The king wants to test how well the court torturer can conduct interrogations with torture.

The king has 30 criminals sentenced to death. What experiment can a king perform to test that the torturer can get the right information from people who are very interested in telling a lie?

It's not only about whether torturer is able to force a person to tell something, it's more about a problem with "telling anything to stop a torture", so a test is needed to show that people would not just tell exactly what the torturer wanted to hear, aka false positives.

  • $\begingroup$ Comments are not for extended discussion; this conversation has been moved to chat. $\endgroup$
    – L.Dutch
    Commented Nov 15, 2022 at 15:57

5 Answers 5


This works pretty much the same as any other experiment:

If your setting takes place in a world like the 1600s or later, your researches will use some variation of the modern scientific method to conduct this experiment. If your setting is more primitive than that, then the question of if torture works is more likely be discussed philosophically, and conclusions drawn based purely on theory, or they will study observations made in a poorly controlled way by just looking at the outcomes of recorded torturing. If your experiment is done at any point after the medieval period, there are a number of controls you might expect to see depending on the exact era your setting is in.

Control Group

If your experiment is done after the mid-1700s, then you will probably see a control group used.

It is not enough to torture people for facts, you need to compare the results to people who are not being tortured to determine its effectiveness. You need to split your death row inmates into 2 groups, those you will torture, and those you will interrogate without torture. If you only study 30 convicts, and get 15 true confessions, then an unscrupulous researcher might make the claim that torture has a 50% success rate. But if you also interrogated 30 convicts and got 20 true confessions without torture, then you would prove that torture actually has a negative outcome.

Verifiable Test

Verifiability has been a part of pretty much every variation of the scientific method dating all the way back to the ancient world.

Interrogating people for information that you can't prove is not a valid experiment. You need to to create interrogation questions that can be proven. For example. You can't interrogate people to find out who thier accomplices are because this usually can not be verified. Instead you need to interrogate them for information that you already know about them.

Confounding Variables

Since the early 1900s, confounding variables have been an important consideration in the scientific method.

This is the hardest part. You need to interrogate each prisoner for as similar information as possible, but if your exact question only applies to a small portion of the population, then that group of people might respond differently to torture than people in general which can skew your results. The questions also need to be something that most people will not want to tell you, but they usually know the answer to, and can be confirmed. For purposes of this experiment, it should also be something that they don't know that you can confirm because if they know that you know, then they might be less willing to lie.

I would find the closest relation to the person (wife, child, parent, etc.) when arrested, and try get them to disclose information about that person like thier address or place of work. Next you create a reason for them to not want to tell you this information. Tell them that you found out that they are involved in a crime and you need information to find them. This should constitute a "normal" level of loyalty you would need to overcome with torture. While some differences may emerge between people interviewed for wives VS children VS parents a good random distribution should account for this, but if you are only asking about people's mothers, then you may anciently skew your data based on how people who have living mothers and dead or abandoning mothers treat loyalty, since you'd have to filter out all convicts who have dead or abandoning mothers from the experiment giving you a non-random sample population.

Sample Size

If your setting is like the late 1800s or later, the researchers will likely take level of certainty into account. Before then, it was not uncommon for studies to include problematically small sample groups like 30. But if you do this, there is a good chance the outcome of the study will prove that torture is effective even though it is not.

In these experiments, some people will role over with out caring about the person you've selected, and some people may not actually know the truth, and some people will come from different backgrounds that introduce extraneous variables that you cant possibly account for. This means that a larger test group is going to be very important. With 15 people, the control group could have a vastly different number of people who don't know or don't care or have different core values, but with 400 people each, these outliers should cancel each other out.

The smaller the number of people you test, the more likely it is that randomness will create a false conclusion. 15 test subjects and 15 control subjects is simply way too small to be able to trust the outcome of the experiment. At only 15 test subjects, there is a 25.82% chance that any outcome you derive will simply be wrong. The preferred level of confidence for most experiments is considered to be at least 95%. To achieve this level of confidence (without any previous research to adjust for your proportion probability), you need a test group of about 400 people. This means you should be using a closer to 800 total convicts in your experiment if possible, not 30.

Using a spread of different torturers and interrogators may also be important to since your study could be skewed by the competency of these people. You can probably get away with a slightly smaller number of these than convicts if your kingdom has a limited number of "detectives" to determine thier population size. For example, if your kingdom only has 200 total detectives, then using a different interrogator for each prisoner is not necessary to get a good cross-section of interrogators. Involving 133 and interrogators, and giving them each 6 people to interrogate will give you about a 95% confidence level that your interrogators are representative of thier population.

Sample Size Calculator


This was not well understood before the 21st Century; so, chances are good that your king will not account for it even if this study were done today.

If you know about P-Hacking, you've probably noticed a flaw in the methodology here. If 133 interrogators gives a 95% certainty and 800 convicts gives a 95% certainty, then I'm actually testing for two things instead of 1 dropping the total degree of certainty for the experiment down to 90.025%. This fake since of certainty is called P-Hacking. Since you can't really do this experiment without testing both different convicts and different interrogators, you need to increase one or both of your pools to reach 95% certainty. The easiest thing to do is invite all 200 interrogators in your kingdom to be part of the study to interrogate 4 people each, then you have 100% representation among interrogators, and 95% among prisoners giving you a 95% certainty. Or you could boost the certainty to about 97.5% for both groups which would take 178 interrogators and 3200 convicts.

That said, this most likely this will not be accounted for unless your setting is in the current day. Otherwise, the experimenters will likely settle for a 90.025% certainty while claiming that they are 95% certain in the results.

It gets even worse if you compound P-Hacking issues with a small sample sizes. If you're only testing 30 prisoners, certainty level actually drops to somewhere between 18% (used a single interrogator for all tests) and 62% (used different interrogators for all tests). While it is counter intuitive that you can be less than 50% certain for a true/false problem, P is not a measure of the outcome of your experiment, but how certain you are that the outcome demonstrates a relationship. If you only test something 1 time, you have a 50% chance of being right, but a 0% chance of proving a trend.

  • 2
    $\begingroup$ It's likely the safest answer, but oh doggy how hard it will be to make the tests verifiable without interacting with the test's value : If you tell the criminals a secret "for the experiment", resisting torture will be futile, as there's no value in resisting telling something your questionner already knows. It'll lead to a near 100% success rate, vs potentially less for interrogation if your criminals actually want to have a king believing too much its torturer "investigations". $\endgroup$ Commented Nov 14, 2022 at 16:56
  • 1
    $\begingroup$ @Tortliena I think not interfering with the values can be adequately done because the prisoners are isolated from society. You can investigate thier relationships without them knowing about it. You can also make sure that the interrogators are not the same people as the researchers so that they don't know the true answers... or even that this is a test where the information is already known. The interrogators may just be doing thier jobs as usual when you come in with a convict for them to get information out of. $\endgroup$
    – Nosajimiki
    Commented Nov 14, 2022 at 17:32
  • 3
    $\begingroup$ You can tell the condemned a secret, make sure that they remember it, then tell them that if they can keep the secret from the interrogator that will be sent to question them for a certain amount of time, their sentence of death will be commuted. That gives them a good reason to want to keep the secret from the interrogator, especially if they see other experimental subjects being released from death row. $\endgroup$
    – Monty Wild
    Commented Nov 14, 2022 at 22:58
  • $\begingroup$ @MontyWild That sounds good in theory... But how much a person would resist torture to save thier own hide may not be at all the same as how much they resist torture to protect others. Since the goal of torture is typically trying to get a person to betray others, you need a scenario that represents this. $\endgroup$
    – Nosajimiki
    Commented Nov 15, 2022 at 14:21

It is generally known that torture 'does not work.' Here a fictional king can conduct a controlled experiment which should show that torture 'does not work' all that often, unless the author wants to make the character improbably perceptive.

In this instance, the king is not interested in 'finding the truth' of anything the 30 victims may know. The king wants to test the ability of the torturer to overcome their resistance, with is the opposite of testing the ability of the victims to resist torture by this specific torturer. So the king might:

  • Identify those of the 30 victims who have genuine affection for some other person. A parent, a child, a spouse or partner, who exactly does not matter.
  • Inform half of those victims that they personally will be executed after torture, no matter what, and that their loved ones will be executed if any only if they reveal a certain codeword. If the victims keep the codeword secret, their loved ones will be released with a small reward for their inconvenience. The second part does not have to be true, but it needs to be told in a convincing way. A king who employs torturers can presumably make credible threats in this regard.
  • For the other half of the sample, make the same promise, just without the codeword. Those have to keep quiet to save their loved ones. (They presumably won't, but this tests if made-up confessions are detected.)
  • Tell the torturer to extract the codeword from each victim. The torturer may or may not be told that there are codeword-less victims.

The torturer gets a positive mark for each successfully extracted codeword and for each case where he or she correctly finds that there was no codeword. There is a negative mark for each case where a wrong or non-existing codeword is extracted. One negative mark counts for more than one positive mark in the evaluation.

The expected outcome should be that just about all victims provide a codeword. A few might keep quiet until they die. A very few might convince the torturer that there is no codeword, especially if the torturer was told of that option. The king will judge the torturer on the success rate.

This 'simulates' the case where a fanatic wants to keep a secret, mixed in with fanatics who do not know the secret. It is not quite the same as a case where a criminal believes that confession leads to a sentence and silence leads to accquittal, but then, your fictional king can't have everything ...

  • 1
    $\begingroup$ This sounds like a good way to train interrogators to be more effective by sluicing out those who know and don't know the truth, but does not establish a control to determine if non-torture interrogation is more or less effective than torture. $\endgroup$
    – Nosajimiki
    Commented Nov 14, 2022 at 16:31
  • $\begingroup$ @Nosajimiki, I didn't think the comparison with other methods was part of the question. Just the detection of false positives as the victims babble to make it stop. $\endgroup$
    – o.m.
    Commented Nov 14, 2022 at 16:34
  • $\begingroup$ Of course, there are very rarely ways to use pain, whether or not it is considered torture, to extract the necessary information giantitp.com/comics/oots0327.html $\endgroup$ Commented Nov 15, 2022 at 2:00

This answer can be based on real-world data... there may be no actual dragons on planet Earth, but there have been countless torturers, tortured, and leaders (who sometimes condone torture).

Summary: It doesn't work.

Occasionally, some captives tell the truth. Occasionally some tell lies. You're able to occasionally verify these truths and falsify these lies... but not from any response or reaction of the tortured.

Even without torture, interrogation (say, of the law enforcement variety) is problematic. Some people simply feel guilty and therefor act guilty, even when perfectly innocent. Many are neurotic, after all. They'll sweat, fidget, their voice will break, they'll have trouble making eye contact, etc. In their heads, they might even wonder if somehow they are guilty and either don't remember it or don't understand it. For that matter, even if someone would assure them that that's not what is being discussed they would still do all of this, they are unable to control themselves and will still exhibit these behaviors. Of course, that doesn't happen as interrogations are usually adversarial.

Torture cranks all of this up to 11. In extreme pain (let's say it, it's really agony), people are no longer thinking rationally. When they're telling their lie/truth, the torture either stops or at least slows down if only so they can choke out the words. And for that moment in time, lies serve just as well as truths.

For obvious reasons, it's impossible to test any of this ethically. That may or may not be something your king is worried about. But beyond the ethics issues, experimental design here is difficult, costly, and slow. He will need to carefully screen "applicants" for his experiments, come up with many of them from many different backgrounds, and subject them to torture. Since he will already know the correct answers, seeing if the torturer's answers match his own, he'll be able to show statistically that the torturer's methods work to within some acceptable margin or not.

This may take years or decades, it will result in thousands of deaths and permanent injuries (PTSD even if nothing physical). He has to find researchers that will remain unbiased somehow despite witnessing this (while weeding out the sort of sociopaths and sadists that will enjoy watching... as they might steer the study in a way that invalidates its results). He has to somehow keep this semi-secret because if "applicants" become aware of the study they will almost certainly strive to avoid "applying".

Of course, all of this will be specific to the torturer... hypothetically another torturer with a different methodology or even just a different personality may get different results. It would take many such studies with many torturers to come up with more general results.

And all of this to determine that torture doesn't work. Supposing he's willing to accept those results. This is going to be one messed up dystopian novel, trotzt.


Check to see if it pans out.

Unfortunately, you haven't provided us with any information about what kind of data the king wants; just that it's data. Therefore, I can't really give you any specialized tests.

That said, verifying information is almost always much easier than inferring it from scratch. For example, say you have three people: Alice, Bob, and Chad. Chad wants to know what Bob's doing. There are two ways he could go about figuring this out.

  • On the one hand, he can just guess as to Bob's intentions. This isn't probably going to be very accurate, since he is more or less unfamiliar with Bob's plans.

  • Alternatively, he can beat Alice into a bloody pulp, "convincing" her to tell him what Bob is doing. This data may or may not be accurate. However, he can use this (possibly inaccurate) data as a basis for further investigation; he now knows where to look. If it turns out that Alice lied to him, he can always repeat the process with Dave (poor Dave!)

See if the stories agree.

The king keeps the prisoners in separate cells, completely preventing them from communicating with each other both before, during, and after they are tortured. Once the torturer has extracted confessions from all the prisoners, he sees to what extent they agree with each other. If they all substantially disagree, chances are that they're all lying. If a good portion agree, chances are that the majority are telling the truth. If they all agree, either the torturer is exceptionally good at his job, or the prisoners came up with a story before they were captured.

  • 1
    $\begingroup$ Would the person who downvoted my answer please explain why he did so? $\endgroup$ Commented Nov 14, 2022 at 17:08
  • 2
    $\begingroup$ I didn't, however one point : The question focus on knowing whether the Dan the torturer can be trusted to worm true news from Alice and Bob. The 30 prisoners are more like material for an "exam", see if Dan gets a A+ or a F- in torture and investigation class. $\endgroup$ Commented Nov 14, 2022 at 18:28


It may seem silly, but tickling is one of the most efficient forms of torture because ticklishness is an involuntary reaction of the hypothalamus. The hypothalamus views the tickling as painful stimuli, and causes the body to react as it would to being tortured with actual pain, though it won’t damage the subject physically. Unless they aren’t ticklish. Then I’d recommend regular torture methods.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .