Future of humanity

Future of humanity names the attempt to reason systematically about which long-run trajectories are open to the human species and its descendants, and how likely each is. It is not forecasting in the ordinary sense: the events at issue are unprecedented, the relevant timescales exceed any dataset, and the most consequential variables are the least predictable. What the field offers instead is a discipline of constraint — physical limits that no technology can exceed, historical base rates that any specific claim must beat, and a taxonomy that keeps arguments from talking past each other.

§Four trajectories

Nick Bostrom's 2009 essay set out the taxonomy the field still uses.¹ Every long-run scenario, he argued, falls into one of four families, distinguished by where the trajectory ends up rather than by how it gets there.

Extinction. Humanity ceases to exist, permanently. The category includes both sudden events and slow declines, and its defining feature is irreversibility: unlike every other outcome, it forecloses all subsequent ones. Derek Parfit's argument that the moral difference between near-extinction and extinction is far larger than the difference in immediate casualties turns on exactly this asymmetry.²

Recurrent collapse. Civilisation repeatedly reaches roughly its present level of technology and repeatedly falls back, without either dying out or advancing further. This scenario is unusual in requiring a mechanism that both prevents permanent extinction and prevents permanent advance, and Bostrom notes that no plausible mechanism has been described. Resource depletion works against it: a collapsed civilisation would rebuild without the accessible fossil energy that fuelled the first ascent.

Plateau. Development continues to some ceiling and stops there, with technology and human capacities remaining broadly at levels a person today would recognise. The plateau might be imposed by physics, by economics, or by deliberate global choice. It is the implicit assumption of most economic and demographic projection, and it is the scenario that receives the least explicit defence.

Posthumanity. At least one central human capacity is raised far beyond its current maximum, producing beings that fall outside the current concept of human — the condition defined in Posthuman. This family includes radically extended healthspan, greatly enhanced cognition, and non-biological substrates.

The taxonomy is exhaustive by construction and says nothing about probability. Its value is in forcing an argument to specify which transition it is claiming, and what would have to be true for that transition to occur.

§What is reasonably well established

A short list of things about the next several decades is on firm ground, and it is worth separating from everything else in this article.

Global population growth is decelerating. Fertility has fallen below replacement in more than half of all countries, and standard projections show a peak around the 2080s followed by decline — a reversal of the assumption that shaped twentieth-century futurism, and one with direct bearing on the objection examined in Overpopulation and life extension. Ageing populations, not growing ones, define the demographic problem of the next fifty years, which is the premise of the argument in The longevity dividend.

Health gains have shifted from mortality to morbidity. Global life expectancy at birth is around seventy-three years, most of the twentieth century's gains came from reducing early-life mortality, and further gains must come from late-life disease — where progress has been slower and where the compression of morbidity that some public-health models predict has not clearly materialised.

Physical limits are calculable and binding. Energy budgets, waste-heat dissipation, the speed of light, and the accelerating expansion of the universe together bound what any civilisation can do; Anders Sandberg and colleagues have argued that intergalactic settlement is not obviously ruled out by physics, while the reachable volume shrinks over cosmological time.³ These constraints are among the few genuinely reliable inputs to long-horizon reasoning.

Climate change is a well-characterised risk that is not, on mainstream assessment, an extinction risk. It is a large-scale harm to human welfare and ecosystems and a plausible contributor to instability, which is a different claim from civilisational termination and should not be merged with it.

§The technological levers

This wiki covers the specific capabilities through which any transition out of the plateau would have to run. Their maturity varies enormously, and the variation matters more than the list.

Aging. The Geroscience hypothesis holds that the processes catalogued in Hallmarks of aging drive most chronic disease jointly, so that intervening upstream would compress many diseases at once. Multiple interventions extend lifespan in mice. None has demonstrated an effect on human aging in a controlled trial, and the interventions with the best human evidence remain exercise and the avoidance of known harms. Longevity escape velocity is the limiting case of this lever and is rejected as a near-term prospect by most biogerontologists.

Genomes. CRISPR–Cas9 and its successors made precise editing routine in the laboratory and have produced approved somatic therapies. Heritable modification is a separate matter, restricted almost everywhere and covered in Human germline editing; selection rather than editing is what is commercially available, and polygenic embryo screening can shift the odds on common traits only slightly, because thousands of variants of small effect underlie the traits people most want to influence.

Reproduction. In vitro gametogenesis would, if it worked in humans, decouple gamete supply from the ovary and change the arithmetic of embryo selection completely. It works in mice. Artificial womb devices support extremely premature lambs and have not been tried in humans.

Nervous systems. Brain–computer interfaces restore communication and control to people with paralysis at rates well below unimpaired function. They do not currently enhance healthy users, and the arguments for merging human and machine cognition rest on capabilities that do not exist.

Minds. Whole brain emulation and mind uploading depend on two premises — that the relevant causal structure is capturable by achievable imaging, and that mind is substrate-independent — which are respectable positions rather than established results. Most neuroscientists regard both as very distant.

Bodies and matter. Organ bioprinting and xenotransplantation address a defined shortage, with real clinical progress and unsolved problems in vascularisation and immunology. Nanomedicine has produced working DNA-based devices in the laboratory and nothing resembling the cell-scale machines proposed in the 1980s.

Space. Decades of spaceflight medicine document what microgravity and radiation do to human physiology over months; the engineered adaptations imagined under the heading of pantropy, and the multigenerational biology a generation ship would require, are unstudied outside fiction and thought experiment.

Machine intelligence. Artificial general intelligence is the lever most likely to change the others, because it acts on research productivity rather than on any single problem. It is also the one whose timeline is least constrained by evidence.

The asymmetry between levers

Every lever above except machine intelligence has a bounded effect on a specific system. AI is the only candidate that could plausibly compound the rate at which the others advance, which is why AGI timelines dominate every serious long-horizon estimate — and why the weakness of the evidence about them propagates into everything else.

§The risks

The case for treating this as urgent rather than academic rests on the claim that anthropogenic risks now exceed natural ones by a wide margin. Toby Ord's survey estimates roughly a one-in-six chance of existential catastrophe this century, dominated by unaligned artificial intelligence and engineered pandemics, with natural risks such as asteroid impact and supervolcanism orders of magnitude smaller.⁴ The estimates are explicitly subjective and have been criticised as unfalsifiable; the ranking is more robust than the numbers, since natural risk rates can be estimated from the geological record and have evidently been low enough to permit two hundred thousand years of human existence.

Engineered pathogens are the clearest biological case, and the mechanisms are covered under dual-use research of concern and Mirror life. Martin Rees's earlier assessment reached a similar conclusion about the century by a different route, emphasising the falling cost and rising accessibility of destructive capability.

The countervailing observation is that the field has a systematic bias toward the dramatic. Slow trajectories — declining fertility, entrenched inequality, institutional decay, value lock-in through durable technology — are less discussed than sudden ones, and there is no reason to think they matter less.

§Reasoning across long horizons

Four methodological commitments distinguish careful work in this area from speculation.

Separate mechanism from timeline. The argument that a capability is physically possible is independent of any claim about when it will exist. Most disputes conflate them, and the timeline half has by far the worse track record, as Accelerating change documents.

Use base rates, and admit when none exists. Reference-class forecasting works when a reference class exists. For extinction, it does not, and the historical absence of extinction is weak evidence because of observation selection: only surviving civilisations produce records.

Beware conjunctive detail. Detailed scenarios feel more plausible and are strictly less probable than their components, a well-documented reasoning failure. A five-step story about how a technology transforms society is less likely than any of its steps.

Distinguish demonstrated from extrapolated. The recurring failure in this literature is treating mouse results, laboratory demonstrations and design studies as though they were positions on a timeline rather than facts about what has been done. Maturity scales such as the technology readiness levels used throughout this wiki are a blunt instrument for this, and better than nothing.

What the record shows

Confident long-horizon technology predictions have been wrong in both directions and at roughly comparable rates. Nuclear energy, space settlement and machine translation were all over-forecast in the mid-twentieth century; the internet, genomic sequencing costs and protein structure prediction were all under-forecast. The lesson is not that optimists are wrong but that specific dates carry almost no information.

§Where the disagreements actually are

Participants agree on far more than the public argument suggests. Nearly everyone accepts that the technologies covered here are physically possible in some form, that timelines are uncertain, and that irreversible outcomes deserve extra weight. The live disputes are narrower.

How much do we owe the far future? Longtermist arguments hold that the number of potential future people is so large that reducing existential risk dominates other priorities. Critics object that this reasoning is dominated by tiny probabilities of enormous values, and that it licenses present sacrifices for speculative benefit.

Is technological progress the right lever? Transhumanists and accelerationists treat capability growth as the primary route to good outcomes. Bioconservatives and the Precautionary principle tradition hold that some capabilities should not be developed. A third position proposes to alter the ordering of arrivals rather than the total, which most participants nominally endorse and few institutions implement.

Who benefits? The distributional question examined in Access and inequality cuts across all four trajectories. A posthuman transition available to a small fraction of people is a different outcome from a universal one, and the taxonomy above does not distinguish them.

§Open problems

Three problems are unresolved and structurally hard rather than merely unfinished.

The plateau's mechanism is undertheorised. If neither collapse nor transformation occurs, something must be holding development at a ceiling for a very long time, and no one has described what. The scenario receives the least attention precisely because it is the least interesting to argue about, which is a poor reason.

The evidence base for probability estimates does not exist. Numbers like one in six are elicited judgements presented in the notation of measurement, and the field has not developed a method for improving them that is testable within a human lifetime. Forecasting tournaments have shown short-horizon skill; nothing shows that skill extends to century-scale questions about unprecedented events.

The observer selection problem is unsolved and it undermines the most common empirical argument. That humanity has survived every past hazard tells us little, because we could not be here to observe otherwise — so the historical record cannot distinguish a safe world from a lucky one, and the sharpest disagreement in the field is over how much that consideration should move any estimate.

§See also

• Existential risk
• Posthuman
• Nick Bostrom
• Differential technological development
• Technological singularity
• Artificial general intelligence
• Longevity escape velocity
• Transhumanism

§References