anthropic shadow can be, and 0.1 per cent seems to be a reasonable estimate inside it. Though exact number or range is difficult to estimate. May be Sandberg's work on near-misses in nuclear war would help - when we will have a chance to see it. 

I feel that I didn't answer the whole your question, so can you point what exactly is your point of disagreement. 

I agree there is a difference between:

  • The current temperature (T0).
  • The maximum temperature which would be achieved if we reached net zero today (T1).

The 2nd of these is higher, so the lower bound for the existential additional warming is smaller than the 26.0 ºC I estimated above (for an anthropic shadow larger than 50 %). I also understand T1 may be a function of not only T0, but also of the current composition of the atmosphere, and the rate at which it has been changing. 

However, how large do you think is the difference between T0 and T1? If it is of the order of magnitude of the warming until now relative to pre-industrial levels of 1 ºC, there is still a margin of about 25.0 ºC (= 26.0 - 1) to the existential tipping point.

You mention that we may already have passed the existential tipping point, but that would imply a difference between T1 and T0 of more than 25.0 ºC, which seems very hard to believe.

I think that the difference between tipping point and existential temperature should be clarified. Tipping point is the temperature after which self-sustaining loop of positive feedback starts. In the moisture greenhouse paper it is estimated to be at +4C, after which the temperature jumps to +40C in a few years. If we take +4  C above preindustrial level, it will be 1-3 above current level. 

Thanks for clarifying. I had understood that difference, but for me it is unclear from what you discuss here that the tipping point is only 4 ºC above pre-industrial temperature. Could you link to the specific paper you are referring to?

"Transition to a Moist Greenhouse with CO2 and solar forcing”

Thanks. The results of that article cannot be applied directly to the situation we are in, because the initial temperature of their aqua-planet  is 6 ºC higher than today's mean global temperature. From note (6.93) of What We Owe to the Future (see here):

Hansen et al. 2013, 17. Popp et al. (2016) [the studied you linked to just above] found that if carbon dioxide concentrations reached 1,520 parts per million, a simulated planet would transition to a moist greenhouse state. If we burned all of the fossil fuels, then carbon dioxide concentrations would reach 1,600 parts per million (Lord et al. 2016, Figure 2).

However, the simulated planet’s initial climate was six degrees warmer than today’s Earth. This means that Earth would require a carbon dioxide concentration significantly higher than on the simulated planet to transition to a moist greenhouse.

Indeed, from the Discussion of the article you mention:

A recent study using the same model but in a different version found that the Earth’s climate remains stable for CO2 concentrations of at least 4,480 p.p.m. (ref. 17), whereas our study suggests that such concentrations would lead to a climate transition. Studies of Earth with other GCMs [global climate models] also found the climate to remain stable for higher CO2 concentrations than we do. 

However, the initial climate of our aqua-planet is ~6K warmer than the one of present-day Earth.


If we account for the difference in the initial climates, the results of the two studies are not in contradiction. Indeed, the climate of the model version used in ref. 17 was recently shown to become unstable when the CO2 concentrations were increased from 4,480 to 8,960 p.p.m.

These concentrations of 4,480 and 8,960 p.p.m are 16.0 (=4480/280) and 32.0 (=8960/280) times the pre-industrial concentration, which suggests the existential CO2 concentration is 22.6 (= (16.0*32.0)^0.5) times as high as the pre-industrial one. Given the warming until now relative to pre-industrial levels of 1.04 ºC, and the current concentration of CO2 is 1.48 (= 414/280) times the pre-industrial one, it seems reasonable to expect the existential warming relative to the pre-industrial temperature is about 20 ºC (22.6/1.48*1.04 = 15.9), not 4 ºC.

The relation between warming and CO2 is exponential, s we need to count the number of doublings of CO2. Every doubling gives a constant increase of the temperature. Assuming that each doubling gives 2C and 22= 2exp4.5, we get around 9C above preindustrial level before we reach tipping point.

 In the article the tipping point is above 4C (in the chart) plus 6C from warmer  world = 10C, which gives us approximately the same result as I calculated above. 

I agree with this assessment insofar as we might be far closer to climate tipping points than it at first seems. I am curious about how a society might use and recover from urgent geo-engineering.

It seems like any scenario where urgent geo-engineering is used would have drastic effects. Who do you imagine would make this decision? If it is the governments like the UN what efforts might be made to get them to seriously consider this proposal? If you Imagine that it would be individual actors this seems like it might do more harm than good insofar as it violates the principle of conformity.

On a separate note, it doesn't seem like it would be easy to recover from the use of urgent geo-engineering in such a way that doesn't put us past the tipping point again. It seems like the use of such technology would drastically reduce our ability to research after its use, as it seems like it would totally destroy the supply chain. I am curious if I am missing anything there. 

As a result, I wonder if urgent geoengineering might not be a solution even if we are very close to tipping points. 

If we create artificial nuclear winter - it could be created by one strong actor unilaterally. No coordination is needed. 

Such nuclear winter may last few years and naturally resolve to normality. During this process two things could happen:  either the tipping point conditions also stop, like methane leakage ends. Or we create more permanent solution to our problem like more stable form of geoengienering.

The artificial nuclear winter doesn't need to be very strong (in -2-3 C range), so no major disruption of food production will happen.

I understand it could be done by one strong actor unilaterally, I simply wonder if I could reasonably support such an action being taken unilaterally. This paper is what sold me on this position

I think you are overestimating what could be accomplished during this time period, I imagine that most people would become hostile to any movement which just intentionally triggered a nuclear winter. 

Do you have a source on how disruptive nuclear winters would be to food production, I am skeptical.


On an unrelated note, I see that you sight "Nuclear war near misses and anthropic shadows" which is marked as being in preparation. I wrote an essay that I imagine is similar titled "Nuclear fine-tuning". I am wondering if you have access to this document and if you could send it my way, as I would like to read it to see what gaps in my arguments it might fill in.

My essay can be found here:

If there will obvious global runaway global warming, like +6C everywhere and growing month by month, people will demand "do something" about it and will accept attempts to use nuclear explosions to stop it. 

I don't have access now to the document "Nuclear war near misses and anthropic shadows"

Ah, that's too bad, do you have the email of anyone who would?

Sanberg recently published its summary in twitter. he said that he uses the frequency of near-misses to estimate the power of anthropic shadow and found that near misses was not suppressed during the period of large nuclear stockpiles and  it is evidence against anthropic shadow. I am not sure that it is true, as in early times the policy was more risky.