3 prisoner puzzle

Yes. If he's allowed ask 9998 times (being told N-2 releases). Which would be the analogy. As in the .... 1/4, 1/3, 1/2, 1 part.
Given that the warder can always name 9998 people who it is not, you are saying that by asking the question he increases his chances of dying. Are you sure that is right ?
 
Last edited:
If the scenario were finished and the main prisoner was chosen to die, do we all at least agree that at that point his odds of dying have increased to 1?

As soon as a prisoner has been chosen to die, his odds of dying have increased to 1 from the viewpoint of the person who knows this. If he is not told anything, he will still calculate his odds of dying as 1/3. The point here is that odds change depending on what you know.

Try this: toss a coin, but don't look at how it falls. The probability of it now showing heads up is 1/2. Let someone else look at the coin, but still don't look yourself. The person sees that it has landed heads, for instance. From this person's point of view the probability of the coin showing heads is 1, but from your point of view the probability is still 1/2: you still don't know if heads or tails are showing, and both these possibilities are equally likely.
 
If the scenario were finished and the main prisoner was chosen to die, do we all at least agree that at that point his odds of dying have increased to 1?

Do you still find this confusing?

Look, suppose Dan O. and I had played this game the way it's intended. So Dan picks a number between 1 and 3, randomly. If it's 1, I die, if it's 2, Harry dies, if it's 3, Dick dies.

So if we played many times, in 1/3 of them I would die - no matter what Dan says or doesn't say. Furthermore, when I ask the question, in 1/2 of all those cases Dan says Harry, and in the other 1/2 he says Dick - so his answer in any particular trial contains no information at all about my fate in that particular trial.

Make sense now?
 
Last edited:
The decision tree for the problem in the OP:
----------------------------------------
1/3 Tom will be executed
1/2 Dick will be named
1/2 Harry will be named​
1/3 Dick will be executed
1 Harry will be named​
1/3 Harry will be executed
1 Dick will be named​
----------------------------------------

The outcomes, noted {executed,named}, and their probabilities are as follows:
{Tom,Dick}|1/3 * 1/2 = |1/6
{Tom,Harry}|1/3 * 1/2 = |1/6
{Dick,Harry}|1/3 * 1 = | 1/3
{Harry,Dick}|1/3 * 1 = | 1/3

Eliminating the possibilities where Dick is not named:

{Tom,Dick}|1/6
{Tom,Harry} |1/6
{Dick,Harry} |1/3
{Harry,Dick}|1/3

Of the two that are left, Harry is executed (1/3)/(1/3+1/6) = 2/3, and Tom is executed (1/6)/(1/3+1/6) = 1/3.

Contrast this to the case where the person that is going to be named as going free does not exclude Tom as a possibility:

----------------------------------------
1/3 Tom is executed
1/2 Dick is named
1/2 Harry is named​
1/3 Dick is executed
1/2 Tom is named
1/2 Harry is named​
1/3 Harry is executed
1/2 Tom is named
1/2 Dick is named​
----------------------------------------

Making the outcomes:
{Tom,Dick}|1/3 * 1/2 = |1/6
{Tom,Harry}|1/3 * 1/2 = |1/6
{Dick,Tom}|1/3 * 1/2 = |1/6
{Dick,Harry}|1/3 * 1/2 = |1/6
{Harry,Tom}|1/3 * 1/2 = |1/6
{Harry,Dick}|1/3 * 1/2 = |1/6

Removing the outcomes where Dick is not named:

{Tom,Dick}|1/6
{Tom,Harry} |1/6
{Dick,Tom} |1/6
{Dick,Harry} |1/6
{Harry,Tom} |1/6
{Harry,Dick}|1/6

Tom is executed (1/6)/(2/6) = 1/2, Harry is executed (1/6)/(2/6) = 1/2.
 
Last edited:
Why should it be 1/2?

Well I understand (here in reference to the 10,000 extrapolation) that since asking for the N-2 release names (the specific odds reduction) is entirely pointless (considering the name can’t be him) the realistic odds of the other prisoner (or prisoner(s) who’s fate is unknown) remain what the total was (9999/10000).

Wouldn’t this make the OP more of a trick (or at least pointless)?

Five pages of pointless and still going?
 
Here's how anybody can run the experiment for themselves:

Take a large sheet of paper. Divide it into three columns. At the top of the first column, write "trial number", at the top of the second column, write "condemned" and at the top of the third write "named".

Now do your first trial: roll a die. If it turns up 1 or 2, write Tom's name under "condemned". If it turns up 3 or 4, write Dick's name. If it turns up 5 or 6, write Harry's name.

If Dick is to die, the warden only has the choice of naming Harry as the lucky one who will be spared, so in this case write Harry's name under "named". Similarly, if Harry is to die, the only person who can be named as the lucky one is Dick. If Tom is to die, you need to choose randomly between naming Dick or naming Harry. Toss a coin to do this.

Here's the beginning of my series of trials:

- trial number - | - condemned - | - named -
1|Harry|Dick
2|Tom|Harry
3|Tom|Dick
4|Harry|Dick
5|Dick|Harry
6|Dick|Harry
7|Harry|Dick
8|Dick|Harry
9|Dick|Harry
10|Dick|Harry
11|Harry|Dick
12|Tom|Dick
13|Tom|Dick
14|Dick|Harry
15|Tom|Harry
16|Harry|Dick

If you go on doing this for, say, 300 trials, you should get the following results:

In the "condemned" column, each name will appear approximately 100 times. Each prisoner has a 1/3 chance of being condemned to die. How often does the name Dick appear in the "named" column? It will appear about 150 times, since Dick is named each time Harry was condemned (around 100 times) but he is also named in around half of the cases where Tom was condemned (around 50 times). The same applies to Harry.

Now go back to the original puzzle. The warden named Dick, so just look at all the trials where this happened. There are about 150 of them. In about 100 of these cases, Harry had been condemned, and in around 50 of the cases, Tom had been condemned.

Therefore, after Dick has been named, the probabilities of being the condemned person are 100/150 for Harry and 50/150 for Dick, that is to say 2/3 for Harry and 1/3 for Dick.

If you're not convinced, try the experiment!
 
Well I understand (here in reference to the 10,000 extrapolation) that since asking for the N-2 release names (the specific odds reduction) is entirely pointless (considering the name can’t be him) the realistic odds of the other prisoner (or prisoner(s) who’s fate is unknown) remain what the total was (9999/10000).

Wouldn’t this make the OP more of a trick (or at least pointless)?

It is a trick, but not a pointless one. The trick is that the information the warden gives Tom doesn't tell Tom anything about his own fate.

Five pages of pointless and still going?

If it was obvious, it wouldn't have lasted five pages. If it isn't obvious, it's worth discussing for some.

No one's holding a gun to your head.
 
No one's holding a gun to your head.

How can you be sure?

Well, we've come to agreement twice so far in this thread, maybe I should prepare backing away from my previous post for one more go at it!

The third time is the charm. :eek:

.....perhaps I should give it a page or two first.
 
It is a trick, but not a pointless one. The trick is that the information the warden gives Tom doesn't tell Tom anything about his own fate.
Exactly. And as I pointed out earlier, Tom now knows that Dick has a 0% chance of dying, but Harry has a 2/3 chance of dying. Tom, having not learned anything about his own fate, is still 1/3.
 
I suppose your point is that the OP didn't specify clearly that initially, each prisoner has an equal chance of being the one?

That's pretty lame.

As I said several times in this thread, the gameshow version of this puzzle is much better posed (to avoid silly possibilities like the one I guess you're exploiting). So, my $1 is at your mercy.

I guess everyone knows by now that I am not a gambler. You may send your dollar to the forum as I haul Tom off to the Gallows (it's already morning somewhere in the world). You could also renege on the bet. Maybe the forum management will pitty you and only suspend you for a day or two.

As Lucky pointed out, I already knew who would be executed and if you go back further you will see it sort of slipped out when I told you:
Tom will be executed. You just have to pay the $1.

And just to make sure you didn't weasel out by playing Dick or Harry, I added:
Sol will play Tom first.


I don't suppose anyone else wants to play now. But I did already offer to let you play again. If you want to try for revenge, I'll play again and even add the suggestions of the other forum members...

If you plan to simply pick Tom as the prisoner to be executed every time, or even a majority of the time, then your opponent in this game gets to pick which person will be the one to ask the warden the question.

However, the scenario is very clearly that the selection has been made before Tom questions the warder.

Incidentally, if you really want to do this we obviously need some kind of system that verifies you are choosing who's going to die randomly with 1/3 odds for each (before the question). For example we can pick a trusted third forum member who generates a list of numbers, each being 1, 2, or 3. Or we can find a pseudorandom number generator online and mutually agree on a seed.


So as CurtC suggests, the other player gets to pick which prisoner they will play and as Lucky says, the choice will be made before the prisoner questions the warden on who will be spared.

To satisfy your request, I will provide a commitment that can be verified and cannot be forged by any of us at least so you will know I am not making up the rules as we play. The commitment will be in the form of an MD5 checksum generated like this:

Code:
md5 -s "this is how the victim is choosen"
MD5 ("this is how the victim is choosen") = 2200c3d6145a1eba2d92c12b79b83166

Before the game, I will post the checksum with the message blocked out. After the game I will post the full message so the checksum can be verified. Someone with UNIX should verify the checksum above to prove that it can be verified so the message used in the game cannot be changed once the game starts.
 
Last edited:
Dan, you are definitely not a gambler. My advice: No matter what, don't play for money.
 
Someone with UNIX should verify the checksum above to prove that it can be verified so the message used in the game cannot be changed once the game starts.
I haven't followed the discussion, so I don't know what game you're talking about, but I can at least do this:

(using Cygwin and md5sum on Windows XP)

$ printf "this is how the victim is choosen" | md5sum
2200c3d6145a1eba2d92c12b79b83166 *-

There's no need for Unix: http://en.wikipedia.org/wiki/Md5sum#Windows.

Hm, it would be weird if there's no web based md5 tool. Let's see if I can find one...

http://www.fileformat.info/tool/hash.htm.
 
About the problem mentioned in the OP... This may have been mentioned already. I'm too lazy to read all the posts.

I think the easiest way to find the correct solution is to realize that this problem is equivalent to the original Monty Hall problem, and then consider what happens if you have the contestant (in the Monty Hall problem) play the game many times:

Suppose that before he plays the game the first time, he decides to go with his first choice every time. If the probability that his first choice was correct changes from 1/3 to 1/2 when the host opens one of the other doors, then he will win 1/2 of the times he plays the game. But that implies that his first choice is correct 1/2 of the times, and that's obviously impossible.
 
I guess everyone knows by now that I am not a gambler.

If you play a fair game, you will lose money.

How many times would you be willing to play the following game:

1. You randomly determine the sequence of names for the execution here:
http://www.random.org/integers/?mode=advanced

You will create a list of 10 integers as you're willing to play rounds. You will use a persistent identifier that you will announce in the forum encoded with a hash as previously discussed.

2. We will play the numbers from top to bottom if they are displayed in a single row. For each game with a 1, Tom will be killed. For each game with a 2, Dick will be killed. For each game with a 3, Harry will be killed.

3. Your opponent will start each round by announcing who he plays as. He will then ask you to give the name of one of the other two characters, provided that character will not be killed this round.

4. If the player is not to be killed this round, you will only have one name left that you can give. Otherwise you can chose freely between the two names and give whichever one you prefer. (It would be nice if you could do so randomly, but it is not required.)

5. You bet the player 1$ that his name is on the list for that day. If he is to be killed, he pays you one dollar. If he lives that day, you will pay him one dollar.

6. After the final round, you publish your codeword so that the random number sequence can be controlled.

7. Both sides can appeal the result with a poll here in the forums. If a 2/3 majority agrees that one side has cheated or played unfair, the game is void.

8. If you agreed to play more than 10 games, start over at step 1.

9. All winnings will be paid out to the JREF rather than the individual players.
 
Suppose that before he plays the game the first time, he decides to go with his first choice every time. If the probability that his first choice was correct changes from 1/3 to 1/2 when the host opens one of the other doors, then he will win 1/2 of the times he plays the game. But that implies that his first choice is correct 1/2 of the times, and that's obviously impossible.

This is not the same problem. The prisoner doesn't get to choose his fate - it's already decided.
 
If you play a fair game, you will lose money.

If a game is fair, neither side looses more than the other. I choose to play a game to demonstrate the point that different information was available to the different characters. In particular, there was no information withheld from the warden so the probability of any prisoner being executed was either 0 or 1 from the wardens perspective.

If you go back to the scenario in the OP but instead of just the one prisoner questioning the warden, all three separately question the warden and don't communicate with each other. In all cases, we can look at the information that all three prisoners have and conclude which prisoner will be executed but each prisoner will still only know that they have a 1/3 chance of being executed.
 
If a game is fair, neither side looses more than the other.

I wouldn't define "fair" that way. I was thinking along hte lines of "playing by the rules as a reasonable person would understand them, as opposed to relying on semantical loopholes."

I choose to play a game to demonstrate the point that different information was available to the different characters. In particular, there was no information withheld from the warden so the probability of any prisoner being executed was either 0 or 1 from the wardens perspective.

We all knew that.

If you go back to the scenario in the OP but instead of just the one prisoner questioning the warden, all three separately question the warden and don't communicate with each other. In all cases, we can look at the information that all three prisoners have and conclude which prisoner will be executed but each prisoner will still only know that they have a 1/3 chance of being executed.


So?
 
This is not the same problem. The prisoner doesn't get to choose his fate - it's already decided.
The problems are only superficially different. The hero in this problem can pretend that each prisoner is a door in the Monty Hall problemWP and that he himself is the door that was chosen first. And when the warden gives him the name of a prisoner that will not be executed, he can pretend that that prisoner is the door that was opened to reveal a goat. Now he can guess who will be executed (which ironically corresponds to winning the car). Is it him, or the other guy? If the probability that it's him is P, then the probability that his "first choice" was correct is also P, but that's impossible unless P=1/3.
 
Last edited:
I wouldn't define "fair" that way. I was thinking along hte lines of "playing by the rules as a reasonable person would understand them, as opposed to relying on semantical loopholes."

So when you talk about a fair coin you are thinking along the lines of abiding by the rules of physics and not necessarily that the coin has an equal probability of coming up heads or tails.
 

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom