Probability Riddle--are you a "halfer" or a "thirder"?
maproom.co.uk
maproom.co.uk
1/2 is the answer to "What is the probability that the coin landed Heads?"
1/3 is the answer to "What is the average probability of you getting it right if you say the coin landed Heads?"
So I guess that makes me a "Can you rephrase the question?-er"
This problem seems to be a good thought experiment demonstrating that those two questions are not equivalent. Which only makes one wonder what "probability" really means...
If your goal is to maximize the number of correct guesses across all questionings, then always answer 1/3.
If your goal is to guess the current coin flip as accurately as possible, then always answer 1/2. (I interpret the question as this.)
If the coin comes up tails, you will be asked 1,000,000 times in a row. If the coin comes up heads, you have a 1/1,000,000 chance of being asked.
I wake you up. If you had to guess if the coin was heads or tails, what would you say?
The events in question are not independent. You have a 50% chance of being woken once, and a 50% chance of being woken 1,000,000 times.
Therefore the correct answer is 50%.
Sadly, the problem doesn't define that.
But, 1/2 of the time she will be awoken once, and 1/2 she will be awoken 1,000,000 times. So, let's suppose she is awoken. There are 1,000,001 possible awakenings. The probability of her being in the first of these is 1/2 (Heads). The probability of her being in any particular other is 1/2*(1/1,000,000) (Tails).
Hence, she should answer 1/2 if she wants to guess the current coin flip correctly, since the chance of the coin being Heads on any arbitrary awakening is 1/2.
Then ask yourself, how afraid of dying would you be?
(In particular, the reason why the probabilities don't work out that way is that on any given awakening, you have a 1 in 1,000,001 chance of it being heads, and a 1,000,000 in 1,000,001 chance of it being tails).
If these previous arguments are not convincing, consider this (equivalent) alternative: what if you were woken up once on tails, but only 1 in 1,000,000 times on heads? What would you give as your credence that the coin came up tails then?
It doesn't make a different if you're only doing it once or a million times. The whole point of probability is to figure out what would happen if you did it a million times so that you use the optimal strategy when it happens only once.
If this experiment is repeated many times, "Tails" will be the right answer twice as often.
You can try this with a friend. As someone pointed out later on in the page, you just answer the first question freely and if it's tails are required to answer the second question the same way as you did the first. Or just answer heads every time--no memory wiping required. If it's really 50/50 then answering heads should get you your answer.
Because the interviewer asks questions in an unbalanced manner, the interviewee actually gets more information just by being asked.
http://en.wikipedia.org/wiki/Two_envelopes_problem#A_non-pro...
I think this version makes it clearer: if it's heads she'll be interviewed once, if it's tails she'll be interviewed thousands of times. At any given interview she should assign a very small probability to the possibility of heads.
1/2 1/2
/ \ |
Mon Tue Mon
1/2 1/2 1
T T H
I think I have identified the thirder problem. They are assuming that it's Monday: Mon Tue
3/4 1/4
/ \ |
2/3 1/3 T
| |
H TTry doing the experiment several times and see how well you do picking heads every time you "wake up." There's no need to have memory alteration if you set your strategy beforehand. You'll only be right 1/3 of the time if you bet on heads.
The answer is that at any point in time, absent other information, the coin is equally likely to be heads or tails, but as soon as Beauty receives the new information that she's being interviewed, it changes the known information, making it 1/3 instead of 1/2.
Deleted comment
Deleted comment
We're reading the second green one on that page.
So it's thirds: There are two way to be woken up for tails, and one way for heads.