Stories
Slash Boxes
Comments

SoylentNews is people

posted by martyb on Monday July 08 2019, @11:58AM   Printer-friendly
from the I've-still-got-plenty-of-time dept.

A math equation that predicts the end of humanity:

The most mind-boggling controversy in the contemporary philosophy of science is the "doomsday argument," a claim that a mathematical formula can predict how long the human race will survive. It gives us even odds that our species will meet its end within the next 760 years.

The doomsday argument doesn't tell what's going to kill us — it just gives the date (very, very approximately).

Yet, I [William Poundstone] now believe the doomsday prediction merits serious attention — I've written my latest book about it. Start with J. Richard Gott III. He's a Princeton astrophysicist, one of several scholars who independently formulated the doomsday argument in the last decades of the 20th century. (Others are physicists Holger Bech Nielsen and Brandon Carter and philosopher John Leslie.) In 1969, Gott was a physics undergraduate fresh out of Harvard, spending the summer in Europe. At a visit to the Berlin Wall, he did a quick calculation and announced to a friend: The Berlin Wall will stand at least 2 and 2/3 more years but no more than 24 more years.

Demolition on the wall began 21 years later. This motivated Gott to write his method up. He published it in the journal Nature in 1993. There, Gott wrote of the future of humanity itself. He forecast a 95 percent chance that the human race would cease to exist within 12 to 18,000 years.

Not all Nature readers were convinced. "'There are lies, damn lies and statistics' is one of those colourful phrases that bedevil poor workaday statisticians," biostatistician Steven N. Goodman complained in a letter to Nature. "In my view, the statistical methodology of Gott ... breathes unfortunate new life into the saying."

Yet Gott and his predictions also received favorable attention in the[sic] New York Times[*] and the[sic] New Yorker[*] (where a profile of Gott was titled "How to Predict Everything"). Gott is an engaging storyteller with a Kentucky accent that's survived decades in the Ivy League. He has become a sort of scientific soothsayer, successfully predicting the runs of Broadway plays and when the Chicago White Sox would again win the World Series (they did in 2005).

Can it really be that easy to predict "everything"? It quickly became clear that 1) most scholars believe the doomsday argument is wrong, and 2) there is no consensus on why it's wrong. To this day, Gott's method, and a related one developed by Carter and Leslie, inspire a lively stream of journal articles.

You can read more about the doomsday debate on Quora

[*] The name of these publications do include the word "the" and should, therefore, be capitalized: The New York Times and The New Yorker, respectively.


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 0) by Anonymous Coward on Monday July 08 2019, @12:35PM (17 children)

    by Anonymous Coward on Monday July 08 2019, @12:35PM (#864440)

    There is a 99,999999% chance he is wrong, but if he is right... well there is nobody left to tell it to.

    Seriously, if a field of science that is able to make predictions of this kind it's biology or human sciences. All astrophysisics can tell if there is a 100% chance hit with something from space that will sterilize the planet from all live. All humans have to do is to make sure the species gets a chance on other planets.

  • (Score: 2) by JoeMerchant on Monday July 08 2019, @12:46PM

    by JoeMerchant (3937) on Monday July 08 2019, @12:46PM (#864445)

    There is a 99,999999% chance he is wrong

    Not really - this is like weather forecasting, as TFS says: "lies, damn lies, and statistics." Even when they get it wrong, they point to: "well, that must have been one of the very improbable cases..."

    --
    🌻🌻 [google.com]
  • (Score: 4, Insightful) by ikanreed on Monday July 08 2019, @01:21PM (15 children)

    by ikanreed (3164) Subscriber Badge on Monday July 08 2019, @01:21PM (#864460) Journal

    Your 99.999999% chance he's wrong is wrong. The accuracy of prediction could be as high as 50%.

    The nature of the math isn't imprecise or inaccurate, if you randomly start observe a phenomenon with a duration, the most likely case is that you're currently observing the middle of it. There are lots of problems with the doomsday argument though.

    1. There's lots of unstated premises.
    1a. All of "human civilization" has the same character with respect to durability. It's not like we invented weapons that could kill us all in the last century.
    1b. That the chances of making this randomly making this observation is equal accross all of "human civlization" It's not like we invented fucking observational statistics recently.
    2. By its nature, the prediction must be updated each year we don't cease to exist or it becomes inaccurate. If the prediction is "within 12,000 years" till the end today, then a year from now, you've changed your rate of decay by invalidating some possibilities. And if my math is right, you'd update to "within 12,000.5 years"
    3. The assumption that the nature of the random events that start civilization are comparable to the forces that end it. The general basis of the "Assume you're in the middle of a randomly observed event with duration" is that both the start and stop are dictated by forces that, for lack of a better word, are stochastic. Not necessarily a reasonable assumption.

    • (Score: 4, Informative) by AthanasiusKircher on Monday July 08 2019, @05:53PM (12 children)

      by AthanasiusKircher (5291) on Monday July 08 2019, @05:53PM (#864596) Journal

      The accuracy of prediction could be as high as 50%.

      That's a funny statement. "Could be as high..." -- so, to quantify that "could be," is that a probability of a probability?

      And if we're talking about that, I think the author's argument is even more simplistic and statistically naive. I'm sure he probably tries to address this in his book, but let's just imagine a simple scenario -- I sit at a table and start overturning cards. I will keep turning them over until I run out of cards. You have to guess when this process will come to an end.

      Now, if you know anything about cards, you might guess that I have a deck of 52 standard playing cards. Which, in most of the scenarios that someone starts doing it, is probably true. But we don't know any of that in this author's scenario. We only take into account the length of the past duration for some thing or era or process.

      His logic is that to make an "accurate prediction" 50% of the time, you reason as follows: 50% of the time when you make an observation, you'd randomly end up in the region where between 25% and 75% of the total length of the process has elapsed. So, to estimate the total length of the process, if you were at the 25% low end, you'd have to imagine the process goes on for 3 times as long as it already has. If you happen to be at 75% of the process, then the process will only go on for 1/3 of the length of time it already has.

      So, his logic says that you have a 50% chance of making an "accurate prediction" if you say the range is between 1/3 of the current length and 3 times the current length.

      But is this really an "accurate prediction"? What do we mean by "accurate"? Let's examine the cases where you can guess the expected length after EVERY ONE of the 52 card draws in this process. If by "accurate prediction," you simply mean that you guess a range that includes the number 52, yes, you will do that 50% of the time.

      But that it NOT "accuracy" in the normal way we think of it. After all, I can beat that accuracy easily -- I predict the length of future card draws will be somewhere between 0 and infinity. I'm right 100% of the time. Hah! I beat his method!

      Well, of course that's idiotic. So, how could be better judge his accuracy? We'd need to say that his prediction range doesn't include "silly" values -- values that are way too high or way too low.

      So, let's ask the question again. He makes a prediction after every single card draw. Let's say we allow him a 50% tolerance -- his prediction range's low values can't fall less than 50% of the actual value of 52 (i.e., not less than 26), and his prediction range's upper bound can't fall greater than 150% of the actual value of 52 (i.e., not greater than 78). How often does that happen? The answer is NEVER. If we assume each computation is rounded to the nearest whole card, after 19 cards are drawn, he'd predict an expected total duration range of 25-76 cards. That's the closest he gets. After 20 cards, he'd predict a range of 27-80.

      Thus, this method NEVER achieves a standard of being able to predict a range of values between 50% and 150% of the actual value. Although he says he's right "50% of the time," by a reasonable standard, he's basically wrong 100% of the time.

      Okay, so let's cut him a little more slack. Say we give him a 75% tolerance. Thus, his prediction is accurate IF his range includes the actual number (52) AND never predicts values less than 25% of the total (13) or more than 175% of the total (91). How often does that occur? Well, out of 52 predictions, if I did the math right, he'd get the entire predicted range within 75% tolerance only 10 times, or about 19% of the time.

      So, no, I don't think his method can claim to have an "accuracy of prediction" anywhere near 50%. For any reasonable assumptions about what an "accurate prediction" constitutes, he's only going to achieve it a lot less than 50% of the time, even though he's claiming the method I just outlined should be "accurate" 50% of the time.

      And this shouldn't be surprising that this prediction is so bad, because he's trying to make a prediction using one data point (i.e., the duration of the process/event/thing, with no other knowledge). One cannot extrapolate from one data point: that's pretty much a foundational law of statistics. Now, if he had other information, this changes things. For example, if you knew decks of cards contain 52 cards most of the time, you could make a reasonable guess of the length of this process (unless someone has a double deck or a Pinochle deck or something).

      If you observed this process a few times, you might also notice the 52-card duration. But we have no other data points about the lengths of civilizations (human or otherwise), so we can't draw on that sort of information. Even within the single process, maybe you could draw on other things to make an informed prediction. For example, with the cards, you might notice that there are 4 suits after a while. And you might notice there are only 13 different kinds of cards in each suit. After enough cards have been drawn, you might extrapolate those patterns to make increasingly better predictions. If humanity is governed by similar rules, perhaps we could use them too to make a prediction -- but I don't think anyone knows them.

      In the end, we're left with this bogus method of guessing based on pretty unreasonable assumptions (duration of a thing/event/process is uniform and we are likely to step in and make a guess at any random time from the beginning). And even if those assumptions were true, I just showed the method is pretty meaningless, since it generates really stupid answers the vast majority of the time. As I said, I could achieve 100% accuracy according to this method of claiming "accuracy" simply by predicting that a process will end somewhere between now and infinite time in the future.

      TFA says his method as presented was controversial: "Not all Nature readers were convinced." Really? I find it difficult to believe ANYONE who knows anything about math and statistics takes this seriously, at least the way he has argued it.

      • (Score: 3, Interesting) by ikanreed on Monday July 08 2019, @06:05PM (11 children)

        by ikanreed (3164) Subscriber Badge on Monday July 08 2019, @06:05PM (#864609) Journal

        That's a lot of words to arrive at some very suspect conclusion.

        Yes, it's a hypernaive statistical method. But if you were trying to estimate the duration of a random event, strictly based on one piece of evidence(how long since the same single event started), this is the correct method to do so. And 50% is the chance that it's on the left side of that curve.

        I don't think it's fair to say that's the only piece of information we have, but the methodology is correct.

        • (Score: 2) by AthanasiusKircher on Monday July 08 2019, @06:35PM (4 children)

          by AthanasiusKircher (5291) on Monday July 08 2019, @06:35PM (#864626) Journal

          Yes, it's a hypernaive statistical method.

          That's an understatement.

          But if you were trying to estimate the duration of a random event, strictly based on one piece of evidence(how long since the same single event started), this is the correct method to do so.

          Extrapolation based on a single data point is stupid and generally meaningless with no other information given. Calling any method based on that "correct" is naive.

          And 50% is the chance that it's on the left side of that curve.

          As I said, I have a simple methodology that is 100% accurate, according to that evaluation. Please re-read my post and address that point if you want to argue this seriously.

          I don't think it's fair to say that's the only piece of information we have

          It's NOT the only piece of information we have about human civilization (and arguably we DON'T have even that data point, since it depends on your definition of "beginning of human civilization"), but it seems to be the way the author's statistical methodology claims to work.

          but the methodology is correct.

          Assuming a whole truckload of assumptions (about evenness of temporal distribution, the fact that we are supposedly making a prediction at a "random" point among that temporal distribution, etc.) that we have no justification in assuming.

          Again, I have a method that works 100% of the time. The author's "method" is only meaningful if it actually excludes "silly" answers. It does not, a large percentage of the time. So, it's not a whole lot better than my 100% accurate method.

          • (Score: 2) by ikanreed on Monday July 08 2019, @07:34PM (1 child)

            by ikanreed (3164) Subscriber Badge on Monday July 08 2019, @07:34PM (#864656) Journal

            Yeah, I guess I appreciate all that.

            The problem is trying to outdo the doomsday argument with is a bit like trying fill out the drake equation. You can stick some assumptions in there, but really, you don't know what it takes to start or stop another civilization. Not really.

            Other discussions I'm having today make me long for a very short remainder of humanity.

            • (Score: 0) by Anonymous Coward on Tuesday July 09 2019, @03:06AM

              by Anonymous Coward on Tuesday July 09 2019, @03:06AM (#864835)

              dP/dT = ( P )*( 1 - P )

          • (Score: 2) by FatPhil on Tuesday July 09 2019, @09:15AM (1 child)

            by FatPhil (863) <pc-soylentNO@SPAMasdf.fi> on Tuesday July 09 2019, @09:15AM (#864910) Homepage
            His method is the equivalent of getting to a bus stop, asking a geezer on the bench when the last one was (and getting a numeric answer), not looking at the timetable, and guessing when the next one will be with an attitide towards accuracy similar to someone saying "the next roll of the roulette wheel won't be a 0 or a 00".

            I did once run an online quiz based on the identical principle, when the maths wasn't widely known. You arrive in a new town, and you see buses numbered X. On the assumption that all routes are contiguously numbered, how many routes do you think the town has? The optimal strategy seemed to be to just guess that you've seen one(s) in the middle. Number of seen bus route = time since start of humanity; highest bus number = length of humanity's existence. You can make it more interesting in the bus version by seeing two buses, or even three. (I found out later that this has been applied to chassis numbers on enemy tanks during wars, google/wikipedia is your friend.)
            --
            Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
            • (Score: 2) by AthanasiusKircher on Wednesday July 10 2019, @01:52AM

              by AthanasiusKircher (5291) on Wednesday July 10 2019, @01:52AM (#865257) Journal

              Thanks. Yes, I was familiar with this sort of estimation method before I read the article here. I've just never seen it applied to something so important or taken so seriously. (My first post explained it in more detail both to familiarize people here who might not have heard or it or read TFA, as well as to set up why the method isn't that useful when accuracy matters.)

        • (Score: 2) by AthanasiusKircher on Monday July 08 2019, @07:02PM

          by AthanasiusKircher (5291) on Monday July 08 2019, @07:02PM (#864637) Journal

          Also, just to be clear -- I was not at ALL trying to pick a fight with you in my initial post on this thread. If anything, I was trying to add a more subtle criticism to the list of qualifications for this theory you already started. I only posted it below yours as it seemed one of the most appropriate places among the posts already here.

        • (Score: 2) by aristarchus on Monday July 08 2019, @08:21PM (4 children)

          by aristarchus (2645) on Monday July 08 2019, @08:21PM (#864679) Journal

          Come now! The prediction is based on the fact that humanity exists. The probability of the existence of humanity is therefore 1. The probability of humanity ceasing to exist, if humanity did not exist, =0. Therefore, the probability of humanity ending, once it in fact exists, is >0. This is such a profound conclusion that it behooves the theorist to choose some random value greater than zero, to make it appear to be non-vacuous. The real difficulty is that zero is greater than zero, when we are dealing with probability, since impossible means zero possibility, but not impossible does not mean actual possibility.
          Or, in Modal logic notation
          ◊A = ∼◻∼A
          but
          ∼◻∼A ≠ ◊∼A

          • (Score: 3, Insightful) by ikanreed on Monday July 08 2019, @08:33PM (3 children)

            by ikanreed (3164) Subscriber Badge on Monday July 08 2019, @08:33PM (#864687) Journal

            Uh... We're not proving civilization doesn't exist or is possible to exist. We're discussing entirely hypothetical ways of estimating black swan events. Or broadly the durability of civilization. Like in a very informal sense those relate to existence in a pretty intuitive way. But in a strict syllogistic sense... they're very disconnected.

            Given that, that symbolic logic doesn't help much. It sheds light on the wrong question?

            • (Score: 2) by PartTimeZombie on Monday July 08 2019, @09:47PM (2 children)

              by PartTimeZombie (4827) on Monday July 08 2019, @09:47PM (#864717)

              Are we actually agreed that civilisation does exist though?

              • (Score: 4, Funny) by ikanreed on Monday July 08 2019, @11:28PM (1 child)

                by ikanreed (3164) Subscriber Badge on Monday July 08 2019, @11:28PM (#864754) Journal

                They made 6 games in the series. Unquestionable.

                • (Score: 2) by PartTimeZombie on Tuesday July 09 2019, @12:21AM

                  by PartTimeZombie (4827) on Tuesday July 09 2019, @12:21AM (#864780)

                  Your slanderous tongue will make it all the more satisfying when I stand before the ruins of your capital!

    • (Score: 3, Interesting) by HiThere on Monday July 08 2019, @06:20PM (1 child)

      by HiThere (866) Subscriber Badge on Monday July 08 2019, @06:20PM (#864617) Journal

      As you say, not a good or useful argument. Still, I think is predictions are a bit optimistic. I would say that humanity, as we know it, will end within 50 years, certainly by the end of the century. I decline to say whether this will be a good or a bad thing. Also the mode of ending, as some are catastrophic and others are....less so, and arguably desirable.

      The modes of ending that I've identified are:
      1) war
      2) ecological collapse
      3) extreme introversion (see "The Machine Stops", and update the technology.)
      4) supernormal releasing mechanisms. (Early versions of this include lipstick. Modern versions include porn and FleshLight. Advanced versions are being worked on all over the world.)
      5) goodlife. We become pets of an AI. Perhaps one that thinks of itself as benign towards humans, but it won't be the humanity we have known.

      I'm sure I've missed some. I think that unfortunately the first two are the most likely.

      --
      Javascript is what you use to allow unknown third parties to run software you have no idea about on your computer.
      • (Score: 1, Insightful) by Anonymous Coward on Tuesday July 09 2019, @09:48AM

        by Anonymous Coward on Tuesday July 09 2019, @09:48AM (#864922)

        Neither war nor ecological collapse will end humanity. They may end civilisation and greatly reduce population, but people are tough and adaptable. There will be survivors.
        Life may go back to being "nasty, brutish, and short" but to quote Heinlein "Man is an unspecialized animal. His body is primitive. He can't dig; he can't run very fast; he can't fly. But he can eat anything and he can stay alive where a goat would starve, a lizard would fry, a bird freeze. Instead of special adaptations he has general adaptability"