PDA

View Full Version : 33 Questions?



wiggy
2013-May-25, 09:24 PM
I've had this idea running through my head for ages now.
2^33 is close to 9 billion.

What 33 yes/no questions, ie a 33 binary code could uniquely identify everyone?
I'm guessing 5 or 6 would be geographically related.
2 or 3 could be religious
1 gender
3 season of birth

But I've always wondered if the last question would be like, "is that you george?"

KaiYeves
2013-May-25, 10:07 PM
There was some insurance company that had print ads like this, listing several descriptions and the number of their clients each one applied to until it was narrowed down to one guy. (Number of our customers retiring this year... At the age of 69... From a career in medicine... In the state of Wyoming... And will now bond with their three grandchildren... Aboard their new sailboat... In Tinidad.)

Jens
2013-May-26, 10:36 PM
It's an interesting question. The problem is, you basically need 33 questions that will nearly cut the sample in two. You don't have much leeway in that regard. I can think of a way to do it theoretically, though in practice it would be very difficult, because you would need to know the precise distribution of the human population. The first question would be something like, are you north of say 10 degrees north, and the last question would be something like, are you the one sitting in the chair to the right of the window?

Delvo
2013-May-27, 03:26 AM
You'd need to allow some of the questions to be variable, based on the answers to the previous questions. Otherwise the problem is that most questions don't neatly split the population in half; they'd split it by other ratios like 60/40 or 95/5 in at least some cases, and the fact that some people would fall into the majority on one question after another means it would take more questions than you planned on to include all of them.

Jens
2013-May-27, 04:06 AM
I guess there could be ways to do it other than my first proposal, but they'd be very complicated as well. For example, you could divide the countries of the world into two arbitrary groups arranged so you get 50% of population on each side. So one questions could be, were you born in one of the following countries, and just list off 60 names. And then, of course, are you male would get you a pretty nice split. Then you could find the average height of men and women in the world, and find out if the person is taller or shorter, etc.

wiggy
2013-May-27, 09:37 AM
well I thought
were you born in china
were you born in india
were you born in Indonesia
were you born in pakistan
are you only female (hermaphrodite answer no)
are you blood type A

Its like 20 questions but for the entire population.

But there'd be weird questions you wouldn't think of straight off like "have you ever eaten mango?"

caveman1917
2013-May-27, 11:17 AM
You'd need to allow some of the questions to be variable, based on the answers to the previous questions. Otherwise the problem is that most questions don't neatly split the population in half; they'd split it by other ratios like 60/40 or 95/5 in at least some cases, and the fact that some people would fall into the majority on one question after another means it would take more questions than you planned on to include all of them.

This is probably going to be the main problem here, it's easily demonstrated by an example where say people are distributed as the following matrix
7 3
3 7

Each of the questions "which column are you in" and "which row are you in" splits the sample in half yet there exist dependencies between those questions which means that asking both questions doesn't cut the sample in four. Technically the marginal distribution of each question may be 50-50 but there exist statistical depencies between them. As you go down the binary tree to the bottom we can expect these dependencies to grow - you're looking at continuously smaller and diverse sets of people, yet the number of questions that you need to ask at each level also grows exponentially as you go down that tree. It may well be that you need thousands, if not millions, of questions all depending on the answers given previously to make this work.

caveman1917
2013-May-27, 11:29 AM
It would work if you had a means to apply a total order to the set of people before you started, say ordering people by moment of birth. Then you can give everyone a number as in 12 means you are the 12th oldest person alive. After that it's simply an exercise in looking at their numbers in base 2 and the n-th question would be "is the n-th digit in the binary representation of your assigned number '1'?".

HenrikOlsen
2013-May-27, 11:52 AM
This is probably going to be the main problem here, it's easily demonstrated by an example where say people are distributed as the following matrix
7 3
3 7

Each of the questions "which column are you in" and "which row are you in" splits the sample in half yet there exist dependencies between those questions which means that asking both questions doesn't cut the sample in four. Technically the marginal distribution of each question may be 50-50 but there exist statistical depencies between them. As you go down the binary tree to the bottom we can expect these dependencies to grow - you're looking at continuously smaller and diverse sets of people, yet the number of questions that you need to ask at each level also grows exponentially as you go down that tree. It may well be that you need thousands, if not millions, of questions all depending on the answers given previously to make this work.
That dependency only matters if you're forced to ask the same question at each level regardless of previous answers.
If the question at each level is tailored to the set of people still being considered, then the vastness of question-space virtually guarantees that a near 50-50 split can be achieved with the next question, limiting the number of questions that have to be answered to find any specific person to 33.

If you design a comprehensive questionnaire, then yes, that will require millions, if not billions of questions, but it's only 33 of those that need to be asked for each person.

caveman1917
2013-May-27, 12:14 PM
That dependency only matters if you're forced to ask the same question at each level regardless of previous answers.

I may have misunderstood but i thought that was what the OP was asking and what was being talked about, a unique set of 33 questions that could identify everyone. At least that's how i was considering the problem statement.

HenrikOlsen
2013-May-27, 12:23 PM
I may have misunderstood but i thought that was what the OP was asking and what was being talked about, a unique set of 33 questions that could identify everyone. At least that's how i was considering the problem statement.

I thought
But I've always wondered if the last question would be like, "is that you george?"But I've always wondered if the last question would be like, "is that you george?"
excluded that interpretation.

caveman1917
2013-May-27, 12:45 PM
I thought

But I've always wondered if the last question would be like, "is that you george?"
excluded that interpretation.

The other interpretations were excluded more strongly.

We first had

What 33 yes/no questions, ie a 33 binary code could uniquely identify everyone?

That excludes the interpretation that the OP was asking whether we need to ask more than 33 questions. Given that he is asking about the questions themselves, not about their number, suggests that he thinks it's possible to list 33 questions that would uniquely identify everyone. We can hardly conclude that he's asking us to list the complete questionnaire consisting of millions of questions.

Then we had


But I've always wondered if the last question would be like, "is that you george?"

Which contradicts the above interpretation and makes it seem that no interpretation is consistent with the entire OP. However he stated "i've always wondered if..." so i took the interpretation that the OP is asking whether there exists a set of 33 questions that could uniquely identify everyone or whether (as he is wondering) some questions will have to be individual. Which seemed, to me, the only interpretation that is consistent with the entire OP. As if the OP had stated "What 33 questions could uniquely identify everyone or would there need to be questions that are tailored individually?". And thus that is the question i answered - and then gave an example where it is possible to have a unique set that can identify everyone by imposing a total order on the set of people.

But it's perfectly possible that i completely missed out on something, what interpretation would you give that is consistent with every statement in the OP?

Jens
2013-May-27, 01:44 PM
I think the disagreement between HenrikOlsen and caveman1917 is the result of the OP not being clear. Are you thinking of a set questionnaire, or could questions be made up on the spot in response to answers to previous questions?

HenrikOlsen
2013-May-27, 02:37 PM
I actually think we're in near-complete agreement, except over which of two possible questions was asked.
We even agree on what the answers are to the two questions.

Delvo
2013-May-27, 07:35 PM
OK, so the first question in this person-identifying process should be about which way the subject would interpret the question in the original post here...

Jens
2013-May-28, 04:00 AM
I actually think we're in near-complete agreement, except over which of two possible questions was asked.
We even agree on what the answers are to the two questions.

I agree with that too, and what I meant was that you were in disagreement over what question was being asked. I think I also agree that if you had to have a set questionnaire from the beginning, it would be impossible except in very unusual circumstances. For example, if you lined up all the people of the world and gave each one a ticket with a number on it, and then started asking questions like "is your number greater than n/2"? "Is your number greater than n/2 for the set made up of all the possible numbers that would fit your answer to number 1?" etc. etc. While if one is allowed to make up questions as one goes, it would be possible but still would require some ingenuity. Right?

caveman1917
2013-May-28, 08:18 AM
I agree with that too, and what I meant was that you were in disagreement over what question was being asked. I think I also agree that if you had to have a set questionnaire from the beginning, it would be impossible except in very unusual circumstances. For example, if you lined up all the people of the world and gave each one a ticket with a number on it, and then started asking questions like "is your number greater than n/2"? "Is your number greater than n/2 for the set made up of all the possible numbers that would fit your answer to number 1?" etc. etc. While if one is allowed to make up questions as one goes, it would be possible but still would require some ingenuity. Right?

That is exactly right. You can state that unusual circumstance as "if it is possible to index the set of people". The disagreement over which question was asked isn't so important, since every interpretation has been covered and answered anyway.

wiggy
2013-May-28, 09:36 AM
the same 33 questions for everyone

33 bit code for each person.

caveman1917
2013-May-28, 11:02 AM
the same 33 questions for everyone

33 bit code for each person.

Extremely unlikely to be possible unless you either first order everyone in a list or let the questions depend on the subset of people you're considering. For example you could do something like instead of "is your height above 1.7m?" the question would be "is your height above the median height of the group of people i am now considering?". Though you could argue whether you consider questions of that latter type as "the same for everyone", if you do allow such questions then you can just ask the question "is your height above the median height of the group of people i am now considering?" 33 times.

Solfe
2013-May-28, 12:42 PM
This reminds me of a toy I got to play with. It would identify book characters, concepts or items in less than 20 questions. Each device was themed, say Harry Potter or some other series.

ToSeek
2013-May-28, 06:54 PM
This reminds me of a toy I got to play with. It would identify book characters, concepts or items in less than 20 questions. Each device was themed, say Harry Potter or some other series.

There's a website for that - http://en.akinator.com/ . (I also have the iPad app.)