Originally posted by Palynka
Yeah, I know what you mean... The problem is that in the WCS you always get the same answer ('He is not a truth-teller'😉 from anyone you ask (which will always be liars) about any of the people you ask about (which will always be liars).
We could always pose a second question and ask what is the best strategy for both parties if there is a game between as nd an optimal strategy for the asker that minimizes the EXPECTED number of questions asked?
A probabilistic approach...
If you have lost interest in this, I understand. This is about my last posting on this topic, because delving much further will require tedious math work. But after thinking about it I want to include this idea, because it will mean I can't worry it further. 🙂
I am thinking about a strategy which is immune to the values of P_L and P_T, I think.
Number the people 1 to 100, and pair them off 1 and 2, 3 and 4, etc, on paper. You don't have to tell them they are paired or about the pairing scheme. Ask the odd numbered people about the even numbered; 1 about 2, 3 about 4, etc. Don't ask 2 about 1, 4 about 3, etc. just yet.
If, in any pair, the person you ask calls the other a liar, cross out BOTH of that pair, don't pair them with anyone again, and don't ask either of them from now on. They are both done for the day. They don't have to be told this -- you just won't ask either of them again.
For any pair crossed off, you have avoided asking the next question, which is, what the even-numbered person has to say about the odd-numbered person. Do this with the survivors, and repeat the crossing-off of any pair where the odd-numbered person is called a liar.
This seems counterintuitive because you are crossing off truthers, but you have two extra of them. Two truthers have to be paired up with each other, and so will survive the round. So will any additional paired-up truthers -- and so will any paired up liars, but only if they both lie about each other and call each other truthers.
Now, before starting the next round, redo the pairing systematically.* Renumber reductively (eliminating gaps) and pair 1 with 4, 2 with 5, and so on. Basically you rotate the pairing wheel two spaces from the previous, after renumbering. This process will certainly, at least eventually, separate some previously paired liars and put them with truthers. (I don't know if this can be quantitatied or estimated easily). Those liars are doomed (along with their paired truthers). You still have two extra truthers that will be paired with one another -- they might be different truthers this time -- so at least one pair of truthers will survive, again.
You are using the truthers to 'out' the liars, and you have truthers to spare. Getting rid of truthers you no longer need, also reduces the number of questions you need to ask in future rounds.
Because you are always eliminating pairs, you always have an even number to pair up. In principal on average, in each round this will cut the number of people left, about in half. Of course it could in principle take longer, if it were possible that all the truthers get paired to each other and all the liars get paired to each other, and the liars lie about each other. But the pairing system -- practically, if not easily provably, -- prevents that.
There is always a surplus of at least two truthers. When it gets down to, say, 4 and 2, or 6 and 2, the 2 liars will soon enough be paired with truthers and will be gone, and then the truthers will confirm each other as such by pairing each with each of the others in a final round or so.
Other than simulation, I would not know how how to compare this to other strategies that test various settings of P_L and P_T including 100 and 0%. But it has elements that actively trim the number of future questions needed, even during a round.
OK that's it. Like I said, I'm not about to do the math. It satisfies my desire that the information gained be more useful.
* The renumbering might preferably be random, followed by a fresh 1 and 2, 3 and 4 pairing etc. A refinement to the pairing system could be that if any two people are re-paired, skip questioning them in the round, then re-randomize.