Yes that’s a bit of cherry-picking, but if you want to backtest the method’s power to forecast unexpected outcomes, you have to pick ones that have come out unexpectedly. If you use it against still-surviving long-lived structures like the Great Pyramids or Hadrian’s Wall, the known duration of the upper bound well is outside what we can actually observe for testing purposes.
Actually yes and no. The ESB was built in 1930, so the math isn’t the same. Its 50% longevity forecast in 1981 would have been 17-153 years, so yes, the model would agree with you that it’s probably not coming down anytime soon.
If you looked at the entire cohort of urban US skyscrapers built in 1973, including WTC, with all the information available about skyscraper longevity but no good information about future demolition or disasters, you’d still expect the cohort to be entirely intact after 21 years. There’d be every reason to believe that, and no reason to doubt it. But the method (the “Copernican method” as Gott calls it) would give 50% odds that the cohort would fall below 100% in that timeframe, which did happen.
Of course that doesn’t tell you exactly which building would be lost, or why, or even precisely when. It’s probabilistic not deterministic, and not terribly specific. And there are lots of reasons to doubt its applicability to forecast the end of human reproduction, not least because the feedback loops introduce complexity issues as you mentioned. But Gott’s method seems to be a decent enough tool to assign a probability on an upper bound of duration for outcomes for well-defined phenomena whose ending we otherwise have no reason to expect, or a good way to estimate that ending even if we did expect it.
Shall we pick a different backtest? Something else that was supposed to last a fairly long time and didn’t, or something that’s new and is expected to be around a while? My imagination is failing me at the moment. I’m amused to think there’s a 50% chance the new Las Vegas Sphere will end in 3-30 months. Currently there’s no reason to expect that, but it does provoke some thought about how it could. Structural failure, technological failure? The appeal of the new tech grows stale faster than expected, and the revenue from visitors ceases to justify the costs of operation and maintenance? I find that latter outcome fairly likely, so I’ll be watching with interest.
The analogy I’ve heard associated with Gott’s argument is this: You’re presented with a box of pingpong balls (you can’t quite see how large the box is - maybe one side of it is embedded in a wall, so you don’t know how far back it goes) and you’re told that the balls are numbered from 1 to the total number of balls, and thoroughly mixed. Poke a hole in the box, and grab one ball at random. Its number is 12. How many balls are likely to be in the box? Or the first balls’ number is 4,515,271,009 - how many balls are likely to be in the box? For the first case, you’d expect a number greater than 12 (obviously), but pretty small - it would be very unlikely that if there were 10 million balls, you happened to pull out #12. Likewise for the 4,515,271,009 case, it would be very unlikely that there were only 4515,271,010 balls and you happened to pick the one with the second highest number. Gott applies that reasoning to the human race - if the human race lasting for another 10,000 years would mean that there could be 100 trillion humans in all of history, it’s weird that you and I (the 100 billionth and 100 billionth (and one)) humans were randomly selected to be alive now - far more likely that we’re middle of the road folks in a human race that lasts only 200 years more and had only 172 billion people in total. All his calculations have a lot of assumptions - are we really picked at random from the total population of hypothetical humanity? Are possible human futures all equally likely? (no to that one, certainly), so I don’t sweat about that issue much. But for certain things, it’s a reasonable calculation to do - my house has existed since 1980, so the odds that it will collapse immediately are pretty small (but the odds tthat it will last 150 years are also pretty small).
This is actually a superior formulation to what I’ve been using in this thread. I knew I’d seen it before but couldn’t manage to come up with it as you did.
It’s unlikely that humanity is all that special or privileged (as Copernicus suggested w/r/t space)
The longer a thing has endured so far, the more you can assume it will do so in the future. Its longevity suggests its ability to withstand potentially destructive events, and/or how likely it’s actually to undergo those events.
So an item’s past actually does contain some useful information about its future durability, more so if it’s longer-lived. I’ll say again that I’m not sold on how well this can predict the end of human reproduction.
But then again, as @Sam_Stone called out above, we seem to be at the beginning of a demographic crash, one that was unexpected a generation ago. Actually the opposite was expected.
Of course all populations have boom/bust cycles, but we live in an unprecented global economic order that was built on the back of growth levels that haven’t turned out to be guaranteed. What happens when a global economy built on expectation of permanent growth actually goes into sustained decline? It’s hard to say because we’ve never really seen that. Sure there have been pandemics and plagues and mass death before, but nothing like that has happened in the timeframe of the fully integrated global economy.
What does that look like? Unstoppable downward economic spiral? Famine, shortages, prices shocks, destructive global conflicts? Some of these seem uncomfortably imminent.
This algorithm is flawed with anchoring bias. Assuming that where we are right now is “the norm” and therefore we are confident that we are in the center of history.
We aren’t. We aren’t the norm, no other time was the norm either. There is no norm. We are just wherever we are.
No, that’s not what the algorithm assumes. The entire idea is based on the fact that humanity does not occupy any privileged position position in time or space, that we are neither special nor the norm.
This is explained thoroughly in the article that I’ve now linked 3 times.
Andy_L has shared [above]( In fact it’s the opposite, as explained in the article that I’ve now linked 3 times. ) a different (and arguably better) explanation of how the whole idea assumes & is dependent on humans existing in a non-special and non-privileged frame.
In some article or another, it was noted that if the observer is in a special time, the method won’t work (a doctor delivering babies can’t estimate their lifespan assuming he’s seen them at a random time in their lives - he’s there at the birth).
The article is saying that there is a formula you can use for determining how long something will last based on how long it has already lasted. This is a great insight, and if I was faced with a wager on an independent event (the time until end of the thing) with no knowledge other than how long it has existed, this formula is the way to go.
Two major problems with it:
The range of forecast values is ridiculously large, bordering on unusable. Another 1/3 to 3X of its lifespan isn’t very useful information for decision-making. A 100 year old thing is likely to live at least 30 to 600 years more? What do you do with that?
We actually have a whole lot of information about the possible lifespans of the things we are talking about, and by using that we can probably do better.
Getting back to the OP, the smooth march of progress and the economy is an illusion. If you zoom in you’ll see that almost all the main drivers of change were random events or tipping points that weren’t predicted. Crashes, wars, major inventions, surprise disasters like volcanoes or hurricanes or tsunamis or terrorist attacks.
Twenty years ago we had just come out of 9/11, a war in Afghanistan and Iraq happened, we went through a crash in 2008, the cellphone revolution happened, the sudden rise of social media changed a lot of things, Trump was elected out of nowhere, a major pandemic and lockdown fundamentally changed many things, multiple major wars in Europe happened, Putin invaded Ukraine, and now we have the unexpected rise in AI. I’m probably forgetting some things.
Every decade is full of black swan events. The spaces in between them are generally periods of relative stasis. The sheer quantity of these events, and our complex, partially stochastic response to them, makes any prediction of the future futile outside of some broad strokes, and even those are suspect.
For example, the prediction “the world will be more populated in 50 years” would have been considered trivially obvious through the entire modern era, but may no longer be true because of a sudden demographic collapse that went completely unpredicted. All predictions based on continued population growth failed. And all predictions were, right up until the model broke.
I quote from the paper:
“This statement would have turned out to be true for anyone who visited the wall in the shaded part of the diagram. Because the shaded region is half the bar, we can say that, for half the days of the Berlin Wall’s existence, this prediction would have been correct.”
The method doesn’t work in real time.
Berlin Wall is created in 1961. In 1962 you ask this formula how long the wall will last. It tells you 1/3 a year to 3 years. Because it’s been in existence for a year. The lower bound is of course useless. But maybe the upper bound is right. Nope, the wall is still there in 1964. Now you run the formula and it tells you 1 year to 9 years. That’s not going to wind up right either.
Basically, if you run the formula every year, you’re going to be wrong a bunch of times. You’re always wrong on the low end, and you’re often wrong on the upper end as well. All you know is when it finally ends, is “looking back, the formula was the best at the halfway point.” Who cares. No one knew what the halfway point was AT THE TIME OF THE HALFWAY POINT. There was nothing intrinsic about 1975 that made it clear that it was the “halfway point” of the Berlin Wall. And halfway points may not be particularly meaningful, either. People have an “average lifespan” of 80 or so, but that doesn’t make 40 the halfway point. There is the growth of youth, and the final years have a more accelerated decline. But if you run the formula on a human at 40, you’ll find that the 40 year old human will live anything from 14 (well the 40 year old is already past that) to 120. This is not a very good formula for a human being, is it?
The formula only looks good, to the extent that it does, when you are looking backward completely, and ignoring that people may try to use the formula at any point, not “somehow know” where the actual midpoint is.
Bolding above mine. Keep that part in mind while we review your error below:
This isn’t a correct explanation of the formula. The formula doesn’t tell you there’s a 100% chance the wall is gone in 1/3rd to 3 years. It’s telling you the odds of that are 50%.
More supporting detail follows, as well as how to get to 95% confidence (if you’re willing to accept the wider range it requires).
And I would add, I wouldn’t say the lower bound is entirely useless. It’s just that young objects don’t have a lot of information that speaks to their durability, other than comparing them to known durabilities of similar objects, which has more predictive power when the object doesn’t have a long history to consider.
But that’s not to say the lower bound is totally useless either. It’s not inconceivable that the Berlin Wall could have been torn down in 1962-1965. It wasn’t the Pyramid of Giza, it was just a 12-foot slab wall set up for political reasons that changed quickly and could have changed again, perhaps as a diplomacy tactic or shift in power or just a relocation. A 50% estimate on 3 years wasn’t a ridiculous guess at the time; we only see it that way because we know the right answer was 28 years.
If you predict 50% tails on a fair coin flip, but it comes up heads, that doesn’t mean the forecast was wrong. You weren’t predicting the next outcome, just stating its probability, knowing that forecasted events don’t always happen. Same principle applies here. The only problem is you can prove out the coin flip by doing repeat trials, but we’re not going to re-run 20th-century history repeatedly to see how many times the Berlin Wall got knocked down in 1965 instead of 1990 or 1994. That’s an acknowledged limitation of the method.
I think there’s a difference between depopulation due to lower birth rates vs depopulation vs war or plague (like the Black Death that killed half of Europe). And the rate we use up resources with current population levels is unsustainable anyway.
The whole point of The Black Swan was that people are eager to use math to give a false sense of certainty, especially in economics and some social sciences, often by assuming that events follow a normal distribution or something similar, when in fact many things do not do so. (Extremistan in the original book.)
Few civilizations last as long as the Egyptian or Roman empires. You could take all the civilizations and figure out how long they lasted or have been going. And get some result at such a certainty. But are you assuming civilizations now are similar enough to civilizations before? How useful is it to take the longest civilization and say there is a good X% change ours will not last as long?
Climate change looks at the brief time where temperatures have been recorded and the sudden changes during the Anthropocene. We can make educated guesses, and assess risk to some degree. But we still don’t know what we don’t know.
You’re giving the formula too little credit. Actually it’s right 67% of the time. If something lasts 100 years, it’s right starting in year 33. If something lasts 1,000 years, it’s right in year 333.
But you only know it’s right until you get to year 100 or 1,000, so it’s useless. For the Berlin Wall, in year 27 it’s telling you the wall is going to last 9 to 81 years. Accurate, but useless and ridiculous.
Yeah, saying a wall may or may not be there in 8 to 1000 years isn’t useful.
I want something like in the movies and TV where I can look at a giant 3d holographic ball of incomprehensible nodes that light up bright red and get all agitated, indicating a 99.9999% chance of a terrorist attack right down to the date, time, and description of the “outlier” who plans to implement it.
Again, that’s not how any of it works. If you observe a thing that’s 33 years old, then the formula states there are 50/50 odds that it will last an additional 11-99 years, and the other 50% of the prediction is that it will end either sooner than 11 years or later than 99 years.
I hear the argument “that’s too vague to be useful.” It’s true that for a familiar entity like a wall or a building, there are much better things to base a forecast on than past durability. We know things about the material strength, weather conditions, seismic activity, vehicular activity. But that approach has its own limitations. The Berlin Wall didn’t spontaneously fall down due to an earthquake, or have an accidental collision with a bulldozer. Only with hindsight is it obvious that Berliners would take it apart with hammers in 1990. Some people thought it would be there a century or more, and there was little reason to doubt it. Gott’s method said: think again.
Gott’s method is the best available when we know absolutely nothing else about the problem. I would suggest it’s also useful when there are too many known variables to confidently forecast anything. This method isn’t hamstrung by complexity as suggested elsewhere in the thread; in fact it’s a reasonable fallback method when the thing being forecast is entirely too complex (and environmentally affected) to yield any better confidence.
A thing’s past durability is a reasonable basis to estimate its future durability (if nothing else is known about it). Something that’s lasted a long time probably has a quality that will make it endure longer. Something that’s just recently started existing has no such track record. Look around you in the world, do you find that very old things outnumber very new things? I would suggest not.
If you flipped a known fair coin 10 times, and the probability suggested it ought to come up tails 50% of the time, but you flipped heads 4 times in a row, would you say that forecast is useless or ridiculous?
This is just how probability works. Humans crave deterministic predictions, but probability cannot give you that, nor is it intended to.
For those who are interested in the specific topic of population decline, here’s the ongoing thread where we talk about it in dribs and drabs for a couple of years now.
This seems dubious. In particular, it seems to ignore confounding factors - the existence of birth control and the control it allows for selection of reproduction.
For example, before the twentieth century, families tended to lots of births, with a fair number of infant deaths, a well as the occassional death of the mother in childbirth.
The advent and availability of the pill (and later methods) changed that. In particular, society changed to where having a dozen children is not an asset but a burden.
Women want to have careers, too. This shifts some women to delay starting a family until they graduate from college and establish their career.
Both of those factors reduce the birth rate, but neither drive the birth rate to zero. But they would be huge factors in the birth rate trend to make it look like births will end.
It’s like assuming a linear projection with the assumption nothing will change, even though the curve may be parabolic with an asymptote above zero.
It ignores all factors. Or alternatively, it considers so many factors that they’re impossible to weigh distinctly. 760 years is a lot of time for anything to happen. Think how much has changed between 1264 and 2024. Or since 2020, for that matter. We see some drastic changes of fortune living on this wet rock.
It doesn’t necessarily have to be apocalyptic. Maybe the tech singularity arrives, maybe our existence changes in a way that no longer qualifies as human, maybe there’s no more need for childbearing. Or maybe it turns out to be the end of humans reproducing on earth, and the beginning of reproducing somewhere else. Or maybe a giant asteroid just straightforwardly kills us off.
Gott’s prediction is indifferent, though understandably people are reflexively denying the possibility of an apocalypse in 760 years, partly due to aversion psychology, and partly because the Vox article unfortunately (and incorrectly) framed it that way.
If someone instead suggested “there’s a 50% chance that in 760 years, technological advances will bring about a complete change in human existence and reproduction as we understand it”, I bet a lot of board members would respond “760 years? that’s too pessimistic, we may live to see it in our own lifetimes.” So the right framing can make a huge difference in what kind of predictions you’re willing to entertain.
Although I consider classic Marxism to be pretty well discredited already as a predictive theory, a question I’ve wondered about is “to what extent did a conscious awareness of the tenets of Marxism undermine Marx’s original predictions?” Especially in light of the fact that many people consider today’s version of capitalism to be a pre-emptive counterrevolution.