Towards a Better Theory of Anthropics: Introducing the PPA
[This is the first piece in a series on the Parametric Parameter Assumption (PPA), a new model of anthropics. This particular piece is self-contained and for most people will have all they need to know.
TLDR:
With the right prior, the PPA avoids nearly every major anthropic paradox, including the Doomsday Argument, the Presumptuous Philosopher, and the inability to handle infinite universes.
The PPA divides the universe into two types of uncertainty: epistemic unknowns and randomness (a.k.a aleatoric unknowns). The first is treated SSAishly and the second is treated SIAishly.
This piece also introduces the “Basic Model” of anthropics, a single universe model that, paired with the SSA or SIA, exposes many different anthropic paradoxes simultaneously. None of these paradoxes persist in the PPA.]
This is the first piece in a series introducing the Parametric Process Assumption (PPA), a new model of anthropics which allows us to resolve very nearly all anthropic paradoxes I am aware of. With the right prior, the PPA rejects both the Doomsday Argument and the Presumptuous Philosopher and can operate seamlessly over both finite and infinite universes.
The motivating idea behind the PPA is that there are really two different ways to interpret a statement like “the probability of X”:
The Frequentist approach, where probabilities represent the convergence of random trials. Flip many coins and about half will come up heads.
The Bayesian approach, where probabilities are beliefs. We don’t know if the googolth digit of pi is even or odd, so we might as well say there’s a 50% chance it’s even.
The first interpretation makes more sense for some types of questions, while the second approach makes more sense for others. But it turns out that if you take the standard framing for anthropic problems—where a universe is first produced and then you are assigned an observer identity within it—you will get different answers depending on whether you model the universe in the first way or the second. In fact, a fully Frequentist, infinitely repeating universe yields the SIA, while a fully Bayesian one yields the SSA.
The PPA, however, says that the solution is to compose both interpretations. We imagine the entire universe as a random process progressing according to the Frequentist definition, but the laws of physics—the things which actually control the dynamics of the process, like the weak nuclear force or the strength of gravity—are treated in the Bayesian sense. The Bayesian portion we call the “parameters,” and we refer to the rest as the “process.”
In practice, since anthropics is mostly focused on the number of observers, we actually get to absorb almost all the important aspects of the laws of physics into a single parameter which we’ll call the “propensity for life”—for any given laws of physics, propensity for life is the expected number of times we expect intelligent life to emerge. Anything else, such as how long life lasts once it’s here, we treat as random, since that is governed less by the highest-level laws of physics and more by the local dynamics of whatever patch of spacetime that life happens to exist in.
The end result a model that is neither “halfer” nor “thirder.”[1] Rather, the PPA effectively composes the SIAish process portion inside the SSAish parameter portion, inheriting the positives of both approaches without making any of the sacrifices of other competing models, including the various “double halfer” methods existing in the literature. [2]
That said, the PPA is not without weaknesses. The biggest of these is that, by virtue of defining certain variables as parameters, the PPA gives us an extra degree of freedom. Having this extra decision point opens the PPA up to accusations of overfitting, and this makes it extra important to properly motivate the priors we use. I do not know if all decisions this series makes in that regard will be unimpeachable, but if nothing else the PPA’s complete lack of paradoxes should serve as an existence proof of a better way forward for anthropics.
This introductory piece will be self-contained, and should give you a good idea of how the PPA works, albeit with some details left out. The rest of this series will flesh out and build on these ideas, while also serving as an exploration of anthropics as a whole. Much of what I took for granted when I first started reading about this topic turned out to be wrong or exaggerated, and I suspect I am not the only one who made these mistakes[3].
Motivating the PPA
The best way to motivate the PPA is, naturally, with an appeal to the Sleeping Beauty problem (SBP). Recall that in the SBP, Sleeping Beauty is put to sleep by scientists on a Sunday, after which a coin is flipped. On heads, she is woken on Monday, and on tails she is woken on Monday and Tuesday, with an amnesia drug administered in between so that she has no knowledge of which day it is. The SBP asks “upon waking, what should be the Beauty’s credence that the coin went heads?”
One answer to this question—the “thirder” answer associated with the SIA—is that we ought to give 1⁄3 credence to the coin going heads. There are a few arguments for this stance, but the most relevant one for our purposes is the “argument from repetition.” If we were to imagine the Sleeping Beauty experiment being repeated some arbitrarily large number of times, then we would find that 1⁄3 of Sleeping Beauty’s wake-ups were on heads. Since the beauty is “assigned” one of these wake-ups at random, upon waking she should give 1⁄3 credence to it being a “heads” day.
We’ll discuss the validity of this logic in the context of wake-ups some other time. For now, taking it at face value, this argument seems to be a straightforward proof of the SIA[4].
The impetus behind the PPA (at least this particular version of it) is that, as a matter of fact, generalizing the argument from repetition does not actually yield the exact SIA. It gets results consistent with the SIA for the types of problems that anthropics usually likes to consider, but if you apply it universally that won’t always be the case.
To see why, consider a modification of the Sleeping Beauty problem where instead of flipping a coin, the scientists compute the googolth (10^100) digit of pi. If the googolth digit is even, then Sleeping Beauty is woken up once. If it is odd, she is woken twice. Sleeping Beauty, we can assume, neither knows the googolth digit of pi nor knows how to compute it.
On its face, this is an extremely similar problem to the vanilla SBP. Because Sleeping Beauty has no particular reason to assume the digit of pi is more likely to be even than odd, she could reasonably assign 50% credence to either possibility before the experiment starts. However, if we were to actually apply the argument from repetition to this experiment, we would find that there is no possible way to derive a credence of 1⁄3.
The reason is that the googolth digit of pi does not change between repetitions. It’s not even random in the conventional sense of the word, since you cannot simply “re-roll” pi to get a different sequence of digits. The digits of pi are fixed, unalterable truths about our world. “50% chance of even,” is a statement about belief, not the dynamics of the world.
With that in mind, there are two things that can happen:
If the digit is even: on every repetition, the beauty is woken only on Monday. There is a 100% chance every wake-up is on Monday.
If the digit is odd: on every repetition, the beauty is woken exactly once on Monday and once on Tuesday. There is a 100% chance that half of all wake-ups are on Monday, and the other half are on Tuesday.
We are not being placed randomly in a single sequence, but rather being placed with 50% probability into one of two sequences. This means that the possibility of waking on Tuesday is nested within the possibility of an odd digit—one can only ever compute a 25% chance. Thus, in this scenario, we are forced by the argument from repetition to accept the SSA.
The PPA is what you get if you take repetition to its logical conclusion. It operates by essentially decomposing the universe into two layers. The lower level, the “process” (also known as the “likelihood” in standard statistical parlance), is just a progression of random events happening one after the other, like the repeated flips of Sleeping Beauty’s coin, and we use the SIA to place ourselves within it. The outer layer, the “parameters,” represents facts about the universe which are too fixed to be rightly called random, but which we nevertheless do not know; we handle these according to the SSA.
The Basic Model
One unfortunate side effect of using the PPA is that it becomes much harder to hold discussion through the usual thought experiments like the Sleeping Beauty Problem. It suddenly becomes necessary to specify which parts of the experiments represent parameters and which parts represent processes, and even once you have made these specifications (by, say, replacing a coin flip with a digit of pi), the resulting scenario rarely feels analogous to whatever question you were originally trying to ask.
In this section, we’re going to try to build a better model. Our focus will be on two of the biggest paradoxes in anthropics:
The Doomsday Argument[5], which says that if you’re using the SSA and you discover that you are, say, roughly the one hundred-billionth person ever born, your credence in your species being near extinction should increase.
The Presumptuous Philosopher argument, which says that if you are using the SIA, you should bias towards existing in the universe with the most possible observers.
Throughout this series, we’ll spill a lot of virtual ink on the many subtleties of these two paradoxes, but for now the most important thing to realize is that these two statements are not, in fact, opposites. The Doomsday Argument concerns itself with the amount of time your specific species has left until extinction, and triggers only once you’ve learned information about when you were born relative to the “first” human. The Presumptuous Philosopher effect concerns itself with the amount of life in the entirety of existence, and outside of certain toy problems would not typically disappear just because you learned something about your particular species. Indeed, it is possible for a single assumption of anthropics to produce both effects at once.[6]
The terminology we’re going to use is that the Doomsday Argument pertains to the longevity of your species, whereas the Presumptuous Philosopher pertains to the universe’s propensity for life, or more often just “propensity.” In general, the SSA does a better job for “big universe-level questions” like propensity, but fails at smaller scales where longevity becomes relevant, and the SIA does the reverse.
Our goal will be to build a single universe model that incorporates both of these concepts, which we’re going to call the Basic Model of anthropics. The idea is that we will imagine the universe as a gigantic random process that, through randomness, sometimes produces intelligent species. Like the digits of pi, the approximate size of the universe and its probability of randomly producing life is fixed but unknown, while most other facts—including how long an instance of a species lasts once produced—are genuinely random.
The Basic Model.
Each species emerges somewhere in the universe, exists and grows for some time, and then goes extinct. Each color represents the different physiology/conditions under which the species exist, controlled by the diversity variable.
It’s easiest to illustrate this model through a simplified version, which we’ll call the “grid universe.” Here, we imagine our universe as a big grid of some unknown size. As in the Basic Model, the size of the grid is not random but, again like the digits of pi, somehow fixed but unknown.
What is truly random, however, is the pattern that will fill this grid. Starting at the top, going down row by row, a random process fills the grid with colored pixels:
Pixels repent the randomness in the universe. The grid is filled from top to bottom, representing the progression of our universe from one state to the next over time. The particular pattern of pixels on each row depends only on the row above it.
The purpose of these pixels is to represent the randomness in our universe, from galaxies to planets to the individual reactions on those surfaces. It’s just a bunch of stuff happening, and though it is random is not uniformly so—the colors of the pixels (much like the randomness in our universe) depend on each other in a myriad of complex ways. The one constraint is that—just as time in our universe progresses in only one direction—the colors on each row cannot depend on the colors of any rows below them. The grid is filled top to bottom and the future only ever depends on the past.[7]
At some point, through the progression of this randomness, blue pixels might emerge. The exact likelihood of a blue pixel emerging doesn’t particularly matter, as long as it is relatively low, and we can assume it is known or unknown for our current purposes.
Let’s also say that these blue pixels, unlike all the other pixels, are “self-aware,” in the sense that they get to look at some of the pixels above and to the left and to the right, anywhere but into the future, and ask themselves hard philosophical questions like “why am I here?” and “should I be surprised at what I see?”
Blue pixels also propagate—if a pixel is blue, there is some fixed chance the pixel directly below it will also be blue. The blue pixels only know that whatever the chance of another blue pixel, it is determined primarily by randomness within the grid world (in other words, propagation is more like a coin toss than a digit of pi). For the sake of argument we can suppose that the blue pixels have looked around and decided that there is an approximately 10% chance that a blue pixel won’t be followed by another. The relevant part is that whatever prior the pixel has, it is attributing the final outcome to random variation, not some inherent property of the universe.
We can also assume the grid is large enough—and the pixel’s visual range limited enough—that there is no real chance of the pixel seeing an edge, nor of the chain terminating because it hits the bottom. Other than that, though, just as we can only really observe our local slice of spacetime, the pixels don’t know anything about the size of the grid.
Blue pixels emerge in chains. They represent life emerging randomly.
Now, it’s unclear exactly what mechanism “assigned” consciousness to the blue pixel, or if the word “assign” even makes sense in this context. All the average blue pixel knows is that it exists, it has seen some blue pixels behind it, and there are likely going to be more blue pixels after it. What it wants to know are the two questions we already discussed:
Longevity of the current chain: How many more blue pixels will be produced before the current chain terminates?
Propensity for blue pixel chains: How many other blue pixel chains are out there?
From our perspective as a human outside the grid, certain answers seem obvious. The expected longevity for all chains, for instance, is 10, because the independent 10% chance of terminating means that continuing the chain is equivalent to rolling a 10-sided die over and over. Also, since dice rolls are “memoryless” (they don’t depend on previous rolls—see the Gambler’s Fallacy) if we know a particular chain has already progressed for N pixels, we will expect the whole chain to be N+9 pixels long.
Similarly, from our outside perspective, the propensity to produce blue pixel chains would depend exclusively on the size of the grid and the probability of a blue pixel spontaneously emerging. But if all that happens is that we are shown a randomly selected blue pixel chain and told it was at least size N, we would not have enough evidence to make an informed decision about the size of the grid; the best we might be able to say was that the grid wasn’t so small that blue pixels were unlikely to form. After that, we would just make a reasonable guess, call it a “prior,” and use that to produce our credences.
The interesting thing about these simple conclusions though is that no existing anthropic approach can reproduce them, at least not without unacceptable side-effects. The SSA, especially when your prior gives non-zero weight to small grids, will under-estimate the length of the chain (a.k.a. the Doomsday Argument), while the SIA will strongly overestimate the size of the grid relative to the prior (a.k.a. the Presumptuous Philosopher). Double Halfer methods also fare poorly.[8]
As you might have guessed, the PPA alone reproduces these intuitions, predicting 10 more pixels in the chain while also closely matching our prior on grid size.
The PPA does this by modeling the size of the grid the same way it modeled the googolth digit of pi in our previous example, constructing a prior by assigning different probabilities to different possible sizes. Then, for each possible size, we imagine that the random patterns of the grid are re-sampled an infinite number of times, selecting “ourself” (that is, the blue pixel that we are) at random from the set of all blue pixels in the ever-repeating grid.
In other words, we nest our predictions about the randomness of the pixels inside Bayesian predictions about the size of the grid (with an additional adjustment factor we’ll discuss that compensates for the possibility the grid is empty).
The Math
My goal for most of this series is to keep the use of math isolated to only a few places, but it’s worth it to take a moment to spell out what the PPA’s distribution actually looks like.
All of our anthropic assumptions start with a “universe prior”
The SIA in this case would be the distribution we get when the probability of being observer
This means that if each universe
The only way we can reduce the largeness bias is to reduce the multiplier to something less than
giving
In the PPA, we resolve these paradoxes by breaking our universes into two components, a process component
Here,
Moving some terms around, we get this:
That is, we divide the SIA’s prediction by the expected number of observers at a given parameter setting. In the Basic Model, this means finding the expected number of observers given that the laws of physics have a propensity for creating so-and-so many civilizations, so-and-so many large scale simulations, so-and-so many Boltzmann Brains, etc.
What this does, in essence, is divide out the propensity while leaving everything else intact. You don’t know how many times the observer-generating processes start up, but you act SIAish (and thus in a temporally consistent way) about the process you are currently in.
This is also why (after some careful manipulation involving limits) the PPA is able to effectively handle infinity universes. Only your local neighborhood of spacetime is actually relevant once you divide out propensity.
What the PPA Predicts
The following is a list of some of the biggest contradictions and anthropic paradoxes I’m aware of. As we will discuss, under the right universe model the PPA resolves every single paradox on this list.
Temporal Inconsistency/Telekinesis (Under SSA): Under the SSA, we can take a future event of known probability—like a fair coin flip—and cause our credence in it to change just by pre-committing to certain decisions.
Doomsday (SSA): Since the SSA treats our birth rank as a uniform draw from all humans, we are biased towards smaller universes.
Presumptuousness (SIA): Under the SIA, we have undue bias towards larger universes.
Pascal’s Mugging (SIA): Under the SIA, we can be forced to put full confidence in very unlikely universes if these universes have enough observers. Using a similar mechanism, the SIA forces us to believe with near certainty that we are either part of a simulation or a Boltzmann Brain.
Non-Bayesian Updates (Double Halfer): There are a few different Double Halfer methods, but what they all have in common is that they require something non-Bayesian like ignoring information (Updatelessness) or producing invalid updates (Bostrom’s Hybrid Model)[10].
Infinite Observers: This can mean a few different things in different contexts, and different assumptions all fail in different ways. The PPA cannot handle all possible distributions with infinite observers, but it can handle more than any other method, including all the ones one might expect given our understanding of physics[11].
The catch to this, which we will discuss, is an extra degree of freedom in our model. Specifically, we must decide which variables in our prior are “SSAish” and which are “SIAish.” Much of our discussion will center around motivating this choice.
How to read this Series
This piece has covered most of the important details of the PPA.
The rest of the series will dig in a little deeper, hopefully clarifying or defending any points you didn’t follow or disagreed with. If you don’t have time for all that but still want a little more detail, then this piece makes a good follow up, summarizing and synthesizing all of the previous pieces.
Otherwise, the first two thirds of this series are “low-math”; there are no real equations and all the proofs are visuals worked out on simplified examples. It contains several pieces, like the ones on indexical reasoning or improper priors, which include introductions to intermediate-level concepts that you might already know.
Only the last third of this series is fully technical. These parts will include full equations and derivations for the arguments made in the first portions, as well as handling of a few additional scenarios, such as infinite universes.
The pieces are labeled as follows:
Background [BG]
Low-Math [LM]
Technical [T]
And here is the full series.
[BG] Mosaic Plots for Visualizing Anthropic Distributions: A brief intro to a visual tool we will use heavily
[BG] A Brief Defense of Indexical Reasoning: Why anthropics must be self localizing, plus a visual derivation of Fully Non-indexical Conditioning
[LM] [BG] Parametric Modeling and Types of Uncertainty: Bayesian vs Frequentist definitions of probability, how they apply to the PPA, and an optional introduction to parametric models for those unfamiliar.
[BG] Improper Priors and the Infinite SIA: Rather than breakdown at infinity, sometimes the SIA lets you do math you otherwise couldn’t
[LM] Indexical Models Don’t Have Diversity Biases: Contrary to popular belief, observer experiences are just as varied under both the SSA and SIA
[LM] Temporal Consistency and Texas Sharpshooting: The PPA and SIA are Temporally Consistent; the SSA is Not
[LM] The Fermi Paradox, Fine-Tuning, and the SSA: The SSA is sufficient to believe in a massive universe, plus a discussion on the Fermi paradox
[LM] Putting it all Together: Revisiting the PPA and Basic Model in light of previous pieces
[T] Deriving the PPA: Statement and discussion of the full PPA’s definition
[T] The Basic Model: Full math for all of the arguments made in this piece, excluding derivations of the PPA for infinite universes
[T] The PPA at Infinity: Extending the PPA to infinite universes, along with a discussion on Boltzmann Brains and simulations. This piece is currently WIP but is planned to come out within a week or two of the rest of the series.
I would like to thank Justis Mills for reading and commenting on an early draft of this piece.
- ^
Though the PPA does technically give Sleeping Beauty a 1⁄3 chance of heads, it would be unwise to index too hard on this fact.
- ^
If you are a halfer, you can instead think of the PPA as what you get when you start with the SSA, resample the universe over and over, and then backfill the resulting structure to your original prior. If you are a thirder, then you can simply think of the PPA as the SIA with the “propensity for life” (mean expected number of observers) divided out, keeping all the original dynamics without presumptuousness.
- ^
AI DISCLOSURE: Absolutely no text in this series was written by AI, but most of the figures were either generated completely by AI or with heavy AI assistance.
The vast majority of the text and ideas in this series were fully written/developed without AI several months ago before the existence of models strong enough to meaningfully help. If I had known AI would become so powerful so quickly I would have written this up sooner.
Near the end of writing this series I decided to use Claude to proofread, mostly because recent developments in AI convinced me of the need to publish quickly. AI was used only to flag spelling or factual errors; at no point was it used to make any edits.
While proofreading two of the pieces, Claude ignored my instruction not to provide new ideas and wound up providing some insights that, while minor, were worth mentioning. I have included most of these as footnotes. In general, any time Claude did anything remotely non-trivial, I have tried to point it out so as to make it as clear as possible which contributions are 100% my own.On a few occasions I did have extended conversations with Claude about this work (using Opus 4.8 at the time), but most of these conversations were unfruitful. Claude did make one significant contribution relating to the PPA in infinite universes, which I point out when we reach it.
- ^
I actually don’t believe that this logic makes sense for the “many wake-ups” scenario of the Sleeping Beauty problem, but I do think it would hold for, say, an “incubator” version of the SBP where a new set of Sleeping Beauties is created each time. This is much closer to the situations we care about in this series. I plan to expound on this in a future piece.
I talk about these different scenarios a bit in this comment.
- ^
This is the Carter-Leslie version of the Doomsday Argument. Other versions, such as Gott’s Doomsday Argument, are not built on top of the SSA.
- ^
Using the notation introduced just a little farther down, define an indexical distribution
. This indexical distribution interpolates between the SSA and SIA in such a way that it manages to reproduce a version of basically every paradox from both approaches simultaneously. - ^
Assuming each row depends only on the row above it, the technical term is that the process has the Markov Property.
- ^
Non-indexical Double Halfer approaches will look at a bunch of pixels, devolve into Full Non-indexical Conditioning, and fail the same as the SIA. Bostrom’s Hybrid Model makes the right predictions in this particular case but it can be easily money-pumped in other scenarios.
- ^
The full PPA has the denominator
. This term allows the PPA to behave reasonably in universes that don’t produce any observers, something which both the SSA and SIA already do. - ^
Updatelessness also has the problem that it actually just reduces to the SIA in practice. This is known as “Full Non-indexical Conditioning,” which we’ll briefly discuss later in this series.
- ^
Specifically, we can handle the cases where (1) there are infinitely many species/simulations in the universe and each of these (2) almost surely goes extinct/terminates eventually
- ^
Better yet, assume the grid wraps around on the sides, so there is no edge
- ^
It does add a Doomsday bias, but as long as your species doesn’t permanently grow exponentially, this effect can be mitigated or even asymptotically disappear by all sorts of additional assumptions. Like a good Bayesian prior, the more evidence you get about how long you believe your species will survive, the weaker the anthropic Doomsday effect.
- ^
Though the PPA does technically give Sleeping Beauty a 1⁄3 chance of heads, it would be unwise to index too hard on this fact.
- ^
I want to head off two possible misinterpretations:
No, the PPA does not violate the Principle of Indifference. If you fix all variables and parameters, the PPA is uniform over all observers. This is because it does not reweight on a per-process level, it reweights by the expected number of observers at a fixed parameter setting.
“Subprocesses” do not have to be independent from one another or non-interacting. They are simply the units which we are dividing out of our model. All that’s required is that they be large enough that “how long do they go on” is a meaningful question, and that the expected size of a subprocesses is independent of the total number of subprocesses.
- PPA 3: A Brief Defense of Indexical Reasoning by (14 Oct 2026 15:03 UTC; 6 points)
- PPA 11: The Basic Model by (14 Oct 2026 15:03 UTC; 4 points)
- Sleeping Beauty and the Forever Muffin by (7 Jul 2025 12:05 UTC; 1 point)
- 's comment on Probabilities are not the right concept by (21 Oct 2026 20:59 UTC; 1 point)
Man, you can’t just go around solving anthropics like that! Now what are we going to argue about around here?
Seriously, fantastic work. This is now my go to reference for the subject.
Thanks! Glad you appreciated it. Quite a lot of work went into this.
Nice. I think PPA is compatible with centering-uniformity, which just says some measure over uncentered worlds such that subjective probabilities, conditional on an information state, are SIA-ish over compatible centered worlds, as if was the SIA prior on un-centered worlds.
If I’m understanding the math correctly, the way to do this is to set .
Moreover, like SIA, PPA satisfies an additional constraint on : for two uncentered worlds with the same epistemic parameters, differing only in their aleatoric outcomes, assigns weights proportional to the aleathoric probabilities of those outcomes. This is the sort of constraint we want for ordinary stochastic processes, e.g. ensuring that a physically fair coin which the observer knows has not yet been flipped is believed by the observer to have a 50% chance of heads, assuming the observer knows the universe’s epistemic parameters.
(Main difference between the “subjective probability over centered worlds” perspective and more traditional anthropics, is that in the first case, we interpret the measure in terms of a representation theorem for subjective probabilities over centered worlds, and do not assert that corresponds directly to the subjective distribution from any particular perspective.)
Possibly! Unfortunately I’m still working through your piece, but it looks very interesting!
The PPA definitely has a concept of “uncentering;” the that gets divided out is basically saying that the universe looks the same (in expectation) everywhere, so we’re going to divide out the total size of it. I think this might be the same intuition that you’re tapping into.
In general, it seems like if you want to make an indexical distribution that doesn’t violate the principle of indifference, your only option is to have for some . SSA says , SIA says . So in some sense the main contribution of the PPA is finding an that (a) eliminates the problems with the SIA and SSA, and (b) is actually justifiable from first philosophical principles.
Nice.
I auto-formalized your PPA in my anthropics Github repo here (includes proofs of some of your derivations):
https://github.com/GunnarZarncke/anthropics/blob/master/Anthropics/PPA.lean
I do not agree with this interpretation of DA. For half of the observers, DA predicts longer existence of their group than actual existence. For example, we have 10 ordered observers. For observers from 6th to 10th, the predicted 50 per cent length of the group is between 12 to 20.
So DA is not biased to shorter universes – and thus it is not a paradox. Therefore, we do not need to fight it.
What you’ve described is true, but that’s just a consequence of the SSA being a valid probability distribution. The Doomsday effect is different than that.
Consider a simple universe where in a loop we create an observer then roll a 10-sided die. We repeat this process until we roll a 10 and the process terminates. You wake up as one of these observers. Basically the grid universe in this piece.
The die is memoryless, it always has a 10% chance of rolling a 10. However, under the SSA, if you find out you are the first person, the chance that the next roll will be a 10 becomes 39%. The loop is disproportionately likely to end early.
If you prefer, instead of calling this “Doomsday,” you can call it temporal inconsistency, since it changes the probabilities of future events. Joe Carlsmith called it telekinesis and did a good write up here. You can also see it in the Sleeping Beauty problem where is famously 2⁄3, even if the coin isn’t flipped until after the beauty goes back to sleep on Monday.
(That said, in practice this effect can be a lot smaller than in these toy examples. See this piece).
Most observers are not the first in this example, so we should multiply our surprise (of the fact that it gives 39 instead correct 10 per cent) on my chance being first observer. In other words, we can find specific examples where DA looks wrong, but it is probabilistic argument and it is correct in average.
Things don’t balance out, though. In the 10-sided die example, every single observer in the universe assigns greater than 10% chance to the possibility of the 10-sided die rolling a 10. This effect disappears asymptotically as your observer rank increases, but it is always there. It is only in the limit of infinite observers that the probability reaches exactly 10%.
In fact, one way to derive the PPA (which I talk about here, same piece I linked in my previous comment) is to take the limit of how the SSA behaves in arbitrarily large universes, and then backfill that structure to all possible universes. The PPA is basically just the SSA without temporal inconsistencies.
If we use Gott’s Doomsday argument, then for 15th observer in some long chain it gives chances for 16th observer’s death as 5.9 per cent which is smaller than 10.
Carter-Leslie’s DA indeed gives above 10 per cent for any observer but it uses the fact that observers knows the dice probability = 10. However, it requires that there was only one sample (one chain not many) and it actually exists and is over.
EDIT: Upon rereading your comment, I realize that you were using Gott’s example exactly how he presented it, which means it wasn’t using any of the setup from my 10-sided die example; it’s a completely different thought experiment.
Presumably Gott was saying we should use a prior like 1/N when we have no other information to go on, but if we’re going to stipulate a thought experiment where the probabilities of certain outcomes are known, then we’re not going to use that prior. Or at least, we’re going to have to specify how we’re changing that prior to match the thought experiment.
Anyway, my point below still stands. Gott’s Doomsday Argument is a non-indexical argument built on top of a very specific prior that will behave weirdly when it starts interacting with other evidence because of FNC. Also, since it doesn’t use the SSA, it doesn’t counter the statement that the SSA is Doomsdayish.
That said, I will add a footnote to specify that I mean the Carter-Leslie Doomsday Argument.
--
I have only read about Gott’s Doomsday from this wikipedia page, but as far as I can tell this argument does not use the SSA, so it doesn’t contradict anything I said.
Instead it conditions only on the fact that there have been humans, not that you are specifically human . This makes it a non-indexical double halfer method, since the same logic in the Sleeping Beauty problem would have you conditioning only on “Monday happens in the experiment,” and not “today is Monday.”
I agree that this doesn’t yield Doomsday, but it’s also for all intents and purposes identical to using the SIA once you condition on arbitrary amounts of random information (this is Radford Neal’s FNC), which means it’s presumptuous about the total amount of life in the universe.
I think that two things needs to be addressed: 1 that SSA requires only one chain of observers, not many and 2 that the selection of the observer has to happen only after the creation of the chain is over. Leslie explicitly wrote a lot that his Doomsday argument works only if the world is determinate.
In that case, if we play only once and the game is over and it produced only 6 observer chain, than finding that I am observer number 3 is not surprising and is an argument against that there were 100 observers.
In your example, both conditions 1 and 2 do not hold, so your model is not argument against Carter-Leslie-DA. Moreover, our Earth is a) likely not unique—there should be other habitable planets somewhere b) Its future is not determined because of quantum an chaotic effects and MWI.
However, there is one more version of DA—Universal Doomsday argument—https://arxiv.org/abs/1303.4676 It basically says that when I learn my position n, I can estimate a typical age of civilization, and thus most civilizations in the universe must be Young. As most are young, most die young. As we are typical, we also have high chance to die young.
The Universal DA is better as there is no reading from the future and it accounts the existence of many chains. It can be tested: If I ask myself my age and multiply by 2, I get approximation of typical age humans live (around 100) – and I can use it to estimate my life expectancy.
First of all, let me own up and say that I’m clearly not familiar enough with the different flavors of the DA that have been put forward over the years. I’ll read your sources and see if that develops my thinking.
With that said, I’m having trouble understanding the context of your disagreement. What I said was that the SSA, as defined by many writers, including Nick Bostrom, Radford Neal, as well as almost everyone I’ve encountered here on LessWrong (Joe Carlsmith, Ape in the Coat, etc) produces a bias towards smaller universes and shorter species lengths. This statement is true, and I feel like we’ve already demonstrated that.
It’s also true, as Carter and Leslie evidently already pointed out (I will add a citation to the relevant piece) that as the number of species in the universe increases, the evidence for Doomsday you get from learning your birth rank asymptotically disappears. The problem is that if you’re choosing between “small universe“ and “big universe“ hypotheses, then the SSA will still produce a bias towards the smaller universe. You have to start with the assumption that the universe is big in order to avoid the paradox. If universe size is the thing you want to know,, the SSA doesn’t help you.
I wrote about this (including what happens with many species) a lot in the piece I linked in the other two comments, and in fact this observation is literally what the PPA is: a model of anthropics that behaves like the SSA does in big universes, even when the universe is small.
Finally, you mentioned that the Carter-Leslie Doomsday argument only really applies in a deterministic world. I agree with the intuition behind that, and that’s basically the argument I made in this very piece with the digit-of-pi Sleeping Beauty problem. The PPA is what you get when you apply this determinism argument not to the entire universe, but only to certain high-level aspects of it (i.e. keeping the laws of physics constant but varying the random way they progress).
Which is all to say I think there’s a lot in this piece you may find interesting; it was built on a lot of the intuitions you’re citing.
This comment is a nitpick about your treatment of the sleeping beauty problem. I haven’t read the rest of your series yet, so please let me know if any of my objections are covered later.
The SBP is first defined as:
But later you say:
But these aren’t the same event! SB’s credence that the coin is heads at all should be 1⁄2. Her credence that she would observe heads is 1⁄3, because she gets to do the observation twice as often on tails. You even mention that this doesn’t make sense in the “many wake-ups” variant. In that variant her credence that the coin is heads is still 1⁄2, while her credence that she would observe heads is 1/1000. Observing an event more times doesn’t change the probability of the event actually happening.
For the digit of pi variant, the image is misleading; it should show “Mon - [blank] - Mon - [blank]” etc. SB doesn’t get to live through twice the amount of Mondays in the even branch (or if she does, then it’s a different problem).
Also, what exactly should “probability of Tuesday” be interpreted as?
The probability that she lives through a Tuesday at all is 1 (unless you kill her on Monday)
The probability that she’s ever awake for a Tuesday is 1⁄2 (only happens in the odd branch)
The probability that any specific day she’s awake for is a Tuesday is 1⁄3 (50% chance of waking up for X Mondays, 50% chance of X Mondays + X Tuesdays → X/3X = 1⁄3)
I’m not seeing a definition of “probability of Tuesday” that makes the answer 1⁄4, except for the above mentioned variant where she lives twice as long on the even branch.
Thanks for the comment. I agree the language was slightly imprecise in the parts you quoted. I have tweaked both spots so it’s hopefully clearer.
Well, that’s kind of the crux, isn’t it? Thirders do believe something a bit like this: if more people observe a thing, then that thing is more likely to happen (from the perspective of a person with unknown identity).
Although maybe it’s worth it to take a step back here so I can try to understand what you’re proposing. How is it possible for “her credence that the coin is heads is still 1⁄2, while her credence that she would observe heads is 1/1000?” How do you envision those beliefs operating in practice? Does it mean that she’d be willing to bet 999:1 against heads even knowing full well the true odds were 1:1? You’re describing two completely different sets of beliefs that would presumably require the Beauty to take complete different actions depending on which one she subscribed to. That doesn’t work—she has to pick something and stick to it.
I cover this a little when I start formalizing things, though I leave a fair bit of it implicit, but the way I see it is there are really two distributions: a universe prior, which is how the universe looks to someone outside it, and the indexical distribution, which is how the universe looks from the perspective of a “random” observer. The SSA and SIA are two different ways to create indexical distributions.
The important thing though is it’s impossible to make both distributions (universe and indexical) identical, so you have to pick one. So yes, someone using the SIA must say in the name of self-consistency that they believe that, conditional on the observation they exist, the coin had a 1⁄3 chance of landing heads. They’re also allowed to say that if they didn’t exist, or if they were the scientist standing outside the room and watching, then on that evidence the coin would have gone heads with 50% chance. But, conditional on specific evidence, they must commit to only one single distribution over reality.
I did that on purpose; I think unfortunately you may not have grokked the argument behind the illustration. I tried to improve the wording before the illustration.
We do not need the missing space because we are conditioning only on the wake-ups that actually happen. The illustration is supposed to show that there’s a nested structure: we first break the universe into two possibilities, then divide up those possibilities by the number of wake-ups. If the digit were even, there would only be Monday wake-ups, which makes the relative number of wake-ups on an odd-digit irrelevant because there is no universe in which the odd-digit ever would or could exist. If it helps, try to imagine we were running the following algorithm:
Select “A” or “B” at random
If A, fill a list with 1000 “M”s
If B, fill the same list with 1000 “M”s and 1000 “T”s instead
Pick a letter randomly from the list
This procedure will be SSAish over Ms and Ts.
Also, I think if I filled in the blanks and treated them like gaps in the total probability mass, it would just be confusing. It would imply that “Even” has have half the probability mass of “Odd,” which would just incorrectly imply the SIA.
It means her credence that, upon waking up and not knowing the day, the day is actually Tuesday.
Yeah you’re right, it does start to get weird once you consider betting rules and such. What if we tried the following variant:
SB is woken up on all 1000 days regardless.
On heads, SB is only asked to bet on the coin’s state on day 1.
On tails, SB is asked to bet on the coin’s state every day.
I think this is isomorphic to the original, please tell me if you disagree. But it’s (hopefully) clearer now that even though she should bet tails at 999:1, her actual credence in the state of the coin should still be 1:1. The true odds and the betting odds can diverge if you have a skewed bet.
Yes, that’s all correct. SB’s credence in observing a heads awakening is 1⁄3, the scientist’s credence in observing a heads awakening is 1⁄2. I still disagree that observing an event more makes it actually occur more. Similarly I disagree that “event happens” and “I observe event” are comparable categories.
I think I see the disagreement here. You’re drawing the diagram to imply SSA because you think it’s obviously correct, but I want the diagram to imply SIA because I think that’s obviously correct. I still think 1⁄3 is correct here. The probability mass of even and odd are the same, but the probability mass of even wakeups is half that of odd wakeups. If SB were betting on the day, do you think she should bet 3:1 against Tuesday? If you took this betting rule to your “pick a letter” variant, you would make the most money betting 2:1 against T.
At the time Beauty is deciding whether to take the bet, she knows she has been offered the bet. According to both “halfer” and “thirder” rules, this gives her strong evidence that she is in the “tails” case. Sure, she will have a 50⁄50 distribution prior to knowing she has an opportunity to bet, but by the time she is deciding on the bet, she knows she has the opportunity. So the difference from the original here is that your new problem has a relevant distinction between 2 information states: “after waking up” vs “after being offered the bet”.
You do have a new information state, my question is whether that changes the problem meaningfully, i.e. changes the odds you should have or the odds you should bet at. If SB could reason while unconscious she would have a 50⁄50 distribution prior to being awakened, and then update after waking up. This seems exactly the same as having 50⁄50 after awakening and then updating after being offered the bet. I don’t think consciousness is special in this regard, you can imagine the credences you should assign while unconscious.
Main thing I was trying to contest was this statement:
There is no skew needed here, we note that at the time she bets, her odds are 999:1.
With “consciousness” vs “unconsciousness” the main thing that matters is when the memory wipe happens, which creates the ambiguity about where she continued from; according to thirding, that is where the probability change should happen.
For more on where I’m coming from with this, see Simple Dutch books for Sleeping Beauty halfers. Basically there are some theorems at least for CDT + thirding which relate it to ex ante optimality. If adopting this framework, there need be no conflicts between subjective credences and betting odds. There may be other solutions that work for ex ante optimality, such as EDT + FNC, though that’s less clear.
The thing I’m trying to get at is that her betting odds and the odds of the event happening aren’t the same thing. Her bet is scored every time she bets, but the event only happens once.
I think if you claim that the 999:1 odds reflect something about the actual state of the coin you can end up proving some weird things? E.g. what if SB buys a lottery ticket (let’s say 1⁄1,000,000 odds) and then builds a machine that’ll give her a million wakeups if it wins? She should bet 1:1 that she would observe a winning ticket on wakeup, but surely this doesn’t change whether she’s actually won the lottery in any given iteration.
The claim I’m making is that her betting odds can match her subjective credence on ‘odds of the event happening’ a posteriori.
Like, if a coin has already been flipped, and you know nothing about it, you should default to 50% chance heads. If you have seen it’s heads, you should change your belief to 100% heads. If someone asks you to bet on it, and you know they would be more likely to ask you in the ‘heads’ case (they look at the coin first), then them asking you should update you towoards ‘heads’. These updates affect your posterior even if the right prior was 50%. This can come up easily in non-anthropic cases.
Before sleeping, she should assign 1 in 1 million to her next wakeup being an instance where she won the lottery; we interpret “her next wakeup” in a physical sense of the next time her body awakes. If she adopts ‘thirding’ (which I claim is consistent and allows betting odds to match subjective credences), then when she wakes up, and receives news of being in one of these special awakenings (where there is 1 special awakening in the lose case, 1000000 in the win case), then her subjective credence for having won will be 50%. I claim this is basically reasonable.
The reason you might consider this a paradox is that you’re saying something about “would observe a winning ticket on wakeup”. But this has some ambiguity. If she has not slept the first time yet, and knows her next awakening is ‘special’, then the thirder credence on centered worlds will say, there’s a 999999/1000000 chance her centered world is “I have had no special awakenings yet, and I lost”, and a 1/1000000 chance her centered world is “I have had no special awakenings yet, and I won”, and given that centered world, it is possible to operationalize “my next awakening” as “my first special awakening” and get the non-paradoxical conclusion of “probably on my next awakening, I will have lost”
For more on what I mean by subjective probabilities over centered worlds, see this post.
Actually I thought this through a bit more and you’re right. If she looks at her bank account after the wakeup she should in fact expect to see the money 50% of the time.
Unfortunately, I think this is very much not isomorphic. How you bet is central to the entire debate. That said, this entire series is implicitly using a model where each universe creates indpendent observers and each one bets separately and keeps their money separately. This is actually different than an SBP multiple wake-ups scenario, since in that scenario each “observer” (who are all actually the same person at different times) puts their money into the same pot, so to speak. I agree that in wake-up scenarios only the SIA is reasonable.
Using vanilla SBP was a risk I was aware of but ignored in order to make the exposition simpler; I should have used an “incubator” scenario where each observer is a unique individual. Wake-ups give the impression that each wake-up must have equal weight, because the betting scheme implicitly pays out relative to observer count.
Think of it like this: If I make $1 on Monday and $1 on Tuesday, then I ambiguously have $2. But if I make $1 on Monday and my buddy Bob makes $1 on Tuesday, then we each only have $1, and both of us betting 2:1 would just make both of us poorer. Without everyone actually being the same person, there is nothing constraining observers in different universes to have the same relative weight.
The extreme example of this is if instead of wake-ups, we had clones. The Beauty enters a cloning chamber in building M, a scientist flips a coin, and:
On heads, nothing happens and the Beauty remains in building M
On Tails, a clone is created in building T. So there are now 2 beauties, one in each building.
The Beauty is asked, after the flip but before she learns which building she is in, her credence that the coin went heads. In this case, the solution is as unambiguous as it was in the wake-ups: she must have 50% credence. 1⁄3 will lose her money in expectation. If, for instance, all the clones take a bet losing $10 on heads but paying out $5 on tails, then on heads all the clones (of which there is only 1) will lose $10 and on tails all the clones will only make $5. Vanilla SBP only works because all the “clones” are the same person who keeps all the money.
Anyway, the point of that is to hopefully unstick your intuitions a little. Our universe isn’t a wake-ups or cloning scenario, and so which approach to use is less obvious.
My point was that repeated coin flips are like the wake-up scenario because they give every observer the same weight, since all show up at the same frequency in the long run. But the digit-of-pi scenario doesn’t, since you’re just repeating the same thing over and over. Either half of all observers are on heads, or they all are, and only the ratio matters.
Ok the distinction between cloning and wakeup is really helpful. The correct betting scheme changes depending on whether the other observers are you or somebody else. If you’re using “our universe” to mean “our universe as opposed to a hypothetical different universe” I think it does feel like cloning, or at least not like wakeup.
Let me see if I’m understanding this correctly: you’re saying the digit of pi variant is more like the cloning variant, and odd SB doesn’t care about even SB since their money can’t go into the same pot? In this case I do see how you should bet 1⁄4 for Tuesday.
I should probably also add that the main point of the PPA is that we don’t have to use a single interpretation of probability across our entire model. For things that we believe are truly random, it makes more sense to use the Frequentist interpretation of probability and get SIAish results, while for things that are fixed but unknown, the Bayesian/SSAish approach makes more sense.
Then it just becomes a question of which types of facts fall into which category. I argue that “longevity of our species” is mostly down to randomness, whereas “average number of times life emerges independently in the universe” (“propensity for life as I call it) is more a function of the laws of physics, which are fixed but unknown.
The similarity to cloning/wake-ups is, in my opinion, a bit incidental. I don’t see the overlap between their predictions and the different parts of the PPA as a major crux, which is why I opted to not include that discussion in this series. (Also, I had already spent way too much time on this and needed to cut myself off somewhere).
I think “truly random” vs “fixed but unknown” isn’t the right description of the distinction. It should be more like “fixed but unknown” vs “variable and unknown”. E.g. in the digits of pi case, if you checked a subsequent digit of pi each time you decided whether to wake SB on Tuesday, every individual check is fixed but unknown, but the overall credences and optimal betting scheme are now the same as the coin flip case, since SB will see an even mixture of even and odd digits in her own universe.
A better way to formalize it might be, given your current predictive abilities, how far forward in time can you go before you’d be capable of knowing the result? For digits of pi, you could do all the calculations right now and get all the answers. For the coin, you can’t plausibly calculate the physical trajectory before the coin is flipped. So in this sense I think the longevity of the species and the total number of species that arise are “variable and unknown” because we currently don’t know how to calculate either.
Ape in the coat has a good post on this that I generally agree with. They also have a series on SBP that I found enlightening.
I’ve read a few Ape in the Coat pieces but not that one, I’ll check it out.
Anyway, you’re right. Saying it’s “truly random” is probably not correct; it’s an imprecise simplification that maybe I shouldn’t be using.
Instead, let’s just use the actual Frequentist definition: the long term convergence over multiple repetitions. For a single digit of pi, the convergence is the same as a single sample. For a continuing sequence of digits, the convergence is (presumably, if pi is a normal number) that each digit gets 1⁄10 probability.
Then, when we don’t know what the convergence will actually yield, we switch to using a Bayesian approach.
This is how Bayesian statistics is usually modeled in practice. I did a write up on it as part of the series here.
Thanks, I read the PPA part 4 post and the aleatoric vs epistemic uncertainty thing looks like the physical vs logical uncertainty thing Ape in a coat’s post is calling out as a kinda-fake distinction. I agree on that, see the comments on that post.
If you have disagreements/confusions/comments on it I would be happy to discuss further, but in that case should we move the discussion to the PPA part 4 post instead?
Yup! That’s basically it.
Well, it depends on the type of your world-hypothesis. If you think world is a program that runs uhh the world in which you can be located, then sure.
But if your hypothesis is self first, as in, it’s directly about you and your perceptual inputs, then observing more often essentially makes event objectively happen more often.
[Removed, accidentally replied to the wrong thread]