How the numbers work
Brandon Carter set out the argument in 1983 and Richard Gott gave it its best-known form in Nature in 1993. Roughly 100 billion humans have been born so far. If you are a randomly selected human among all humans who will ever live, there is a 95 per cent chance you fall in the middle 95 per cent of the sequence.
That puts the total number of humans ever at somewhere between about 1.4 billion and 2 trillion, with the doomsday estimate near 1.2 trillion. At current birth rates the species would then have a few thousand years left, and probably much less.
Gott's version is a pure survival argument: take how long a thing has lasted so far, and the 95 per cent confidence interval for its total lifetime runs from a fortieth of that to forty times it. Applied to the human species at 200,000 years, that gives between 5,000 and 8 million years.
The objections that matter
The self-sampling assumption, formalised by Nick Bostrom, is the load-bearing premise: that you should reason as if you were a random observer drawn from all observers. Deny it and the argument dissolves. Defend it and variants of it start producing contradictions, including the 'presumptuous philosopher' who concludes he is Boltzmann brain number one.
The reference class problem is equally serious. Are you a random human, a random primate, a random mammal, a random conscious being, a random civilisation-member? Each choice changes the answer, and there is no principled way to select one that does not beg the question.
There is also a selection effect: any observer who is reasoning about doomsday is, by construction, alive and in a technologically capable civilisation. That is weak evidence about timing, and some authors argue it cancels exactly the effect the argument relies on.
It is not a prediction. It is a statement about where a randomly placed observer should expect to be, and it produces no date, no mechanism and no policy.
It is not refuted by the observation that previous doomsday predictions failed: those were specific forecasts, and this is a claim about reference classes.
It is not established either: the whole inference rests on a probability assignment to observers that nobody has a way to test.
The archive labels this fringe because it is a genuinely contested piece of philosophy with no empirical content to check it against — and because its conclusions depend more on bookkeeping than on the world.
Why it is in the archive
Because it shows how far a plausible symmetry assumption can be pushed, and how quickly the push produces conclusions nobody is willing to accept.
Related: the simulation argument, superrationality.
Sources for this page
- Wikipedia, “Doomsday argument” — the argument, variants and objectionsen.wikipedia.org
- Gott (1993), Nature — “Implications of the Copernican principle for our future prospects”doi.org
- Bostrom (2002), Anthropic Bias — the self-sampling assumptionwww.anthropic-principle.com
- Wikipedia, “Self-sampling assumption” — the premise the argument needsen.wikipedia.org