No.10279
What do you guys think about the Doomsday Argument? Almost everyone that hears this argument immediately objects to it. But their objections are always different, and they never seem convincing to me. It is weird, but it seems logically sound.
One of the most convincing aspects is that you can apply the same logic to other things just fine. If you pick a random object and estimate it's lifespan. Say you pick a tree that's 10 years old, and estimate it will only survive about 10 more years. The 90% confidence intervals will include the true answer 90% of the time. The same math was used in world war 2 to estimate the production german tanks from their serial numbers, and they were right with 1 tank! J. Richard Gott used it to estimate the lifespan of the Berlin wall and it was remarkably close to the true answer.
But there is one tiny flaw in it. It assumes you are randomly chosen from the group of all humans that will ever exist. In fact most humans that have ever lived never would have considered this argument, and so it shouldn't apply to them. It seems to me that it should only apply to the people who have also thought about this argument.
…Which was invented in 1983. Assuming the number of people exposed to this argument remains constant, it implies there is a 50% chance the world will end in 34 years. Which lines up remarkably well with singularity predictions. So does this prove we are fucked?
____________________________
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10282
I don't understand this shit.
>if you are a randomly chosen human, there is a 95% chance you exist in the middle 95% humanity ?
middle 95 percent of what ? Of all the humans that ever lived till the end of the world ?
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10283
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10284
>>10279
Why does it matter whether or not they've heard of the argument? Why would that make it not apply to them? Those are rhetorical questions, I don't think there's an answer.
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10288
>>10279
Applying the argument to just "people exposed to this argument" makes it about the Doomsday Argument itself; it's now a 90% chance you're in the middle 90% of "people exposed to the Doomsday Argument" - sure, you can assume that holds constant, but there's no reason to do so and as long as you don't it's a prediction entirely about the popularity of this one meme.
The thing is, it's a priori. What it does is give you a prior. It's a starting point, like making a Fermi estimate before trying to actually measure a thing - finding actual evidence doesn't quite screen it off, but you can get from any prior to any posterior with the right evidence.
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10289
My objection is to taking "last human" to mean no posthumans will be around. Admittedly, this objection doesn't work very well if your estimate is only 34 years.
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10296
Where did you get the first 90% ?
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10304
If the right reference class is humans + posthumans, and posthumans will never go extinct, then the argument doesn't go through, because there is no "middle 95%".
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10305
>>10304
The argument can be applied to any reference class, but" and posthumans will never go extinct" is obv. begging the question.
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10306
>>10282
I assumed familiarity with the idea and didn't really try to explain it. See the wikipedia for a full explanation: https://en.wikipedia.org/wiki/Doomsday_argument
>>10284
If you only apply the argument to people who have heard of the argument, it is guaranteed to be correct. Exactly 5% of people who think about the meta-doomsday argument will be in the first 5% of people to do so.
You can't apply the same guarantees to the regular argument. After all the doomsday argument only can come about late in humanity's history. Because for most of time humans didn't have an understanding of probability or math necessary to make the argument, or ability to spread the idea if it did occur. So most people to make the argument today will be more towards the tail end of humanity than they estimate.
It also clears up the issue with reference class very nicely. Should non behaviorally modern humans be included in the argument? What about prehumans? Could you even include chimpanzees and our chimpanzee like ancestors? Just making about beings that actually do make the doomsday argument clears up reference class issues nicely IMO.
>>10288
>>10289
>>10304
The meta doomsday argument just implies people will stop thinking about the doomsday argument, not necessarily go extinct. This could imply a permanent set back to civilization that doesn't wipe us out. Or it could imply a singularity where the future is secured and people/beings stop trying to predict doomsday. Or the world goes on as is, but the doomsday argument falls out of favor.
But I don't find those possibilities convincing. Civilization permanently set back and no one ever reinvents books or probability theory? Post humans, with all their intelligence (and huge numbers!), never think about anthropic arguments, even just for fun? Ever? The world goes on as is but no one ever finds or rediscovers this simple idea and shares it like I am doing now?
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10310
The Doomsday argument is a frequentist argument. It's just frequentist confidence intervals you are measuring. Robin Hanson and Andrew Gelman have both criticized the argument on that basis. To do it like a proper bayesian we must specify a prior.
A good default prior for something like this is the pareto distribution. It's the distribution that creates power laws and appears everywhere. It allows both large and small numbers to be relatively probable. It has one parameter, and because we use base 10, let's set it to 1/10th. Arbitrary but it must be about that small for large values to be relatively plausible.
So let's say the true number of humans that will ever live is N.
[N ~ pareto(1/10)]
Now we need a likelihood function to update on. We have just one observation. That we drew a number between 1 and N, and it came up as 180 Billion (the number of people that have ever lived.) This means we can assign 0 to all possible values of N less than that. And for every other value, larger values are proportionally implausible relative to smaller values. Being 180 Billion out of 180 Billion possibilities is twice as likely as being 180 Billion out of 360 Billion possibilities.
Multiplying these together and getting a probability function took some algebra but I solved it. It gives the median number of years of civilization left as 1,689 at the current rate of population growth (95% < 23,000.)
Doing the same math on the meta doomsday argument seems like it would be a bit harder because I don't know the number of people exposed. But assuming it's the same every year since 1983, the math works out so it's invariant to the exact number. Pareto distributions are pretty awesome like that. And it gives a median estimate of 41 years left (95% < 580).
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10311
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10312
>>10311
Extra support for the argument comes from the fact that AI looks dangerous for humans. Humanity is likely to be replaced/annihilated by "humans 2.0"
Intelligent civilization will go on.
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10313
>>10310
This is the kind of gratuitous addition of explanation and insight that I would like to see more of here.
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10316
>>10310
Since your prior kind of has to come from somewhere frequentist (unless "your ass" isn't frequentist), why use Pareto as opposed to just taking the "Doomsday Argument" for your prior?
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10318
>>10316
It's typical in bayesian methods to use some kind of uninformative prior that allows a very large range of values roughly equal probability.
A true flat prior that says any value is equally likely is too much though. It's not mathematically correct. You can't draw a random real number from the space of all real numbers because most of them are ungodly huge. I also don't think it makes any sense to a priori assume the number of humans that will ever exist is much larger than the number of atoms in the universe.
I think it's fine to assume that increasingly huge orders of magnitude should have increasingly small probabilities, proportional to the order of magnitude.
Pareto distributions also have a lot of nice properties like being scale invariant. My analysis doesn't change at all if you multiply the number of humans by some constant. The number of expected humans is also multiplied by that constant.
I admit the 1/10 parameter was arbitrary. But testing it with different parameters doesn't change the result that much really. I just use that whenever I need some uninformative prior over some parameter that could be really big.
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10340
So you are just plucking that prior out of your ass and the Doomsday argument just reflects that prior back at you. i.e. you're assuming the rough outline of your conclusion.
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.10341
>>10340
ALL statistical methods pluck priors out of someone's ass. Bayesians are just explicit about it. You can't update without a prior.
And no it doesn't reflect the prior back at you. The prior is super vague and assigns a ton of probability mass to obscenely high numbers. The likelihood function is just so strong that most of that mass disappears.
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
