Nick Bostrom

When we create the first superintelligent entity, we might make a mistake and give it goals that lead it to annihilate humankind, assuming its enormous intellectual advantage gives it the power to do so. For example, we could mistakenly elevate a subgoal to the status of a supergoal. We tell it to solve a mathematical problem, and it complies by turning all the matter in the solar system into a giant calculating device, in the process killing the person who asked the question.

3 thoughts on “Nick Bostrom

  1. shinichi Post author

    Existential Risks

    Analyzing Human Extinction Scenarios and Related Hazards

    by Nick Bostrom

    Because of accelerating technological progress, humankind may be rapidly approaching a critical phase in its career. In addition to well-known threats such as nuclear holocaust, the prospects of radically transforming technologies like nanotech systems and machine intelligence present us with unprecedented opportunities and risks. Our future, and whether we will have a future at all, may well be determined by how we deal with these challenges. In the case of radically transforming technologies, a better understanding of the transition dynamics from a human to a “posthuman” society is needed. Of particular importance is to know where the pitfalls are: the ways in which things could go terminally wrong. While we have had long exposure to various personal, local, and endurable global hazards, this paper analyzes a recently emerging category: that of existential risks. These are threats that could cause our extinction or destroy the potential of Earth-originating intelligent life. Some of these threats are relatively well known while others, including some of the gravest, have gone almost unrecognized. Existential risks have a cluster of features that make ordinary risk management ineffective. A final section of this paper discusses several ethical and policy implications. A clearer understanding of the threat picture will enable us to formulate better strategies.

  2. shinichi Post author

    We shall use the following four categories to classify existential risks:

    Bangs – Earth-originating intelligent life goes extinct in relatively sudden disaster resulting from either an accident or a deliberate act of destruction.

    Crunches – The potential of humankind to develop into posthumanity is permanently thwarted although human life continues in some form.

    Shrieks – Some form of posthumanity is attained but it is an extremely narrow band of what is possible and desirable.

    Whimpers – A posthuman civilization arises but evolves in a direction that leads gradually but irrevocably to either the complete disappearance of the things we value or to a state where those things are realized to only a minuscule degree of what could have been achieved.


    4    Bangs

    This is the most obvious kind of existential risk. It is conceptually easy to understand. Below are some possible ways for the world to end in a bang. I have tried to rank them roughly in order of how probable they are, in my estimation, to cause the extinction of Earth-originating intelligent life; but my intention with the ordering is more to provide a basis for further discussion than to make any firm assertions.

    4.1  Deliberate misuse of nanotechnology

    In a mature form, molecular nanotechnology will enable the construction of bacterium-scale self-replicating mechanical robots that can feed on dirt or other organic matter. Such replicators could eat up the biosphere or destroy it by other means such as by poisoning it, burning it, or blocking out sunlight. A person of malicious intent in possession of this technology might cause the extinction of intelligent life on Earth by releasing such nanobots into the environment.

    The technology to produce a destructive nanobot seems considerably easier to develop than the technology to create an effective defense against such an attack (a global nanotech immune system, an “active shield”). It is therefore likely that there will be a period of vulnerability during which this technology must be prevented from coming into the wrong hands. Yet the technology could prove hard to regulate, since it doesn’t require rare radioactive isotopes or large, easily identifiable manufacturing plants, as does production of nuclear weapons.

    Even if effective defenses against a limited nanotech attack are developed before dangerous replicators are designed and acquired by suicidal regimes or terrorists, there will still be the danger of an arms race between states possessing nanotechnology. It has been argued that molecular manufacturing would lead to both arms race instability and crisis instability, to a higher degree than was the case with nuclear weapons. Arms race instability means that there would be dominant incentives for each competitor to escalate its armaments, leading to a runaway arms race. Crisis instability means that there would be dominant incentives for striking first. Two roughly balanced rivals acquiring nanotechnology would, on this view, begin a massive buildup of armaments and weapons development programs that would continue until a crisis occurs and war breaks out, potentially causing global terminal destruction. That the arms race could have been predicted is no guarantee that an international security system will be created ahead of time to prevent this disaster from happening. The nuclear arms race between the US and the USSR was predicted but occurred nevertheless.

    4.2  Nuclear holocaust

    The US and Russia still have huge stockpiles of nuclear weapons. But would an all-out nuclear war really exterminate humankind? Note that: (i) For there to be an existential risk it suffices that we can’t be sure that it wouldn’t. (ii) The climatic effects of a large nuclear war are not well known (there is the possibility of a nuclear winter). (iii) Future arms races between other nations cannot be ruled out and these could lead to even greater arsenals than those present at the height of the Cold War. The world’s supply of plutonium has been increasing steadily to about two thousand tons, some ten times as much as remains tied up in warheads. (iv) Even if some humans survive the short-term effects of a nuclear war, it could lead to the collapse of civilization. A human race living under stone-age conditions may or may not be more resilient to extinction than other animal species.

    4.3  We’re living in a simulation and it gets shut down

    A case can be made that the hypothesis that we are living in a computer simulation should be given a significant probability. The basic idea behind this so-called “Simulation argument” is that vast amounts of computing power may become available in the future, and that it could be used, among other things, to run large numbers of fine-grained simulations of past human civilizations. Under some not-too-implausible assumptions, the result can be that almost all minds like ours are simulated minds, and that we should therefore assign a significant probability to being such computer-emulated minds rather than the (subjectively indistinguishable) minds of originally evolved creatures. And if we are, we suffer the risk that the simulation may be shut down at any time. A decision to terminate our simulation may be prompted by our actions or by exogenous factors.

    While to some it may seem frivolous to list such a radical or “philosophical” hypothesis next the concrete threat of nuclear holocaust, we must seek to base these evaluations on reasons rather than untutored intuition. Until a refutation appears of the argument presented in, it would intellectually dishonest to neglect to mention simulation-shutdown as a potential extinction mode.

    4.4  Badly programmed superintelligence

    When we create the first superintelligent entity, we might make a mistake and give it goals that lead it to annihilate humankind, assuming its enormous intellectual advantage gives it the power to do so. For example, we could mistakenly elevate a subgoal to the status of a supergoal. We tell it to solve a mathematical problem, and it complies by turning all the matter in the solar system into a giant calculating device, in the process killing the person who asked the question.

    4.5  Genetically engineered biological agent

    With the fabulous advances in genetic technology currently taking place, it may become possible for a tyrant, terrorist, or lunatic to create a doomsday virus, an organism that combines long latency with high virulence and mortality.

    Dangerous viruses can even be spawned unintentionally, as Australian researchers recently demonstrated when they created a modified mousepox virus with 100% mortality while trying to design a contraceptive virus for mice for use in pest control. While this particular virus doesn’t affect humans, it is suspected that an analogous alteration would increase the mortality of the human smallpox virus. What underscores the future hazard here is that the research was quickly published in the open scientific literature. It is hard to see how information generated in open biotech research programs could be contained no matter how grave the potential danger that it poses; and the same holds for research in nanotechnology.

    Genetic medicine will also lead to better cures and vaccines, but there is no guarantee that defense will always keep pace with offense. (Even the accidentally created mousepox virus had a 50% mortality rate on vaccinated mice.) Eventually, worry about biological weapons may be put to rest through the development of nanomedicine, but while nanotechnology has enormous long-term potential for medicine it carries its own hazards.

    4.6  Accidental misuse of nanotechnology (“gray goo”)

    The possibility of accidents can never be completely ruled out. However, there are many ways of making sure, through responsible engineering practices, that species-destroying accidents do not occur. One could avoid using self-replication; one could make nanobots dependent on some rare feedstock chemical that doesn’t exist in the wild; one could confine them to sealed environments; one could design them in such a way that any mutation was overwhelmingly likely to cause a nanobot to completely cease to function. Accidental misuse is therefore a smaller concern than malicious misuse.

    However, the distinction between the accidental and the deliberate can become blurred. While “in principle” it seems possible to make terminal nanotechnological accidents extremely improbable, the actual circumstances may not permit this ideal level of security to be realized. Compare nanotechnology with nuclear technology. From an engineering perspective, it is of course perfectly possible to use nuclear technology only for peaceful purposes such as nuclear reactors, which have a zero chance of destroying the whole planet. Yet in practice it may be very hard to avoid nuclear technology also being used to build nuclear weapons, leading to an arms race. With large nuclear arsenals on hair-trigger alert, there is inevitably a significant risk of accidental war. The same can happen with nanotechnology: it may be pressed into serving military objectives in a way that carries unavoidable risks of serious accidents.

    In some situations it can even be strategically advantageous to deliberately make one’s technology or control systems risky, for example in order to make a “threat that leaves something to chance”.

    4.7  Something unforeseen

    We need a catch-all category. It would be foolish to be confident that we have already imagined and anticipated all significant risks. Future technological or scientific developments may very well reveal novel ways of destroying the world.

    Some foreseen hazards (hence not members of the current category) which have been excluded from the list of bangs on grounds that they seem too unlikely to cause a global terminal disaster are: solar flares, supernovae, black hole explosions or mergers, gamma-ray bursts, galactic center outbursts, supervolcanos, loss of biodiversity, buildup of air pollution, gradual loss of human fertility, and various religious doomsday scenarios. The hypothesis that we will one day become “illuminated” and commit collective suicide or stop reproducing, as supporters of VHEMT (The Voluntary Human Extinction Movement) hope, appears unlikely. If it really were better not to exist (as Silenus told king Midas in the Greek myth, and as Arthur Schopenhauer argued although for reasons specific to his philosophical system he didn’t advocate suicide), then we should not count this scenario as an existential disaster. The assumption that it is not worse to be alive should be regarded as an implicit assumption in the definition of Bangs. Erroneous collective suicide is an existential risk albeit one whose probability seems extremely slight.

    4.8  Physics disasters

    The Manhattan Project bomb-builders’ concern about an A-bomb-derived atmospheric conflagration has contemporary analogues.

    There have been speculations that future high-energy particle accelerator experiments may cause a breakdown of a metastable vacuum state that our part of the cosmos might be in, converting it into a “true” vacuum of lower energy density. This would result in an expanding bubble of total destruction that would sweep through the galaxy and beyond at the speed of light, tearing all matter apart as it proceeds.

    Another conceivability is that accelerator experiments might produce negatively charged stable “strangelets” (a hypothetical form of nuclear matter) or create a mini black hole that would sink to the center of the Earth and start accreting the rest of the planet.

    These outcomes seem to be impossible given our best current physical theories. But the reason we do the experiments is precisely that we don’t really know what will happen. A more reassuring argument is that the energy densities attained in present day accelerators are far lower than those that occur naturally in collisions between cosmic rays. It’s possible, however, that factors other than energy density are relevant for these hypothetical processes, and that those factors will be brought together in novel ways in future experiments.

    The main reason for concern in the “physics disasters” category is the meta-level observation that discoveries of all sorts of weird physical phenomena are made all the time, so even if right now all the particular physics disasters we have conceived of were absurdly improbable or impossible, there could be other more realistic failure-modes waiting to be uncovered. The ones listed here are merely illustrations of the general case.

    4.9  Naturally occurring disease

    What if AIDS was as contagious as the common cold?

    There are several features of today’s world that may make a global pandemic more likely than ever before. Travel, food-trade, and urban dwelling have all increased dramatically in modern times, making it easier for a new disease to quickly infect a large fraction of the world’s population.

    4.10  Asteroid or comet impact

    There is a real but very small risk that we will be wiped out by the impact of an asteroid or comet.

    In order to cause the extinction of human life, the impacting body would probably have to be greater than 1 km in diameter (and probably 3 – 10 km). There have been at least five and maybe well over a dozen mass extinctions on Earth, and at least some of these were probably caused by impacts. In particular, the K/T extinction 65 million years ago, in which the dinosaurs went extinct, has been linked to the impact of an asteroid between 10 and 15 km in diameter on the Yucatan peninsula. It is estimated that a 1 km or greater body collides with Earth about once every 0.5 million years. We have only catalogued a small fraction of the potentially hazardous bodies.

    If we were to detect an approaching body in time, we would have a good chance of diverting it by intercepting it with a rocket loaded with a nuclear bomb.

    4.11  Runaway global warming

    One scenario is that the release of greenhouse gases into the atmosphere turns out to be a strongly self-reinforcing feedback process. Maybe this is what happened on Venus, which now has an atmosphere dense with CO2 and a temperature of about 450O C. Hopefully, however, we will have technological means of counteracting such a trend by the time it would start getting truly dangerous.

    5    Crunches

    While some of the events described in the previous section would be certain to actually wipe out Homo sapiens (e.g. a breakdown of a meta-stable vacuum state) others could potentially be survived (such as an all-out nuclear war). If modern civilization were to collapse, however, it is not completely certain that it would arise again even if the human species survived. We may have used up too many of the easily available resources a primitive society would need to use to work itself up to our level of technology. A primitive human society may or may not be more likely to face extinction than any other animal species. But let’s not try that experiment.

    If the primitive society lives on but fails to ever get back to current technological levels, let alone go beyond it, then we have an example of a crunch. Here are some potential causes of a crunch:

    5.1  Resource depletion or ecological destruction

    The natural resources needed to sustain a high-tech civilization are being used up. If some other cataclysm destroys the technology we have, it may not be possible to climb back up to present levels if natural conditions are less favorable than they were for our ancestors, for example if the most easily exploitable coal, oil, and mineral resources have been depleted. (On the other hand, if plenty of information about our technological feats is preserved, that could make a rebirth of civilization easier.)

    5.2  Misguided world government or another static social equilibrium stops technological progress

    One could imagine a fundamentalist religious or ecological movement one day coming to dominate the world. If by that time there are means of making such a world government stable against insurrections (by advanced surveillance or mind-control technologies), this might permanently put a lid on humanity’s potential to develop to a posthuman level. Aldous Huxley’s Brave New World is a well-known scenario of this type.

    A world government may not be the only form of stable social equilibrium that could permanently thwart progress. Many regions of the world today have great difficulty building institutions that can support high growth. And historically, there are many places where progress stood still or retreated for significant periods of time. Economic and technological progress may not be as inevitable as is appears to us.

    5.3  “Dysgenic” pressures

    It is possible that advanced civilized society is dependent on there being a sufficiently large fraction of intellectually talented individuals. Currently it seems that there is a negative correlation in some places between intellectual achievement and fertility. If such selection were to operate over a long period of time, we might evolve into a less brainy but more fertile species, homo philoprogenitus (“lover of many offspring”).

    However, contrary to what such considerations might lead one to suspect, IQ scores have actually been increasing dramatically over the past century. This is known as the Flynn effect. It’s not yet settled whether this corresponds to real gains in important intellectual functions.

    Moreover, genetic engineering is rapidly approaching the point where it will become possible to give parents the choice of endowing their offspring with genes that correlate with intellectual capacity, physical health, longevity, and other desirable traits.

    In any case, the time-scale for human natural genetic evolution seems much too grand for such developments to have any significant effect before other developments will have made the issue moot.

    5.4  Technological arrest

    The sheer technological difficulties in making the transition to the posthuman world might turn out to be so great that we never get there.

    5.5  Something unforeseen

    As before, a catch-all.

    Overall, the probability of a crunch seems much smaller than that of a bang. We should keep the possibility in mind but not let it play a dominant role in our thinking at this point. If technological and economical development were to slow down substantially for some reason, then we would have to take a closer look at the crunch scenarios.

    6    Shrieks

    Determining which scenarios are shrieks is made more difficult by the inclusion of the notion of desirability in the definition. Unless we know what is “desirable”, we cannot tell which scenarios are shrieks. However, there are some scenarios that would count as shrieks under most reasonable interpretations.

    6.1  Take-over by a transcending upload

    Suppose uploads come before human-level artificial intelligence. An upload is a mind that has been transferred from a biological brain to a computer that emulates the computational processes that took place in the original biological neural network. A successful uploading process would preserve the original mind’s memories, skills, values, and consciousness. Uploading a mind will make it much easier to enhance its intelligence, by running it faster, adding additional computational resources, or streamlining its architecture. One could imagine that enhancing an upload beyond a certain point will result in a positive feedback loop, where the enhanced upload is able to figure out ways of making itself even smarter; and the smarter successor version is in turn even better at designing an improved version of itself, and so on. If this runaway process is sudden, it could result in one upload reaching superhuman levels of intelligence while everybody else remains at a roughly human level. Such enormous intellectual superiority may well give it correspondingly great power. It could rapidly invent new technologies or perfect nanotechnological designs, for example. If the transcending upload is bent on preventing others from getting the opportunity to upload, it might do so.

    The posthuman world may then be a reflection of one particular egoistical upload’s preferences (which in a worst case scenario would be worse than worthless). Such a world may well be a realization of only a tiny part of what would have been possible and desirable. This end is a shriek.

    6.2  Flawed superintelligence

    Again, there is the possibility that a badly programmed superintelligence takes over and implements the faulty goals it has erroneously been given.

    6.3  Repressive totalitarian global regime

    Similarly, one can imagine that an intolerant world government, based perhaps on mistaken religious or ethical convictions, is formed, is stable, and decides to realize only a very small part of all the good things a posthuman world could contain.

    Such a world government could conceivably be formed by a small group of people if they were in control of the first superintelligence and could select its goals. If the superintelligence arises suddenly and becomes powerful enough to take over the world, the posthuman world may reflect only the idiosyncratic values of the owners or designers of this superintelligence. Depending on what those values are, this scenario would count as a shriek.

    6.4  Something unforeseen.

    The catch-all.

    These shriek scenarios appear to have substantial probability and thus should be taken seriously in our strategic planning.

    One could argue that one value that makes up a large portion of what we would consider desirable in a posthuman world is that it contains as many as possible of those persons who are currently alive. After all, many of us want very much not to die (at least not yet) and to have the chance of becoming posthumans. If we accept this, then any scenario in which the transition to the posthuman world is delayed for long enough that almost all current humans are dead before it happens would be a shriek. Failing a breakthrough in life-extension or widespread adoption of cryonics, then even a smooth transition to a fully developed posthuman eighty years from now would constitute a major existential risk, if we define “desirable” with special reference to the people who are currently alive. This “if”, however, is loaded with a profound axiological problem that we shall not try to resolve here.

    7    Whimpers

    If things go well, we may one day run up against fundamental physical limits. Even though the universe appears to be infinite, the portion of the universe that we could potentially colonize is (given our admittedly very limited current understanding of the situation) finite, and we will therefore eventually exhaust all available resources or the resources will spontaneously decay through the gradual decrease of negentropy and the associated decay of matter into radiation. But here we are talking astronomical time-scales. An ending of this sort may indeed be the best we can hope for, so it would be misleading to count it as an existential risk. It does not qualify as a whimper because humanity could on this scenario have realized a good part of its potential.

    Two whimpers (apart form the usual catch-all hypothesis) appear to have significant probability:

    7.1  Our potential or even our core values are eroded by evolutionary development

    This scenario is conceptually more complicated than the other existential risks we have considered (together perhaps with the “We are living in a simulation that gets shut down” bang scenario). It is explored in more detail in a companion paper. An outline of that paper is provided in an Appendix.

    A related scenario is described in, which argues that our “cosmic commons” could be burnt up in a colonization race. Selection would favor those replicators that spend all their resources on sending out further colonization probes.

    Although the time it would take for a whimper of this kind to play itself out may be relatively long, it could still have important policy implications because near-term choices may determine whether we will go down a track that inevitably leads to this outcome. Once the evolutionary process is set in motion or a cosmic colonization race begun, it could prove difficult or impossible to halt it. It may well be that the only feasible way of avoiding a whimper is to prevent these chains of events from ever starting to unwind.

    7.2  Killed by an extraterrestrial civilization

    The probability of running into aliens any time soon appears to be very small.

    If things go well, however, and we develop into an intergalactic civilization, we may one day in the distant future encounter aliens. If they were hostile and if (for some unknown reason) they had significantly better technology than we will have by then, they may begin the process of conquering us. Alternatively, if they trigger a phase transition of the vacuum through their high-energy physics experiments (see the Bangs section) we may one day face the consequences. Because the spatial extent of our civilization at that stage would likely be very large, the conquest or destruction would take relatively long to complete, making this scenario a whimper rather than a bang.

    7.3  Something unforeseen

    The catch-all hypothesis.

    The first of these whimper scenarios should be a weighty concern when formulating long-term strategy. Dealing with the second whimper is something we can safely delegate to future generations (since there’s nothing we can do about it now anyway).

  3. shinichi Post author





Leave a Reply

Your email address will not be published. Required fields are marked *