Recently I read Wei Dai's "What Are Probabilities, Anyway?" post, and just wanted to see if the LessWrong community has now reached a consensus on the interpretation of probability. I am of the impression that most LWers endorse multiverse theories, whether it be the Many-Worlds Interpretation of quantum mechanics or the more radical Mathematical Universe Hypothesis by Max Tegmark. In both these cases, there exists additional worlds where all possible outcome occur (though the MWI is only restricted to "possible outcomes" that fit our current physics). An interpretation that many people seems to prefer is that all possible worlds are real and probability acts as a kind of normative "measure of caring" instead of a type of "objective reality-fluid", but I have my reservations about this. I'd love to hear the community's opinion on this today and whether or not there has been any new progress.

New Answer
Ask Related Question
New Comment

7 Answers sorted by

A probability is a quantified possibility. That doesn't tell you much about the ontology of probability, because it doesn't tell you much about the ontology of possibility.

A possibility could be a feature of reality, (a propensity) or it could be a merely apparent result of ignorance. (Knightian uncertainty). These are very different theoretically, but hard to distinguish practically.

Concrete examples of real possibilities include MWI branches or metaphysical possible worlds. Neither the meaning nor the usefulness of possibility depends on such realities. Ignorance is always with us, so possibility is always with us.

The quantification of possibility is a separate issue.

Probability is not always with us because it requires the extra element of quantification. Physics supplies no meaning or method for determining the probability of decoherent branches, for all that it is a topic of great philosophical interest.

If you are interested in the objective probability of the coin flip,the it only has one value because it is only one event. In a deterministic universe the objective probability is 1, in a suitably indeterministic universe it is always 0.5.

The assignment of probability 1 to an event that has happened is also subjective. You don't know that it had to occur with complete inevitability, ie you don't know that it has a conditional probability of 1 relative to the preceding state of the universe. You are setting it to 1 because you are proceeding as if it has occurred. Although determinism is equivalent to the claim that everything happens with probability 1.0, the fact that 1's appear in probability calculations does not imply determinism.

If you think the questions “what will it be” and “what was it” are different, you are dealing with subjective probability, because the difference the passage of time makes is a difference in the information available to you, the subject.

Failing to distinguish objective and subjective probability leads to confusion. For instance, the sleeping beauty paradox is only a paradox if you expect all observers to calculate the same probability despite the different information available to them. The trick is to drop the assumption that there is a "the" probability everyone has to agree on.

The ways possibilities are quantified varies, too. Bayesianism and frequentism are different methods of quantification that often agree. Bayesianism permits a subjective element without excluding objectivity. Frequentist assumptions can be used as priors, and Bayes can converge in frequentist conclusions.

The objective basis of frequentism is the measure of an equivalence class. The objective basis of frequentism is not an objectively existing possibility, ie. frequentism is defineable in a deterministic universe. However , for a measure to be translated into a probability, something has to happen. An urn might contain a certain ratio of black balls to white balls , but that is not a (frequentist) probability, until you imagine a ball being drawn at random. Randomness, even imaginary randomness, introduces the necessary element of multiple, non trivial possibilities.

Objective and subjective probability are not mutually exclusive opposites. Strict determinism does not imply any level of information for any observer, so observers can still apply probabilities based on their ignorance of how things turn out

All concepts, including probabilities, are tools for making better predictions of the anticipated experiences. One can discuss whether a single universe, a Tegmark multiverse, MWI or something else is a more convenient and consistent way of describing "reality", but that has no bearing on "what probability really means" or "what electromagnetic field really means" or "what consciousness really means". All these are concepts (models) that are useful within their domain of applicability, and not very much outside of it. There is no one universal concept that is always useful in every case.

Nobody does nothing but passively make predictions. Agents also change things. That's why they are called "agents", not "patients".

Definitely, I don't think I said anything to the contrary. We are embedded agents in most realist models of where observations come from, so the predictions include those of our own actions, at all levels. Scott Alexander's discussions of the brain modeled as an anticipated error minimizer go into some details there. 

If you want some philosophical foundation, I suppose the first step is to decide whether probability is in the mind, or in the territory.

The MWI/Tegmark approaches are trying to find probability in the territory; to define it as a frequency among some possible worlds.

The opposite approach is to define probabilities as a tool to deal with uncertainty. You can have uncertainty regardless of the underlying structure of the universe. (You could even have uncertainty about the underlying structure of the universe.) If thinking creatures can evolve in Conway's deterministic Game of Life, they too could be uncertain about some stuff, and could invent probability to describe their uncertainty.

"In the mind" and "in the territory" aren't mutually exclusive , and words dont have to have a single meaning.

Huh, I thought that many people supported both a Tegmark IV multiverse as well as a Bayesian interpretation of probability theory, yet you list them as opposite approaches?

I suppose my current philosophy is that the Tegmark IV multiverse does exist, and probability refers to the credence I should lend to each possible world that I could be embedded in (this assumes that "I" am localized to only one possible world). This seems to incorporate both of the approaches that you listed as "opposite".

To be precise, I think that MWI is probably true in our reality, but I think that probability is subjective and unrelated to whether MWI is true or false. Like, if the MWI is true, then for a hypothetical omniscient and perfectly calibrated being probabilities would be equal to frequencies of Everett branches. But if e.g. Copenhagen interpretation is true, then for the same being, probabilities would be equal to... uhm, probabilities in the collapse. And if we lived in the Conway's Game of Life, then I guess the hypothetical omniscient being could predict everything with 100% certainty, so the concept of probability would not make sense for them, but it would still make sense for beings with imperfect knowledge. In other words, probability is in the mind, but hypothetically speaking if your mind is god-like then your probability reflects something in the territory (because what else it could be?). I don't think we have a substantial disagreement.
Why not summarise that as "in the mind and in the territory".
A better summary is "in the mind and maybe parts of the territory".  It definitely exists in the mind.  We don't know how to test when and whether it's in the territory. 
In the mind that is trying to reflect the territory.

There is also the somewhat boring answer that probability can refer to anything which obeys the axioms of probability.

Be careful with ideas of what "most LWers" or "the LessWrong community" believes.  There's a LOT written here, some of which is useful, some interesting, and some kind of pointless.  The fact that something gets explored and not specifically refuted may only mean that it's fun to explore, and the scepics don't find it worth the effort to engage.

I'm pretty agnostic on MWI.  Not because I have strong counterarguments, but because it's hard to know what "true" or "real" even means outside my light cone.  Fortunately, it doesn't matter.  Bayesean probability is about prediction or knowledge of future experiences, not about reality (whatever that is).

One might be so bold as to add that people outside the LW community could possibly have something worthwhile to say.

I mean, sure, but I'm not sure how that helps with this question.
The question of what probability really means? Why shouldn't someone who studies it be able to help?
Oh!  yeah, would be helpful to include references to outside sources, huh?  Guess that advice goes for both of us.  I'm too lazy, why don't you go ahead?

I subscribe to the Jaynes/Laplace view of probabilities, namely that they exist in the mind and result from changes in information rather than changes in the world, let alone multi-worlds.

Imagine I tell you about an urn with black and red balls, without an additional detail. You can provide a probability of getting a black or red ball (50/50). 
As I provide more information ("there's 5 red balls but 50 black balls", "the red balls are sitting on top of the pile", ...) your probability assignment will change without the physical urn having changed at all. 
As your knowledge of the urn and the selection mechanism becomes more complete, your uncertainty decreases and your confidence/probability levels grow.

Again, the existence of in-the-mind probability doesn't imply the non existence of objective probability.

New to LessWrong?