Talk:Boy or girl paradox

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia
(Redirected from Talk:Boy or Girl paradox)

There is no ambiguity[edit]

I have deleted all mentions of the ambiguity, which does not exist. The question is a simple "what is the probability of A given B." How "you found B" is not of concern, because:

1. This is a problem about an abstract measure space in probability. The question is posed about the odds under naive mathematical assumptions, such as the fact that men and women are distributed uniformly, which is obviously not the case. The exact answer in the "real world", which would be an extremely long decimal, is of no relevance.

2. Even if you were to do some experiment in the "real world", you would surely do it in the way that generates a ~1/3 occurrence, because that is the way that is better suited to answer the question as it is asked about the abstract model of genders.

Please stop reverting my edits. — Preceding unsigned comment added by 216.164.249.213 (talk) 11:48, 11 December 2021 (UTC)[reply]

What you think, or I think, or any other editor thinks, is all irrelevant. Our task in Wikipedia is to reproduce as clearly as possible what reliable sources say. Peter coxhead (talk) 20:13, 12 December 2021 (UTC)[reply]

@Peter coxhead you don't seem to understand. The concern I am raising, and the purported ambiguity, has nothing to do with natural language. It is just wrong. It would be like citing the claim that 1+1=3 to a psychologist. It can be (and has been) explicitly shown that there is nothing unclear about the wording of the problem, per both analysis and other questions of the same structure. The idea that there is any ambiguity is a "cope" from people who solved the problem incorrectly. I can "cite" you countless phds, even math ones, who originally got the monty hall problem wrong, and displayed similar levels of confidence and dissonance to these phds, but yet they are still wrong. A claim being patently wrong overshadows however many sources you can produce for it, we don't need to post and cite every silly mistake an academic has ever made to wikipedia. If you would like me to write a cited section on the imaginary ambiguity I will do so, but until then, it's staying deleted. — Preceding unsigned comment added by 216.164.249.213 (talk) 23:50, 13 December 2021 (UTC)[reply]

The question is presented in natural language, it's nonsensical to claim that it has nothing to do with natural language or that you can by mathematics alone prove that there is no ambiguity in the metalanguage. BlueBanana (talk) 00:13, 21 December 2021 (UTC)[reply]
Precisely. But the key issue remains that some reliable sources say it is ambiguous. We are obliged to present all views in reliable sources, so if sources that say it's not ambiguous are under-represented, they should be added. But this cannot justify removing other views. In the old slogan, Wikipedia is about verifiability not truth. Peter coxhead (talk) 10:05, 21 December 2021 (UTC)[reply]

@blueBanana I knew someone would say this, which is why I addressed it several times with differing examples. The question is asked in natural language, but the purported ambiguity is not the result of natural language, it is the result of a "cope" by people who answered the problem incorrectly. Do you find it the least bit odd that of all of the academics claiming the problem is ambiguous, none have backgrounds that suggest they would understand the nuance in probability in the first place? That's because it doesn't exist, but as the above user stated wikipedia (clearly) does not care about truth, so i will give this up, have fun with your silly "pseudoscientific" circlejerk (you guys like this word, right?) Btw it is grand that wikipedia talks about science and what is or isn't "scientific" when it's users and the academics they cite haven't a clue what it is or how it works. — Preceding unsigned comment added by 216.164.249.213 (talk) 23:03, 5 January 2022 (UTC)[reply]

And do you not find it the least bit odd that of all the academics only examining the question with mathematics, none have backgrounds that suggest they would understand the nuance of natural language, reference, or meaning? You didn't address anything, you haven't done anything to even attempt showing there is no ambiguity in the question (unless you have written more unindented and unsigned comments here, but I won't bother guessing which ones might or might not be written by you). If you happen to read this again, I challenge you to give me any three propositional statements in natural language with no ambiguity in them. BlueBanana (talk) 01:13, 20 January 2022 (UTC)[reply]

As someone who also has a mathematical background, I want to add a few points in this discussion with our discontented anonymous editor. First of all: as an encyclopaedic entry on the problem, in-depth academic discussions of it can hardly be neglected, even if someone thinks they are 'irrelevant to the mathematics of the problem' (quotation for specific emphasis, not a quote). People have discussed this problem from philosophical viewpoints and wether those are really relevant to solving the original puzzle is now irrelevant, they are strongly related to the puzzle and are part of its history. This is the most appropriate place on wikipedia to present these analyses. However, I also agree that, in a way 'in all its simplicity', 'how' the information was obtained is irrelevant for the idealised abstract mathematical puzzle, for which the answer is 1/3, perhaps this ('mathematical') viewpoint could be added more clearly in the article, with citation: (not sure many mathematicians bothered to write about it, however) I managed to find a source, although not superbly academical, it is a book by a professor of mathematics which contains a discussion of this problem and a variation among other, various mathematical problems. The book is FUNdamental Mathematics, by David Eelbode, the problem is discussed on page 164 and onwards a bit. Perhaps other academic sources exist that make claims about how 'unambiguous and straightforward' the 'true' answer is, but please cite them then and they still wouldn't mean the rest of the entry can be deleted, as that is all relevant information. (A small note: the book I mentioned does also acknowledge the academic discussion of the problem by cognitive psychologists and philosophers. This wikipedia entry is not for people who think: "it's easy, just 1/3 because out of the equiprobable options Gg Bb Bg Gb I know it's not Gg and that is all, so I can't discriminate between the others and then the answer (naively) should be 1/3". That wouldn't take a whole entry and people who are interested in this should be presented with the relevant information from the academic discussion, at the very least 'as well', though I am personally not opposed to adding something like 'and a mathematician would just say it's 1/3, because simple idealised puzzle and whatnot'.) I also understand the frustration of the editor, but (addressing them) if you really want to get frustrated over ambiguity in seemingly trivial probability puzzles, go check out the Sleeping Beauty problem, though be careful not to waste too much time on these things. I recently added an extra reference there in 'Other works...' that quite thoroughly explores its relation to other probability problems and paradoxes such as Bertrand's boxes from Sylvia Wenmackers, which is a relatively recent addition in the discussion of the problem. (For other editors, I will add this to the talk page of the SB problem if I have time, but I wanted to add some things there as well, it seems the wiki entry has been edited quite a lot lately and is perhaps in need of some revision.) A last word to anonymous (only IP address known): I really do understand how annoying this can be, but try not to get aggressive, we're all just trying to be polite here even though it's the internet. The problem with some things you said is that these articles aren't regarded as 'mistakes' by these academics. And even if they were, unless a better paper points this out, they are the only sources available to quote on the matter. I would like to see an article that says 'it actually isn't ambiguous, as is claimed by, among others, … , because it is unnecessary to introduce information on 'how' the given in the puzzle is obtained, and introducing this as extra information further complicates the problem, in an interesting way, one might add, but it is no longer the original puzzle', however such statements would be considered original research at this point and as such (frustrating as it may be) have no place in a wikipedia article or long discussion on this site (those simply are the policies, you're welcome to call that part of 'circlejerking' as well, but encyclopaedias and summarising works often don't really include original research and I, for one, don't think that referencing more reliable sources in the construction of articles here is a bad method, although we could argue about the reliability of both some sources and some whole pieces of wikipedia, but that doesn't immediately seem relevant, or at least outside of the scope of the discussion of this particular page here). — Preceding unsigned comment added by MathsWolf (talkcontribs) 07:26, 15 January 2022 (UTC)[reply]

Actually, I think I should come to the defence of the anonymous editor a little bit more. Wikipedia has plenty of excellent articles on mathematics, but sadly a near trivial application to a simple puzzle is too much 'original research' to get a proper mention between the philosophy and psychologists papers on this simplest (allowing some hyperbole) of all probability puzzles that have been called 'paradoxes'. Consider a sample space . The event space is its power set and a probability measure for the event space is fully defined by defining it for the basic events that are all singletons containing one element from the sample space (e.g. ), which form a partition of the event space: . This completely sets up the probability space and the basic ('naive') mathematical assumptions for the problem. The puzzle now says that 'at least one of the children is a boy', so it is not , but rather it is one option out of . This means that a certain event is already given as being the case though: . This can be written in set notation as an event , which is the complementary event of and which we define as for notational purposes. By the axioms of probability measures (or Kolmogorov axioms of probability theory) we get: . Now we only still require the conditional probability . Here because , where the last equality holds because . The conclusion is . Sorry for all the pretty custom notation, that could be cleaned up a little, but the point stands: the solution is mathematically unambiguously decided by a straightforward application of the rules of probability theory. Philosophy and psychology are not further relevant to the correct interpreting and solving of the problem or puzzle, though they may have interesting things to say about it, about how people try to solve it and deal with it and about related things. Now we just have to find a paper where someone says exactly what I said above, so it can be used as a source to write this in the article. Of course the discussion of the ambiguity shouldn't be removed, it is still relevant, but purely considered as a mathematical problem it seems apt to show the solution and the purely mathematical point of view as well, which may alleviate some confusion and frustration readers might (otherwise) have. I have an Important note on this! I keep reading about the 'extreme assumption' that is supposedly necessary. However in the approach outlined above, no such assumption is necessary, rather , as a consequence, because of the definition of the events. Most important to notice is that 'At Least One Girl' and 'At Least One Boy' are not complementary events, which means that it is false that the probability of one implies the probability of the other by the complement rule. They can perfectly both be one. Here one might say the question of interpretation rears its ugly head again, because what I have discussed here applies to wether 'ATOB' or 'ATOG' are the case, not about wether we have received this information. However in all texts it appears to be a silent assumption that you are given either the info that there is at least one boy or the info that there is at least one girl. Never once is it considered that you could receive both these pieces of information, because they in fact overlap in the two cases BG and GB. Sure, it may seem ridiculous that you would just be given the answer to 'what the exact case is' in a probability puzzle, but that does not mean it should be neglected, definitely not if it significantly influences the relevant probabilities. The fact is that 'wether we have/get' certain info is confused with 'is the case', which causes the philosophical, linguistic and then also psychological relevance and confusion. The fact is that, considered as a maths puzzle, this is irrelevant when correctly approached. When we want to go deeper and consider a specific situation for which we would actually think up experimental settings, which, granted, certain alternative wordings of the puzzle may lead to that, then the discussion is still relevant and interesting in all its aspects. The fact remains that when simply stated as a maths puzzle, no extra info, no story, then the whole statistical experimental setup possibilities as described in the article, and all the 'extreme' assumptions are entirely unnecessary, because there is indeed no ambiguity. The ambiguity gets added by considering slightly altered questions, trying to find 'equivalent' wordings of the original question to answer, because the original one only lends itself to the above abstract mathematical method, and not so much to a more intuitive approach. However the problem then arises that not exactly purely 'equivalent' wordings are found, new nuances are added and these do indeed change the problem or add ambiguity that was not originally present, although many people will naturally add this in. I do think the whole discussion is very interesting and perhaps a lot of good research has even come out of it, discussions on methods of sampling and how you have to be careful, that different approaches might give different results. However for the original puzzle, "Mr. Smith has two children. At least one of them is a boy. What is the probability that both children are boys?", is not ambiguous. 'How the information was obtained' is entirely irrelevant, it introduces parameters or variables that cannot be known and assumptions must be made about, which sparks the discussion. However this can be avoided altogether with the mathematical analysis as outlined above. In the 'real world' it could/would be relevant, but that's not the puzzle. An interesting remark still to close this: in the discussion of the Sleeping Beauty problem, some 'solutions' have been given that approach it exactly as I have approached this problem here. The issue with that is assigning probabilities to Monday and Tuesday, which leads to the difference between the halfer and thirder and also to the double halfer who has a problem with that. Additionally 'being the case' and 'having as information (epistemically)' is also a problem there, however I won't go rambling on anymore here about that. — Preceding unsigned comment added by MathsWolf (talkcontribs) 12:14, 15 January 2022 (UTC)[reply]

It is true that there is a clear and objective answer to what is the probability of , given the probability space you described. However, it is not clear or unambiguous at all, that such a question is what the question "Mr. Smith has two children. At least one of them is a boy. What is the probability that both children are boys?", presented in natural language, refers to.
  • If one first finds out that Mr. Smith has a son, and later finds out they have another child, this is a situation that can, in natural language, be described with "Mr. Smith has two children. At least one of them is a boy."
  • The phrase "Mr. Smith has two children. At least one of them is a boy." can, then, refer to a situation in which .
Thus, as the phrase "Mr. Smith has two children. At least one of them is a boy." can refer to multiple contradicting propositions, it is ambiguous, and similarly, the question "Mr. Smith has two children. At least one of them is a boy. What is the probability that both children are boys?" is ambiguous as well. BlueBanana (talk) 15:41, 19 January 2022 (UTC)[reply]

Ok, I know I'm writing a lot (I'm getting messier and less rigorous), it's just that I thought I had something to add above, but then I kept adding bits and now in review everything I wrote after the purely mathematical working towards a solution is often times badly worded and perhaps not entirely correct. My main point about 'being the case' (which is a problematic one, even more so for approaches to and philosophy of statistics when you consider I put that as opposed to 'having information of'. Perhaps you should interpret 'being the case' here also more as 'could be the case', 'potentially', with a certain probability.) More to the point though, I think I can also mathematically explore this problem and ambiguity without making it too problematic in any way, hopefully at least, and my above solution stands, but it goes a bit into the 'having information' bit. I write to mean 'being/having been told not Gg (so at least one boy is strictly equivalent to that)', or in other words 'having/gaining the information that…'. Then instead of what we did above, all of which still counts, we assume that setup (although I will use a bit more convenient notation now), we now ask . We work out:

Being told that it is not Gg will definitely happen in case of Bb and definitely not in case of Gg. So we now get the result (skipping calculations):

In this case it is true there are two 'free variables', two parameters that are essentially unknown. They are, however, also entirely irrelevant to the question posed. After all, what I am actually assessing is the probability of 'being told this', but the original problem made no mention of 'being told' in this sense, there was no little story in which Mr. Smith tells you about how at least one of his kids is a boy. You just get the dry minimal mathematical knowledge in a short statement. No mister Smith, no one telling you anything 'in the story'. The 'informational' stuff can lead to a problem though. If you'd want an answer of 1/3 here, both probabilities still left in right hand side of the equation above would have to be one, like in the 'being the case' stuff. But the same would go for the girl case. Which would mean in case of Bg or Gb, you would always get told both. Which means if you're not told both, it's not one of those, so in the case discussed here, it would… hmm, I think I missed something, I also completely ignored the case that you 'weren't told anything', plus my own statements seem more and more philosophically problematic and ambiguously interpretable, how applicable and to what they are becomes difficult to assess. Just the setup of the puzzle would also have to be taken into account, definitely if you'd want to apply 'being told' on what is actually told to you in the puzzle. You'd have to assume things like 'they'd either only write at least one boy or at least one girl, one or the other, exclusively, always' and then about which when, probabilities… but there may be alternatives, if you want to try to treat it as openly and generally as possible at first. Sorry for the ranting, I'm out of time to seriously spend on this right now. My useful contributions probably ended after the purely mathematical working out of the question, but in my attempt to add nuance and discuss it further I ran into exactly the problems that are the problems with these sorts of paradoxes. Perhaps I will revisit this some other time with a clearer head, but I still think that the purely mathematical approach that does not concern itself with the 'information' is correct and solves the problem entirely unambiguously, as it was worded. When we start pondering other concerns though, it does become interesting and immediately problematic. — Preceding unsigned comment added by MathsWolf (talkcontribs) 14:57, 15 January 2022 (UTC)[reply]

Proposal and argumentation for (minor?) rewrite of the article as of 2022[edit]

I will start off by saying I do not intend to 'remove' a discussion of ambiguity from the article. However even the Bar-Hillel and Falk article says in its conclusion that the problem statement, as it is given here on wikipedia as well, unequivocally results in 1/3. We have plenty of sources of psychological papers that investigated people's misinterpreting and using heuristics that lead to faulty outcomes, strictly speaking. When considered as a 'textbook' abstract and idealised mathematical puzzle, the answer is clear and straightforward. The point of much of all the discussion though, of course, is 'real life' as well as alternative formulations of the puzzle that imply different models for the solution. The article, however, does not cover 'alternative formulations' so much as that it gives readers the impressions that the second formulation can validly be interpreted in a way that gives 1/2. However many of the sources used in this discussion are concerned with how alternative formulations give rise to different models, not with how this formulation gives rise to different interpretation. There is only one logically correct interpretation of the current statement. In real life, when concerned with statistical experiments, it is true that there are extra concerns and how the information came about and how we received it, as among others Bar-Hillel and Falk have more properly discussed, is indeed very relevant and necessary knowledge and people should be aware of this. So, perhaps contrary to what I have said before, I do think that point needs to be stressed, about real world statistical experiments. The reason for this, however, is not that 'formulation/question two' (as given in the article now and as I have previously discussed) 'is ambiguous', because it is not and different answers are achieved only by an ignorance of probability theory or a wilful misinterpretation in which extra unnecessary assumptions about an imaginary, fictitious situation are added to the original problem statement to change it so that another answer is obtained. The whole point is that in reality, the details matter. Because the details and the strictly logical treatment of the problem, insofar that is possible (there are problems that can be stated ambiguously, of course), matter, however, I think the current treatment of the problem given in the article is a bad one. I also have a strictly logical argument for that which I will give later. Readers are given the impression that they would be justified in misinterpreting strict logical conditions, even though none of the reliable sources claim that that is allowed and it not only misrepresents the controversy, which also largely has to do with different formulations, but it goes against the very point that papers on this subject make: the way in which information was obtained in statistical experiment matters. However, the Bar-Hillel and Falk paper itself mentions that there are problems in which the conditioning event can be identified with the information itself, directly and automatically: textbook problems. They continue to provide exactly the formulation currently given in the article as an example of such a problem in which this is the case. In these problems 'imagining how the statistical experiment happened' is unnecessary and doing so can change the probability, because a straightforward sample space was already implied in which the 'given' with application of the correct mathematical rules can be used to obtain the answer, while when you imagine and start adding in extra elements you may change what you are actually doing. The fact that you must be aware of the effect of extra information is the point and is what is always being stressed. Here I use 'extra information' also to mean the entire method by which certain specific information was obtained. The conclusion of the Bar-Hillel and Falk paper contains two main points: one is precisely the same as the point Bertrand wanted to originally make with his problem of the boxes and coins, known Bertrand's boxes, which is that you can't just assume a change to the sample space changes the probability you're looking for, because probabilities of specific events may be changed in such a way that they leave your event of interest unaffected in the end. The other point they make is about what they consider to be a second 'rule of thumb', which may also be part of the research of the 2004 paper of Fox and Levav, though they call these 'heuristics' as is common in such psychological studies. The second rule of thumb, or heuristic, is that 'information is irrelevant if if it doesn't matter which of its alternative values is supplied'. The authors correctly identify this as an extremely faulty heuristic that would lead to a lot of wrong outcomes, as it has for a lot of people who tried problems like these. Many people, even on these talk pages, have argued using exactly that rule. However it is not a logical rule, because 'a specific value is given' is not the same as 'more vague general information about possible values is given that is less restrictive'. I will add in my logical argument here. It is not so much of an argument as it is a demonstration:

Compare the above derivation with the following:

What is the point of this? Why am I writing these short logical statements, what is the meaning? In the first case above, the premise says that one specific individual, one specific child is a boy. One out of the two distinguishable specific individuals that are the two children of Mr. Smith. By applying the logical rule of addition, for example, we can entirely easily logically derive a new logical statement that represents 'at least one of the two children is a boy'. In other words, we can derive the statement, the information, from different more specific information. However when this statement is the only premise that one has, it is impossible to derive with the rules of classical logic (propositional, predicate, set, class or otherwise) that one specific individual of the two is indeed a boy. In other words it is logically impossible to capture the specificity of which child is a boy if only the purely logical information "'at least one is a boy' is given. Another confusion that arises is the difference between 'equivalence' between logical statements and 'derivability', or rather between 'replacement rules' and 'derivation rules'. Some logical statements are equivalent and can be modified by valid rules and changed into one another, while given certain premises there are also rules that allow us to derive new logical statements that are however not logically equivalent to the former ones, but rather 'deducible' given the former. They also hold, given the premises, but they are not of exactly the same form or something equivalent to that. So just because statements are consistent, just because some view are coherent, doesn't mean they are all logically equivalent, not at all even. To show this even more clearly, considering traditional classical logic, we can also do the following:

Many will recognise the law of the excluded middle from propositional logic. 'Of course this holds' and with the specific propositions, which are predicates for child one being a boy, it holds if the first premise is given, it can be derived/deduced from it by logical rules. This is valid logical reasoning. However we can not, with the same logic, do the reverse movement, if 2 were given, we could not deduce 1. I am not suggesting we put all this intro to formal logic into the article, it is just here to help me make my point about some erroneous reasoning people have displayed about this puzzle and to make sure no one would start that again as a response. The information supplied in the current formulation is only that the statement holds for the problem we are considering. It does not allow us, strictly logically speaking, to validly derive that one specific child is a boy. (I would almost say we should somehow consider it as an ensemble because of this, in a way, and treat it appropriately accordingly, but this might lead us too far and may not really be relevant, although may lead to interesting other variations and treatments. But disregard this.) That (one specific child is a boy) would be an assumption, that can find its place in 'how the information was obtained', because if this is assumed, the information originally given can be obtained indeed, deduced even, and it would also hold, but only the original information necessarily holds, it is not sufficient reason for the assumption to be true and the question arises wether such an assumption is (even) necessary (do we not already have sufficient information to solve the puzzle, unequivocally?). However, as even Bar-Hillel and Falk pointed out: in a textbook mathematical puzzle of the form 'what is the probability that Mr. Smith's children are both boys' given 'at least one of the children is a boy'", the information supplied can directly, automatically, be used as a conditioning event and will, unequivocally (they used that word)(under the standard (common) assumptions) lead to 1/3. So the problem is that an intuitive conceptual interpretation makes it into a problem in which the method of obtaining the information is relevant for the outcome, which a strictly logical 'textbook' interpretation does not. The problem, as it is written in its current form in the wikipedia article, is hardly (I argue 'not at all') ambiguous. The ambiguity concerns 'if the problem was formulated differently' and is mainly relevant in statements such as 'we see Mr. Smith walking with a child'. My issue with the current article is that such formulations and treatment of their parameterised solutions, to show that certain assumptions of 'how information was obtained' must still be made (I am talking about things like the fact that 'the walking companion was chosen' which has certain probability that we must assume) and is relevant to the final result. My criticism of the current article is mainly that it says 'two different procedures for determining "at least one is a boy" could lead to the exact same wording of the problem'. I disagree with this. The problem as it is currently worded in the article (Mr. Smith has two children. At least one of them is a boy. What is the probability that both children are boys?) is almost exactly a textbook problem and should thus ('unequivocally', if one remembers) have solution 1/3. It is however true that the statement would also logically hold and be derivable under different possible circumstances, in different possible situations that lead to obtaining this information. However these are not the problem that is given, this is not the question that is actually asked. We should both make clear that there is a textbook formulation which has a well determined answer and that any seriously less clear formulation that sets things up in a way more similar to an actual statistical experiment or observation, requires knowledge of the exact procedure and may not be solvable without additional information or assumptions. Many of these different formulations start with deceivingly similar, yet different givens and therefore often lead to different solutions and conclusions or final results. The point is about what matters for the solution method. This is not stressed enough in the article and it is stressed with regards to entirely the wrong question and in the wrong way. The article, its explanations and notation are pretty terrible in that way as it stands. I understand that this article has a long and difficult history with many disputes and discussions leading to what it is today. However I think it can be better than it currently is. Starting with a more complete and correct representation of what is actually said in the Bar-Hillel and Falk paper, but otherwise also with regards to the discussion of formulations and solution methods, because 'interpretation and ambiguity' are not truly the corse issue and should not be the takeaway, the way these are currently stressed and treated may lead people to think justified precisely these fallacious heuristics and approaches and they will walk away with a misunderstanding of the mathematics, or no grasp of it at all, but rather thinking 'I can just interpret it however I want and get my answer', which is not quite exactly the case. I don't want people to be misinformed by this article and walk away with an even more deep misunderstanding of probability and statistics, while this article is precisely an opportunity to help clarify these things, if it has clear explanations. I do not want to push a non-neutral point of view. I just want that the article highlights the actually important points and does so in a way that is correct. Claiming that the problem, as currently stated, is ambiguous, is incorrect and even in contradiction with many of the most reliable sources that are referenced in the article. However I do not want to set out on my own as a relatively new editor and get editing on this much debated article. So I wonder wether any other editors want to discuss this and collaborate making changes, because I am not comfortable making any major changes before this has passed by anyone else and then get a load of NPOV, vandalism and other accusations thrown at my head. I don't want to vandalise, but I think changes for the better could be made. But I am not experienced enough to set out on my own to do this without it first having been discussed with an editor who has some relation to this article. I am generally careful in my edits and try to just be outspoken on the talk page, hoping much talk here can lead to small but meaningful improvements to the quality of the article, that have been well considered by multiple editors. It is easier to be captivated by a subject than to write well about it and you always need feedback from other editors. MathsWolf (talk) 07:35, 19 January 2022 (UTC)[reply]

As one final addition, I will add a problem that is isomorphic and that is as close as possible to the 'textbook' like formulation, but now with the question expressed in terms of systems and probabilities of failure, or rather components being 'good' or 'bad' for simplicity (it may just be vaguely familiar for some people who have had probability courses in an engineering background). The problem is very simplistic. I do this just to show that straightforward examples of 'learning information' like this are possible and so some people could find the problem misleading because of the specific things it talks about and their unacquaintedness/unfamiliarity with and grasp of the necessary mathematics, but that is not a reason for an approach to be 'correct'. Consider a serial system S (we could do a parallel one and change the question a bit, mutatis mutandis, as well) with components A and C as follows (I can't draw things very well here, I should learn how to show things schematically on wikipedia so I can represent things in ways that are easier to see and understand):

    | ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅| S
  --|------| ̅A̅ ̅|----| ̅C̅ ̅|------|--
    |__________________________|

We shall assume (the working or failing of) these two components to be independent. If at least one of the two components fails, the system fails as a whole. As long as both components work, the whole system functions. Each component can fail or not fail (so work) separately. For achieving similarity (even (almost) equivalence/isomorphism) with the situation with children who are boys or girls we assume that each component is independent of the other and that they are either a good or a bad component and that being good or bad is equiprobable. If the component is bad, it fails and thus the system fails, if it is good, it works, and if both components are good (so work), the whole system functions and doesn't fail. We can denote the possibilities this gives us in a multitude of different way, but in the end it is a matter of notation and we will write GG to denote that A is good and C is good, the first letter denotes the first component, A and the second letter denotes the second component C. We could treat this in a different way and give the events different names but this changes absolutely nothing about the mathematically technical meaning of the whole thing, the result etc. so we'll go with this (for reasons that are hopefully already obvious). The system will now fail if A or C is bad, so either A is bad or C is bad or they both are, denoted by cases BG, GB and BB. Now assume that we know the system failed, or you are given as information to solve this little textbook problem that if the system failed (which is entirely equivalent to 'at least one of A or C is bad' and to nothing else, nothing more and/or nothing less). Now that information can be immediately used as 'given', taken to constitute an event over which is conditioned. In our sample space, this excludes the event of GG (from the prior, for (getting) the posterior). The given event has probability of 3/4, I will denote it with an F for conciseness of notation. (It is obviously a terrible system, but hey that's the problem we were given and set out to solve, not the design we'd accept.)

That's it for my discussion of this besides purely practically 'how are we going to improve the article'. MathsWolf (talk) 07:35, 19 January 2022 (UTC)[reply]

I said I was done with this, but I just quickly reviewed something about the different formulations used in different sources and this seemed really important to mention and is relevant to how the article should really be reviewed. Grinstead and Snell discuss the formulation: "Consider a family with two children. Given that one of the children is a boy, what is the probability that both children are boys?" In this formulation of the question, there really is an ambiguity, and so it is the one most relevant to the article and should definitely be included, because in this case it is indeed really not clear wether we are allowed to assume that it is a specific child who is a boy and so we could bring up the whole argument of 'how the information was obtained' again, this time with the assumption being absolutely necessary for the interpretative part, because it is not so straightforwardly translatable. So now it becomes dependent upon the assumption and so is truly, definitely ambiguous. I think I will already add this formulation and perhaps a short explanation to the article, just a minimal edit for the time being. MathsWolf (talk) 11:44, 19 January 2022 (UTC)[reply]

What's the difference between that formulation and the original one? They seem completely analogous in their ambiguity of reference to me. Is there a difference between stating "A" and "given that A", or is the inference of including the words "at least" (which are implicit in this formulation as well from the facts that one children is boy, and more than one children may or may not be boys) in the original formulation relevant? BlueBanana (talk) 15:53, 19 January 2022 (UTC)[reply]
I was about to make a small further note on this. However I will first address the 'at least' point, though it relates to what I'm about to say further. First of all 'at least one (of the two)' translates into a rather specific logical expression that could be represented set theoretically or class logically in which case it can stand for an event that can be conditionalised over and that expression is not equivalent to the assumption that a specific given child is a boy and that the other one is unknown. The reason they are not equivalent is because the problem with any specific child is, in its most simplistic form, isomorphic to the 'oldest child' problem, however it requires one child to be labeled and declared to be a boy, this taken as a premise and not assuming anything about the truth of the other child being a boy. This requires the 'other child' to be a possible label though, which is only possible in a simple way and not problematic if there is a known specific child who is a boy and another unknown one. However this means that instead of assuming and using as a given , to represent 'at least one of the two children in question, may be just one, or just the other, or both', one would assume a statement where the labelling means something slightly different, namely in which stands for the 'known child'. However this is only possible if in 'how we came to know, how we learned' there is indeed a known child that we are able to label. In absence of this as a necessity, doing so would be making a very specific assumption and conditionalising over that is no longer the same conditionalisation as we would be making over the minimal information that was originally stated in the problem. There are a few notes on this that I was going to add related directly to these last few sentences. The point is that under the standard, basic, common assumptions for equiprobability and independence to get a most simple idealised mathematical model, the question sets up only a minimal space, wait a moment before I go on further about that I really should make those 'notes' I was mentioning. There may well be considered to be an epistemic difference between 'I learn A' and 'I know A'. In this case, if the statement said 'you learn(ed) A', then you would be entirely justified to ask 'how did I learn A'? Because learning would imply a procedure and then you can think up a story, a specific situation, in which case exactly what you imagine and assume matters and can make a difference for the final result, in a way because with different assumptions you're actually solving different problems. In the case of 'you know A' you could argue: 'I could also ask: "How did I come to know A?" ', however in the case of a problem statement such as this, you are essentially given a proposition to be supposed true, nothing more. 'How I came to know' only matters in the story about Mr. Smith, which, because it is an abstract puzzle, has nothing necessarily to do with your reality, and so the answer for you would be 'because I read it in the problem statement', but that does not really have anything further to do with the world of Mr. Smith, you are not necessarily in that same world, you only have this one piece of information about it. It is a bit silly to ask 'but how did the person who wrote the question come to it?', because they can just have made up the question. Given only this basic information you cannot update in any other way than over without further assumptions, but it is possible to treat this as the 'given' and calculate that conditional probability, without further specific assumptions. Otherwise you start adding in your own hypothetical propositions and what you are calculating is the probability, would those be given and supposed to be included in the 'given' condition event. That is what you do when you assume a specific child. Specifically for what you asked: is the probability of 'A given B', so the formulation 'given B' implies exactly what conditional probability you are supposed to calculate. However the difference now is that while the formulation of Grinstead and Snell is 'textbook' in wording it with 'given', the point here actually is that the formulation of specifically what is given is actually logically ambiguous, as opposed to 'at least one' which is not ambiguous and can be translated uniquely (save for/up to equivalence) into a logical statement. Because 'given that one of the children is a boy' seems like it at least might imply that you would be justified in assuming that they are talking about a specific child, that 'one of the children' can be assumed a specific one. In either case the statement that 'at least one of them is a boy' holds, because in one case it is specifically given as the premise to be assumed true, while in the other it can be derived, but is not equivalent to what is supposed to be originally given. The necessity arises to 'suppose' one or the other, what is specifically the case is not clear and so it is not 'logically' as straightforward in this case, which opens up the question to complication because it really is an interpretative problem, which means not one unique answer exists, different are possible depending on the interpretation (because to be able to translate the (problem) statement, you need to interpret it and there is no one simple interpretation possible). I think there is something to be said for the fact that 'in reality' if someone were to tell you this problem, with reference to definitely real people, describing a real case, then a lot of extra things do indeed come into play. How did they learn this? Why did they tell you this and specifically this? You can then take a Bayesian approach and interpret all of that in terms of degrees of belief, often applying a flat prior/principle of indifference as you see fit (which may at times be debatable). Statistically speaking, the Bayesian approach would be equivalent for a specific problem, however when considering an abstract mathematical puzzle with a simple problem statement I think perhaps because of the specific philosophical approach to the interpretation, being an interpretation of probability, a philosophy of it in a sense perhaps, approaching it in a Bayesian way may tempt one to treat the problem as asking about their degree of belief, given certain info, about something, in which case one would have to imagine it, as if it were a sort of story, in a kind of realistic way, in which we are part of the story and Mr. Smith is part of ours. Then we must imagine a way in which this information could have come about. In this case, the problem is not well determined because that information is absent and so it cannot be solved, unless additional assumptions are made. The problem I have with this is that this assumes a very specific conceptualisation of the problem about which there was nothing in the problem statement. Treat it as an abstract idealised mathematical puzzle and interpret the given statements purely, strictly logically and the problem is solvable (straightforwardly) as I have argued, however this only goes for a 'textbook'-like formulation where the sentence to be translated into the relevant symbolic expression can straightforwardly be translated. If the problem implies a conceptualisation as mentioned in any way, the problem becomes different. The Grinstead and Snell formulation seems like a textbook formulation, but it cannot be treated straightforwardly as a textbook problem because the 'given' does not have a single strict(ly) logical meaning. 'At least one of them is a boy' does correspond to a logical statement, and assuming one specific child to be a boy in that case is not a correct/adequate logical translation, because there is no specificity given. That would be part of 'how the information was obtained', which as Bar-Hillel and Falk in their own discussion of ambiguity state, is not necessary to further specify only in one kind of problem: problems like the ones found in textbooks. Here we can assume the most simplistic model and identify the event for the conditional purely and strictly with what is specifically given. In this case 'at least one' is not ambiguous, however 'one of the children' is because 'one of the children' can be taken to be a deictic element, an indexical, to mean/refer to 'a specific one', which can be read as meaning 'one of the children is known to be a boy', which makes it a specific one (allowing us to assume 'either it is the youngest or it is the oldest' (not inclusive but exclusive or, difference with 'at least') and we are given it is specifically one of these cases, and which we choose does not alter the result, so we calculate assuming one of them, which makes it identical to 'the oldest is a boy', this is because 'the one about which we know' is now a specific label that makes sense, while there is no such individual in 'at least one'), which is not logically implied nor present in 'at least one', in which logically there is no alternative non-equivalent proposition or predicate sentence that could be meant in the metalanguage, without adding in further assumptions that are unnecessary for the solving of the textbook problem. MathsWolf (talk) 21:21, 19 January 2022 (UTC)[reply]
To summarize (I'll try to keep this brief), your argument seems to rely heavily on that there is a way to approach the question as if it was a textbook problem, and with that approach there is one correct way to interpret its propositional meaning - that is, were it a textbook question, there presumably would be a meaning and a correct answer associated with that meaning that the writer of the question tried to convey. However, while arguably true and possible to defend, this assumption is far from consensus within fields such as philosophy of language. By saying that a structure of a sentence and the words that it is constructed of determine its meaning, you're simultaneously assuming a prescriptivist view to grammar, and a view to philosophy of language that is close to that of Russell's, but by bringing up its context (which you're doing, even if the said context is "as a question that is removed from any contexts (real life situations in which it'd be asked)) you're getting closer to use theories of meaning, like that of Wittgenstein. The co-usage of the latter two would be especially difficult to justify to defend your specific interpretation. In the hypothetical scenario where this was a test or textbook question, you might get better score by arguing that the question is not ambiguous, but saying that a sentence has a specific meaning because assuming that meaning in a specific context would be beneficial is a view that, again, is not that out-there, but is not widely accepted either, and that is what your argument would have to boil down to, because there is no other justification for favoring the context of your choosing. Besides, if you accept the argument thus far, you might want to consider the possibility that your professor would disagree on your interpretation of the meaning of the question presented in natural language (we're running into the issues of self-referential theories of meaning here).
To give and elaborate on an opposing view of the question, it seems weird to me to say that one couldn't in a textbook question argue that "[a natural language description]" doesn't mean the same as "a situation that could in natural language be described as [a natural language description]". The different phrasing does give the sign a different intension and mind, and according to some, alters its meaning, but I don't see that as an issue as I'm not arguing for either interpretation, but merely for the possibility of both. Either way, referring to my earlier comment in There is no ambiguity, this is the situation with the phrase "at least one of the children is a boy", as in natural language that is a phrasing that could refer to a situation where it is known which child is that of whom the gender is known, so it feels unintuitive that it could not be translated backwards the same. There is no good explanation for why it'd have different implications than the alternate phrasing, and while its connotations are different, those connotations would be subjective, and thus that it'd feel more natural to you to use such phrase to imply that it specifically is not the case that any given child is a boy, is subjective to you which makes it ambiguous. The phrase "at least one" does not feel unambiguous to you, but that it feels that way to you makes it ambiguous. BlueBanana (talk) 00:54, 20 January 2022 (UTC)[reply]
To add to the last comment, I see how "at least one of the children is a boy" has connotations leaning towards "the amount of children whose gender is boy is known to be one or greater", and how while "a specific one of the children is known to be a boy" implies "the amount of children whose gender is boy is known to be one or greater", the opposite is not true; nevertheless, the connotation merely leans that way, and "at least one of the children is a boy" does not unambiguously mean the same (insofar as much as two statements of natural language can ever mean exactly the same thing or statements of natural language can even have unambiguous meanings to begin with). BlueBanana (talk) 01:25, 20 January 2022 (UTC)[reply]
I agree with most of what you said, probably about everything regarding the philosophical issues and philosophy of natural language. I still think that the 'unambiguous' (which yes, it is eternally debatable considering all possible philosophical views) view is still defendable considering that the 'context' originally is a mathematical puzzle. If we don't assume the 'basic common(-ground) assumptions' that give the BB, BG, GB, GG model (or one that includes that), it is also possible to treat the question differently. Once the interpretation of the whole question becomes more relaxed, there are indeed many possibilities. My point relies on the fact that all of these need to make additional assumptions or dream up a story in which assumptions are worked in to be able to make their interpretation well-defined and come to a concrete result. Meanwhile this interpretation is minimal and everyone agrees it is at least a possibility, at least a possible interpretation. I had a few more technical, logical and mathematical arguments, but I won't go into that, because if you make it a broader philosophical question there will always be someone who disagrees and there are indeed plenty of interesting branches of philosophy that take a different view. It can be debated wether the problem statement has a unique translation in an advanced intensional logic, logics that try to be more complete logics of all of 'natural language' usually can't enforce a specific translation, indeed, because the choosing of the logical model is a question that can be debated. (Very interestingly related to this, but entirely a side-note, in case it might interest you personally, is that in Eastern-Orthodox theology the models of modal logic that are usually used in much of philosophy and philosophy of science cannot be used, because within it the philosophical view is held that only god can be a cause of necessity, and so the fact that 'necessity implies necessarily necessary' as in specific modal logics is taken to be a law, cannot be allowed within a model that would be used in this theology.) So to conclude, definitely if we agree on a view such as 'meaning is use', and at least some people would, then the fact that at least some people would use, or at least interpret it in a specific way, means that this couldn't be excluded. So if we want to include or at least allow such points of view, for neutral point of view discussion, then what you have argued is correct, we must definitely still allow it to be called 'ambiguous'. I just thought that the way 'ambiguity' is stressed throughout the article was a bit much, mainly because definitely in a mathematical context 'at least one' is not (ok 'usually') taken to mean 'assume a specific one', mainly in this case this seemed problematic to me, because in the 'standard' interpretation, there is no real issue with solving the problem, while allowing different intensional readings means that these must either include additional assumptions or such assumptions must be made. But it is true that natural language can contain more information that is lost when a certain strict logical translation within a chosen logical system is chosen. It just seems to me (yes 'seems to me', although a lot of the psychological researches referenced as sources seem to be about this) that a lot of people who try such puzzles, without much of a mathematical background or acquaintance with probability theory, often make very bad assumptions in that they use methods that are not valid by the rules of probability theory and rather just work with some intuitive conceptual, vague and often ill-defined approach to the problem and interpretation of it. Though I should add that people may agree to find multiple interpretations reasonable and can agree that different interpretations can lead to different solutions and perhaps at least one of the studies, specifically the one by Fox and Levav, can be criticised from a philosophical and logical point of view in that they seem to write that 1/3, for their own specific formulation, which is not only extremely ambiguous but actually should give rise to a much more complicated analysis than the simplistic one, is the 'correct' answer and one half is incorrect. Of course this doesn't influence their psychological study, but the claim is somewhat problematic, because they have even added in 'misleading' elements on purpose that could really mess with interpretation, and so that is where your argument becomes extremely relevant. I would maintain that within the context of an abstract and idealised mathematical puzzle, the question is not truly ambiguous, but exactly the fact that this is debated makes it difficult, impossible by the principles of wikipedia to let the article truly reflect that view and I think that's probably for the better here, it should maintain a neutral point of view. The only change I would only really 'definitely' still like to see to the article then is an addition of more explicit calculations in some parts, because it really shows the probability theoretical side to it, methods that are definitely not erroneous. More specifically I would like to add in a parameterised form of the problem Bar-Hillel and Falk discuss, where 'being told' explicitly needs to be taken into account, which leads to what is currently only described in words in that section. I think showing the symbolical calculation would clarify that section, at least for some people, and show just how the calculations are influence by changing interpretations. I can add that in, I don't think it would be very difficult and then perhaps you could review it? Except of course if you already think that it would be an unnecessary, bad addition. — Preceding signed comment added by MathsWolf (talkcontribs) 02:08, 20 January 2022 (UTC)[reply]
Those are really good points, it's especially true that indeed people tend to either think that the answer is 1/3 or that there is ambiguity in whether it's 1/3 or 1/2, while it'd be difficult to argue that 1/2 is the only correct answer. In that sense 1/3 does have a special status over 1/2, so if there's a source for that point I'd like to see it added in defense of 1/3. Also though it's still a value judgement to assume minimal knowledge, a case can definitely be made for it, so while I'd prefer the article to stress the ambiguity more rather than too little, I'm not against sourced arguments for either side.
I'm usually against adding too much mathematics into articles, but here I think as well that the article would benefit from more explicitness, though I worry I might be biased as I have something of a grasp on probability theory, and keeping the article accessible to readers with minimal background knowledge should be a priority. I think WP guidelines regarding articles regarding mathematics should be consulted, but I (tentatively) support the idea. BlueBanana (talk) 10:02, 20 January 2022 (UTC)[reply]

Ambiguity doesn't refer to the given wording[edit]

The sources that call the Tuesday Boy problem ambiguous refer to the wording "one of them is a boy", the ambiguity being about whether one should assume that "one of them" is fixed before asking what sex that child is. This ambiguity is no longer present when the problem is formulated the way it is here, as "at least one of them is a boy". Someone seems to have changed the wording of the problem to eliminate ambiguities (another ambiguity being whether "one" means "exactly one" or "at least one", where the assumption "at least one" jars with the assumption that "two children" means "exactly two"), but the claim that the wording is ambiguous is still in the article, which is inconsistent and quite confusing.--2.204.226.15 (talk) 08:56, 3 August 2022 (UTC)[reply]

Which sources are you referring to exactly? I don't see many sources about the Tuesday Boy problem that are referenced in the article here except Falk's, which is good, mainly an excellent discussion, but can't possibly be represented at length and in mathematical depth in this article. It also seems to focus on making very explicit the procedure by which the probability experiment/process can be described (or what kind of probabilistic filters are applied to the information, which is itself of course relevant information), not to leave room for ambiguity towards the calculations and to make sure all relevant information is clearly available and how variations in them can change the problem and the result. It doesn't seem to mention the wording and ambiguity you refer to though. I mainly agree with what you're saying, but on the subject of interpretation I've already had a very interesting, but far too long discussion here, even if more could be said this might not be the place if it's not immediately relevant for writing the article. Perhaps we should review this a bit though, focussing on the representation of the sources if what you're saying is right. I do indeed seem to remember an older discussion here about the original wording, it might be (if I recall correctly, I may be off) that Gardner's version was considered to have a 'grammatical error' because of saying 'the other one' without necessarily clearly specifying 'one' first, that might have led to a clear reformulation here on wikipedia in which 'at least' slipped in to reduce ambiguity, which I'd say is a good wording, but as you point out since ambiguity and wording are an important part of the discussion and history of this problem, we should probably point those out a bit more clearly and explicitly, citing exactly as originally worded in each work what their formulations were. Still, since I can't see what sources you are referring to specifically now, if you could tell me which ones you mean, we could look into it better to see if we could improve it. (Although for your note on 'another ambiguity', in mathematical terminology the distinction is made for absolute clarity, but given the context interpreting it as 'exactly one' would make the problem trivial and reduce it to the basic assumptions, while in the other reading it is solvable too, where not reading 'two children' as 'exactly two' would make the problem wildly underdetermined and unsolvable without additional information or assumptions, so I'm not so sure it's an entirely reasonable reading of the problem to claim these elements are truly that ambiguous. But of course that is up for debate and precisely the claimed ambiguity problem that remains.) I'm not sure it is entirely inconsistent in the article on wikipedia as it stands, because here and there it already uses or refers to different wordings of the problem. If you mean by 'inconsistent' that not just one wording is used for the whole wikipedia article, I'm afraid that can't be avoided as the problem has known multiple variations with rewordings in various different sources and so it is relevant to the problem and the claimed ambiguity that at different times and for different variants, different words have been used with varying degrees of ambiguity. As long as at the appropriate place in the article, the corresponding version is referenced or written, this isn't inconsistent but a correct representation. MathsWolf (talk) 07:08, 4 August 2022 (UTC)[reply]

Yudkowsky's version[edit]

According to Eliezer Yudkowsky, in the correct version of this question, you ask "Is at least one a boy?"

In a malformed version, the person tells you, without prompting, "At least one is a boy." This gives a different answer. cagliost (talk) 21:38, 29 January 2024 (UTC)[reply]

"the second child"[edit]

The last sentence of the lead currently says that "[The] answer [1/2] is intuitive if the question leads the reader to believe that there are two equally likely possibilities for the sex of the second child". However, to me it is not clear what "the second child" means. The question talks about "at least one" and "both" children, not "the first" and "the second". (At least one of [the children] is a boy. What is the probability that both children are boys?) Thus, the last sentence of the lead does not seem to be properly worded in relation to the problem. Any input about the best way to fix this? —St.Nerol (talk, contribs) 10:56, 9 February 2024 (UTC)[reply]

I agree that there is reason to object to the old wording, but I don't think the new wording fixes things. To understand why the old sentence was written the way it was, one would most likely need to read references 2, 8, and 9, which is something I haven't done. If I had to guess at the meaning of the old wording, it would be something like, "The answer 1/2 is intuitive if the question leads the reader to believe that there is an identifiable child whose sex is not yet determined, with equal probabilities for the sex of that child." With the new wording I am unable to guess what is meant. Usually equal probabilities of boy and girl is an assumption that holds for children in general in both versions of the problem, so the new wording must be saying more than that. But I can't figure out what. Will Orrick (talk) 19:08, 11 February 2024 (UTC)[reply]
I have reverted the change until a proper review of the three cited sources can be done. Will Orrick (talk) 05:08, 14 February 2024 (UTC)[reply]