“Since it is equally likely that those of us living today are in the first or second half of all past and future human births...”
Something about this predicate seems suspect. I can understand the reasoning, but it seems like we’re mixing two separate domains of probability to arrive at the conclusion, much like the common wrong answer to the Monty Hall problem.
One key problem with the premise is the assumption that your (lack of) priors indicate a flat distribution. It sounds reasonable at face-value, but in the bayesian framework it is actually a rather strong assumption that can be leveraged to arrive at far less reasonable conclusions.
In lieu of German Tank serial numbers, Gott's original paper uses projections from Ehrlich and Ehrlich's 1990 Population Explosion paper that the world's population will top out at around 10 billion, then essentially cause a dumpster fire and crash and burn. Obviously extremely speculative stuff that's been sensationalized here.
I can't get past the paywall to read the original Population Explosion paper that Gott uses, but the dumpster fire premise seems to comes from the assumption of tragedy-of-the-commons-style environmental degradation. The Wikipedia on Ehrlich offers the following comment:
> When is an area overpopulated? When its population can't be maintained without rapidly depleting nonrenewable resources (or converting renewable resources into nonrenewable ones) and without degrading the capacity of the environment to support the population. In short, if the long-term carrying capacity of an area is clearly being degraded by its current human occupants, that area is overpopulated.
I agree with you that it is questionable set of assumptions, though I think when considering some of issues humanity is facing in the realm of climate change and water shortages, it's not an entirely untenable hypothesis.
Actually, this is much more interesting than the doomsday math. Like does somebody have the mental model of a bunch of software engineers shoveling equations into a furnace and it powering a data center?
What exactly did they have in mind when they decided on that phrasing?
It's true though. The first large, publicly seen use of Bayes' theorem in computing was in spam filtering, but it is being used for much more things nowadays, many of which impact our lives significantly. Okay, so the phrasing was hyperbolic; it isn't used for every single algorithm obviously.
So let's say we did this way back in time, when only 100 humans had been born. They could have been in the 1st half or the 2nd half. In either case, this would seem to predict that no more than 100 more would be born, which is clearly absurd.
I think you're missing that this columnist may in fact be a sensationalist hack. As far as I can tell, nowhere in Gott's paper does he imply that we will end in less than a thousand years. From a quick scan of the paper, Gott's most incendiary statement is that there is a 95% confidence interval that our remaining species lifetime is in the range of 12 to 7.8e6 years.
I think it's that you're unlikely to be alive at a time when only 100 humans had been born so you probably wouldn't be doing the calculation. Or, put another way, if you did do it then, you'd be in the minority of people who got the wrong answer. It's taking a sample of 1 (ourself) from the entire history of all humans. If that one person is randomly chosen, they're most likely to be alive in the fat middle of the population history, not at the thin tails (assuming there is a thin tail at the end?). I'm not sure how picking ourself is random though, maybe because we didn't decide when to be born?
There's an assumption here that you're sampling an ergodic process.[1] 'All tanks the Germans have built so far' satisfies this assumption (assuming equal likelihood of capture). 'All tanks the Germans will ever build' doesn't. In other words, you can't sample the 200Bth human because they haven't been born yet, and this isn't evidence they will never be born.
It can be limit process though, like a sigmoid, that eventually becomes stationary. That is a big assumption though. Evolution was punctuated by such seeming stable domains only to reset everything once conditions changed... Or push past previously known barriers. This including technical progress early in human history. The population function over long time is not exactly smooth either.
Im pretty sure I watched a numberphile video on this. I'm having trouble googling it. Anyone know the video I'm talking about?
This is the sort of thing that belongs on HN, but not WSJ. WSJ should be talking about the privacy issues they vaguely touch on before going back to talking about doomsday by statistical argument.
21 comments
[ 2.9 ms ] story [ 59.9 ms ] threadSomething about this predicate seems suspect. I can understand the reasoning, but it seems like we’re mixing two separate domains of probability to arrive at the conclusion, much like the common wrong answer to the Monty Hall problem.
edit: it only works for German tanks because you can reasonably assume that the Germans aren't warehousing significantly more tanks than you've seen.
Seems like a silly guess. Plus bad climate and famines and wars can stop it much earlier than those 10 billion.
> When is an area overpopulated? When its population can't be maintained without rapidly depleting nonrenewable resources (or converting renewable resources into nonrenewable ones) and without degrading the capacity of the environment to support the population. In short, if the long-term carrying capacity of an area is clearly being degraded by its current human occupants, that area is overpopulated.
I agree with you that it is questionable set of assumptions, though I think when considering some of issues humanity is facing in the realm of climate change and water shortages, it's not an entirely untenable hypothesis.
Uh, what? I closed the tab at this point.
It implies that there is a class of algorithms which are "the ones used today" and that they run on a similar set of equations.
This was written by a moron and I can't be bothered.
What exactly did they have in mind when they decided on that phrasing?
What am I missing?
If anyone wants to check for themselves, hopefully it's not against HN rules to link the paper: https://sci-hub.tw/10.1038/363315a0
[1] https://en.wikipedia.org/wiki/Stationary_ergodic_process
This is the sort of thing that belongs on HN, but not WSJ. WSJ should be talking about the privacy issues they vaguely touch on before going back to talking about doomsday by statistical argument.