never mind
Nate Silver presented a book length treatment of this subject: The Signal and the Noise. He surveys forecasting in a range of contexts including elections, weather, sports, and geology. Recommended.
One theme: there’s a difference between forecasts and predictions. The former are possible. The latter mostly isn’t. What’s the difference? Forecasts are probabilistic. I could say that there’s roughly a 10% chance of recession between July 2024 and July 2025. That’s a forecast. A prediction would be, “There won’t be a recession this year.”
Admittedly you can map one concept on to another. “I am over 99.999% sure there will be at least one full moon within the next 45 days”, is pretty similar to a prediction. But in practice near 100% predictions are uninteresting.
At any rate, I think humanity can make steady progress on the forecast problem, quantum butterflies notwithstanding. Collapsing 90% probabilities into 99.999999% probabilities seems harder to me: I don’t see reliable and nontrivial prediction coming any time soon.
Interesting and relevant article: A math equation that predicts the end of humanity
I wouldn’t be honest if I said I understood this well enough to even summarize it well. But I’ll try anyhow:
- the basic idea is that if you can estimate the past duration of a thing, then you can estimate its future duration.
- this theory says there’s a 50% chance of human births ending in the next 760 years, based on best estimates of current birth rate vs. total number of births in human history.
As I said, it’s way above my pay grade, but the reasoning seems pretty tight to me. I’m curious to hear what our resident math nerds think about this one.
It depends if we’re talking about absolute or probabilistic answers.
The world is likely a chaotic system, being as it contains numerous chaotic systems. Therefore, the only way that even a super-duper genius brain could predict world events more than a trivial amount in the future, would be to have perfect knowledge of events right down to the particle level, which is impossible.
OTOH there’s no reason, hypothetically, that a fantastically good model of human societies, psychology, economies and technology couldn’t deliver accurate probabilities of the chances of events like WWIII in a given year.
Yeah, that was the inspiration of my post. Gott’s 1993 paper sparked a lot of responses and rejoinders.
As I see it, the underlying issue is that we have a sample size of one (only one humanity) though we have a decent sample of years. But! The years aren’t independent observations - they are correlated with one another. What happens in one year affects what happens in the next. So the effective sample size is smaller than it might first appear. Maybe decades are less correlated with one another - but there are fewer of those.
So, to wing it, we’ve survived almost 8 decades since Hiroshima. That’s a little reassuring, but not a lot reassuring IMHO: 8 is a very small sample. All that said, I disavow expertise on this subject.
I think he’s pulling numbers out of his ass. Why would one assume the Berlin Wall would stand for at most 3x as long as it has been erected any more than you would assume the Great Wall of China would stand for another 20 years after it was built? Why would one assume that the total number of humans will cap a 200 billion vs 200 trillion?
It just seems random and arbitrary to me.
Haha, I came here after reading your post title specifically to ask if your real name was Hari Seldon.
I hadn’t heard anyone was producing the Foundation novels as a TV serial, thanks for mentioning.
Yeah, it’s on Apple TV. They are working on season 3 now. It’s not bad. They take a lot of liberties with the books, but I think that’s ok. Hari does shit on his own philosophy a lot, picking various “chosen ones” and whatever to lead this and that. Lee Pace (Ronan the Accuser from the MCU) is fun as the Emperor of the Galaxy (technically “Empire” is used in the show to both refer to the empire as a political entity and as a honorific title addressing its ruler(s), implying the emperor is the living embodiment of the empire I suppose. It makes sense in the show but it’s very different from the books.)
That’s not the assumption. Just like we can say that a fair coin flip has a 50% chance of landing tails, for any durable phenomenon, there’s a 50% chance that you are seeing it at a time when its future duration is between 1/3rd and 3x of its past duration. That’s not the controversial part; it’s straightforward probability math. The article has a graphic to help explain it, since probability is often not intuitive.
So if we apply that to the Ming section of the wall, first completed 654 years ago, then there’s a 50% probability that section will cease to exist in the next 218-1962 years. The 95% confidence estimate would be between 16 years and 25,506 years, if I followed the math correctly.
That’s not the assumption. The assumption is that if 100 billion people have ever been born, then at current birth rates of 130mm/yr, that number will double in 760 years. Although there are some quibbles to be had about the 100 billion count – what exactly counts as a person, and are we confident we estimated the cumulative count correctly? – the calculation is straightforward arithmetic.
What I have trouble following is how Gott connects the above arithmetic to the probability theory described in the first para. If you accept the demographic estimates on past birth count and current birth rate, then obviously we double it in 760 years. But how does Gott arrive at saying there’s a 50% chance that humans stop being born at the doubling point? That’s the part that seems hand-wavy to me.
It’s clear how the theory applies to how likely a durable object will survive X number of years. I don’t see how it applies cleanly to how likely humanity will continue having X number of births, but again I’m not even close to being an expert in this area.
And in (at least) two of those decades there were periods that got very close to total war, which is even less reassuring
The idea of a predictive model for the future of mankind runs smack into the Madman Theory, which dates back to Machiavelli.
While Machiavelli (and Henry Kissinger) intended to use the theory only as a negotiating tool, there’s nothing to stop a genuine madman in power from blowing us all up. For that matter there’s nothing to prevent a madman from violently eliminating all their rivals, and having it turn out that one of the rivals was even more of a madman; stopping that madman would actually save humanity.
ETA: As @Andy_L pointed out while I was typing this. We’ve come close to nuclear war a couple of times even with rational actors having their fingers on the button. It would have taken only one misheard order or one battlefield commander panicking to set things on a different path.
Yeah, using the model I described above, there’s a 50% chance that the era of nuclear rivalry between the US and USSR/Russia ends in the next 25-225 years.
Does the end of nuclear rivalry mean peace and plowshares? The equation doesn’t tell us anything about that. It could mean peaceful de-escalation. I’d love to think so, at least.
This sounds to me like handwaving ‘armchair philosophy’. Why 50%?
Rather like the arguments that purport to show that it is very probable that we are living in a simulation.
It would be interesting to do some backtesting of the former concept: assuming we could find predictions to work with?
50% is the confidence you get when the multiples are 1/3 and 3, and was chosen as a simple way to demonstrate the calculation. You could also do 95% probability, and the range would be 1/39th duration to 39x duration. That makes the range a lot bigger, and of course a more vague prediction will be less useful, but it serves to establish an upper bound.
The upper bound for the 95% figure on humanity’s birthrate falling to zero was 18,000 years. Not that it will happen in exactly 18,000 years, but there’s a 95% chance we stop reproducing sometime between now and then (if you accept the figures on how much we’ve already reproduced and how fast we’re currently reproducing)
If read the linked article (here it is again) you can see how this backtested against the lifespan of the Berlin Wall. He predicted it would last between 32 months and 24 years (from 1969). It was demolished 21 years later.
I’m not buying it. But it should be fairly easy to test. There are numerous historical events and constructions that we should be able to use to test the methodology. They seem to have put a lot of weight behind the ‘accurate’ prediction of the Berlin wall coming down, but there have been thousands of other constructions we could test.
First, that’s a huge range of prediction. Now ask, at what point in history would this have been correct for the Great Pyramids? How about Hadrian’s wall? Or the length of various empires?
If our mathematicion was around 100 years after the great pyramid was finished, would he really believe that a good answer for its lifespan would be 30 years to 600 years?
It seems to me that they are trying to apply statistical probabilities to a complex adaptive system (or many of them). For example, none of the statistics up to 1970 hinted at a global demographic collapse, but it was just a generation away.
Complex Adaptive system do not follow bell curves, and any trends visible in them are subject to change at any time. They have tipping points, and apparent stable equilibria can collapse without warning. And since they adapt to changes in unpredictable ways, you can’t predict them.
As the article says, the math is really for cases where you have no information, so you make assumptions and the math tells you the range of possible answers ( a very wide range). But that’s not the case with civilization. Our future distributions are not just random, but more of a directed random walk with feedback about the validity of choices. It’s just very different.
No, I get it (I think). But saying the Berlin Wall has a 50/50 chance of lasting between another 2.67 and 24 years or a 95% chance of lasting up to 312 years isn’t particularly useful.
I opened the thread hoping that he was!
Correction, that’s 50% odds for 30 years to 600 years, and 50% outside that range. (Unfortunately the article seems at several different points to conflate the upper bound of the probable range with a specific end-of-life prediction, probably due to a combination of journalistic hype and layman confusion).
A mathematician might observe that 95% confidence would be 2.5-3900 years and 99% would be 1-9900 years. There’s really nothing to believe or not believe, especially with figures that are larger than an historical epoch. That’s just what they are. An engineer who wanted to be immortalized for correctness would probably throw out the lower bound and make practical recommendations based on the upper bound of 99% confidence.
As you say, the useful range is maybe too large to be practical, and the article unfortunately implies a greater certainty than turns out to be the case. The backtests are interesting though. For them to be really informative, you have to pick things that are already gone, preferably unexpectedly.
If you looked at the World Trade Center in 1980 when it was 7 years old, you’d laugh at anyone who gave 50/50 odds that it would be gone in just 21 years or less. You’d bet large against that, and I’d have joined you to collect an easy bet, because there was no conceivable reason to imagine it wouldn’t last for its designed lifespan of 100 years or whatever. But they’d have taken our money, because it was indeed gone 21 years later.
But that’s cherry picking. To be more valid, you’d look,at the entire cohort of skyscrapers. But even that’s not valid, because we have a lot of information about how long such structures tend to survive.
The same math that worked for the WTC when applied to the Empire State Building at the same time in its life would have been wrong, at least in the 50/50 case, and there is no reason to believe it’s coming down any time soon. Predicting the lifespan of the Colosseum or the Acropolis would have been wildly wrong. But other structures designed to last as long vanished a long time ago.