Interesting comment by Gregory Cochran on torture not being useless as is often claimed.
Torture can be used to effectively extract information. I can give you lots of examples from WWII. People often say that it’s ineffective, but they’re lying or deluded. Mind you, you have to use it carefully, but that’s true of satellite photography. Note: my saying that it works does not mean that I approve of it.
… At the Battle of Midway, two American fliers, whose planes had been shot down near the Japanese carriers, were pulled out of the water and threatened with death unless they revealed the position of the American carriers. They did so, and were then promptly executed. Later, at Guadalcanal, the Japanese captured an American soldier who told them about a planned offensive – with that knowledge the Japanese withdrew from the area about to be attacked. I don’t why he talked [the guy didn’t survive] – maybe a Japanese interrogator spent a long time building a bond of trust with that Marine. But probably not. For one thing, time was short. I see people saying that building such a bond is in the long run more effective, but of course in war, time is often short.
You could consider the various agents that the Germans inserted into England: the British captured almost every one of them, and gave them the choice of cooperation (which included active participation in British deception schemes) or execution. Most cooperated.
The Germans tortured members of the various underground groups in Europe – and some of them never broke. But some did. You may have heard of Jean Moulin not breaking under torture, even unto death: but the Gestapo caught him because Jean Multon did break. To avoid being tortured, Multon agreed to work for the Gestapo. Over the next few days he led his captors to more than 100 members of the Resistance in Marseilles. He then gave away more in Lyons. Some of those he betrayed themselves broke under torture by the Gestapo. Things snowballed, and the whole network was torn to pieces.
People often argue that people under torture will say anything that their interrogators want to hear, and are thus useless as sources of information. There is something to that, but to a large degree that depends on what goals the interrogators actually have. For example, in the Iraq war, American higher-ups often didn’t want information – they wanted their fantasies confirmed. They knew that anti-American guerrillas couldn’t be motivated by nationalism or Islam – they had to be paid Baathist agents. Or there had to a connection between Saddam and Al-Qaeda. Whatever. Most told something close to the truth, but that wasn’t good enough, and so, torture. In much the same way, Stalin tortured until he got what he wanted – false confessions for show trials, rather than actual information about Trotskyist conspiracies (that didn’t even exist). Most people broke – I remember that a Chekist said, admiringly, that Lev Landau held out a long time – three broken ribs before giving in. The Japs at Midway wanted real info, not ammunition for their fantasies.
If an interrogator wants valid information, he can see if the stories of several different prisoners agree. He can see if their story checks with other sources of information. etc. It’s like any other kind of intelligence.
At least some of the arguments about the effectiveness of torture are obviously false, not even meant to make sense. For example, I have seen people argue that torture is pointless because the same information is always available by other means. Of course, since the products of various kinds of intelligence often overlap, you could use that argument to claim that any flavor of intelligence [ cryptanalysis, sigint, satellite recon, etc) is useless. But multiple leads build confidence. Sometimes, you can get information via torture available in no other way. If you are smart, and if information is what you really want.
This seems an insightful and true statement. We seem to like “protecting” ought by making false claims about what is.
We seem to like “protecting” ought by making false claims about what is.
Possibly related to the halo or overjustification effects; arguments as soldiers seems especially applicable—admitting that torture may actually work is stabbing one’s other anti-torture arguments in the back.
I read somewhere that lying takes more cognitive effort than telling the truth. So it might follow that if someone is already under a lot of stress—being tortured—then they are more likely to tell the truth.
On the other hand, telling the truth can take more effort than just saying something. Very modest levels of stress or fatigue make it harder for me to remember where, when, and with whom something happened.
I agree that it is a PC thing to say now in the US liberal circles that torture doesn’t work. The original context was different, however: torture is not necessarily more effective than other interrogation techniques, and is often worse and less reliable, so, given its high ethical cost to the interrogator, it should not be a first-line interrogation technique. This eventually morphed into the (mostly liberal) meme “torture is always bad, regardless of the situation”. This is not very surprising, lots of delicate issues end up in a silly or simplistic Schelling point, like no-spanking, zero-tolerance of drugs, no physical contact between students in school, age restrictions on sex, drinking, etc.
FM 34-52 Intelligence Interrogation, the United States Army field manual, explains that torture “is a poor technique that yields unreliable results, may damage subsequent collection efforts, and can induce the source to say what he thinks the interrogator wants to hear.”[4] Not only is torture ineffective at gathering reliable information, but it also increases the difficulty of gathering information from a source in the future.
We seem to like “protecting” ought by making false claims about what is.
This has interesting implications for consequentialism vs. deontology. Consequentialists, at least around here, like to accuse deontologists of jumping through eleborate hoops with their rules to get the consequences they want. However, it is just as common (probably more so) for consequentialists to jump through hoops with their utility function (and even their predictions) to be able to obey the deontological rules they secretly want.
This seems an insightful and true statement. We seem to like “protecting” ought by making false claims about what is.
Certainly true—I believe a lot of claims about the healthiness of vegetarianism fall into that category.
Another problem is taking something that’s true in some cases, or even frequently, and claiming that it’s universal. In the case of torture, it’s one thing to claim that torture rarely produces good information, and another to claim that it never does.
The point on torture being useful seems really obvious in hindsight. Before reading this I pretty much believed it was useless. I think it settled my head in the mid 2000s, arriving straight from political debates. Apparently knowing history can be useful!
Overall his comment is interesting but I think the article has more important implications, someone should post it. So I did. (^_^)
I don’t see anything insightful about the statement. It rather trival to point out that there were events were torture produced valuable information. Nobody denies that point.
It rather sounds like he doesn’t understand the position against which he’s arguing.
If an interrogator wants valid information, he can see if the stories of several different prisoners agree. He can see if their story checks with other sources of information. etc. It’s like any other kind of intelligence.
It’s not like any other kind of intelligence. This ignores the psychological effects of the torture on the person doing the torturing. Interrogators feel power over a prisioner and get information from them. That makes them spend to much attention of that information in contrast to other information.
This ignores the psychological effects of the torture on the person doing the torturing. Interrogators feel power over a prisioner and get information from them. That makes them spend to much attention of that information in contrast to other information.
And this is different from someone who, say, spends a lot of effort turning an agent, or designing a spy satellite, how?
Beating someone else up triggers primal instincts. Designing a spy satelite or using it’s information doesn’t.
There’s motivated reasoning involving is assessing the information that you get by doing immoral things as high value.
Pretending that there are no revelant psychological effects from the torture on the person doing the torturing just indicates unfamiliarity with the arguments for the position that torture isn’t effective.
I would add that that as far as the description of the battle of Midway in the comment goes, threating people with execution isn’t something that in the US would be officially torture. Prosecutors in Texas do it all the time to get people to agree to plea bargains.
It’s disgusting but not on the same level as putting electrodes on someone’s genitals. It also doesn’t have the same effects on the people doing the threating as allowing them to inflict physical pain.
If you threaten someone with death unless he gives you information you also don’t have the same problem of false information that someone will give you to make the pain stop immediately.
As far as the other example in that battle goes, the author of the comment doesn’t even know whether torture was used and seems to think that there are no psychological tricks that you can play to get information in a short amount of time. Again an indication of not having read much about how interrogation works.
Here on Lesswrong we have AI players who get gatekeepers to let the AI go in two hours of text based communication. As far as I understand Eliezer did that feat without having professional grade training in interrogation. If you accept that’s possible in two hours, do you really think that a professional can’t get useful information from a prisioner in a few hours without using torture?
As far as the other example in that battle goes, the author of the comment doesn’t even know whether torture was used and seems to think that there are no psychological tricks that you can play to get information in a short amount of time.
From what I heard, most of said psychological tricks relay on the person you’re interrogating not knowing that you’re not willing to torture them.
Here on Lesswrong we have AI players who get gatekeepers to let the AI go in two hours of text based communication.
Not reliably. This worked on about half the people.
If you accept that’s possible in two hours, do you really think that a professional can’t get useful information from a prisioner in a few hours without using torture?
Depending on the prisoner. There are certainly many cases of prisoners who don’t talk. If the prisoners are say religious fanatics loyal to their cause, this is certainly very hard.
From what I heard, most of said psychological tricks relay on the person you’re interrogating not knowing that you’re not willing to torture them.
Being able to read bodylanguage very well is also a road to information. You can use Barnum statements to give the subject the impression that you have more knowledge than you really have and then they aren’t doing anything wrong if they tell you what you know already.
Depending on the prisoner. There are certainly many cases of prisoners who don’t talk. If the prisoners are say religious fanatics loyal to their cause, this is certainly very hard.
In the case in the comment the example was an American soldier who probably doesn’t count as religious fanatic. The person who wrote it suggested that the fast transfer of information is evidence of there being torture involved.
It was further evidence for my claim that the person who wrote the supposedly insightful comment didn’t research this topic well.
I case wasn’t that there certain evidence that torture doesn’t work but that the person who wrote the comment isn’t familiar with the subject matter and as a result the comment doesn’t count as insightful.
Not reliably. This worked on about half the people.
Similarly, basilisks would work as motivation to develop a certain kind of FAI but there’s a ban on discussing them here. Why? Isn’t it worth credibly threatening to torture people for 50 years to eventually save some large number of future people from dust specks (or worse) by more rapidly developing FAI?
It’s possible that the harm to society of knowing about and expecting torture is greater than the benefit of using torture. In that case, torturing in absolute secret seems to be the way to maximize utility. Not particularly comforting.
A) It’s not credible
B)The basilisk only “works” on a very few people and as far as I can tell it only makes them upset and unhappy rather than working as hard as they can on FAI.
C) Getting people on your side is pretty important. Telling people they will be tortured if they don’t get on your side is not a very good move for a small organization.
It’s possible that the harm to society of knowing about and expecting torture is greater than the benefit of using torture. In that case, torturing in absolute secret seems to be the way to maximize utility. Not particularly comforting.
Um, the threat of torture only works if people know about the threat.
Interesting comment by Gregory Cochran on torture not being useless as is often claimed.
This seems an insightful and true statement. We seem to like “protecting” ought by making false claims about what is.
Possibly related to the halo or overjustification effects; arguments as soldiers seems especially applicable—admitting that torture may actually work is stabbing one’s other anti-torture arguments in the back.
I read somewhere that lying takes more cognitive effort than telling the truth. So it might follow that if someone is already under a lot of stress—being tortured—then they are more likely to tell the truth.
On the other hand, telling the truth can take more effort than just saying something. Very modest levels of stress or fatigue make it harder for me to remember where, when, and with whom something happened.
I agree that it is a PC thing to say now in the US liberal circles that torture doesn’t work. The original context was different, however: torture is not necessarily more effective than other interrogation techniques, and is often worse and less reliable, so, given its high ethical cost to the interrogator, it should not be a first-line interrogation technique. This eventually morphed into the (mostly liberal) meme “torture is always bad, regardless of the situation”. This is not very surprising, lots of delicate issues end up in a silly or simplistic Schelling point, like no-spanking, zero-tolerance of drugs, no physical contact between students in school, age restrictions on sex, drinking, etc.
Could you provide evidence for this claim?
Going by the links on Wikipedia. A quote:
test
This has interesting implications for consequentialism vs. deontology. Consequentialists, at least around here, like to accuse deontologists of jumping through eleborate hoops with their rules to get the consequences they want. However, it is just as common (probably more so) for consequentialists to jump through hoops with their utility function (and even their predictions) to be able to obey the deontological rules they secretly want.
Real humans are neither consequentialists nor deontologists, so pretending to be one of these results in arguments like that.
Certainly true—I believe a lot of claims about the healthiness of vegetarianism fall into that category.
Another problem is taking something that’s true in some cases, or even frequently, and claiming that it’s universal. In the case of torture, it’s one thing to claim that torture rarely produces good information, and another to claim that it never does.
Arguments as soldiers in regards to universities divesting from fossil fuels.
The point on torture being useful seems really obvious in hindsight. Before reading this I pretty much believed it was useless. I think it settled my head in the mid 2000s, arriving straight from political debates. Apparently knowing history can be useful!
Overall his comment is interesting but I think the article has more important implications, someone should post it. So I did. (^_^)
I don’t see anything insightful about the statement. It rather trival to point out that there were events were torture produced valuable information. Nobody denies that point. It rather sounds like he doesn’t understand the position against which he’s arguing.
It’s not like any other kind of intelligence. This ignores the psychological effects of the torture on the person doing the torturing. Interrogators feel power over a prisioner and get information from them. That makes them spend to much attention of that information in contrast to other information.
And this is different from someone who, say, spends a lot of effort turning an agent, or designing a spy satellite, how?
Beating someone else up triggers primal instincts. Designing a spy satelite or using it’s information doesn’t.
There’s motivated reasoning involving is assessing the information that you get by doing immoral things as high value.
Pretending that there are no revelant psychological effects from the torture on the person doing the torturing just indicates unfamiliarity with the arguments for the position that torture isn’t effective.
I would add that that as far as the description of the battle of Midway in the comment goes, threating people with execution isn’t something that in the US would be officially torture. Prosecutors in Texas do it all the time to get people to agree to plea bargains. It’s disgusting but not on the same level as putting electrodes on someone’s genitals. It also doesn’t have the same effects on the people doing the threating as allowing them to inflict physical pain.
If you threaten someone with death unless he gives you information you also don’t have the same problem of false information that someone will give you to make the pain stop immediately.
As far as the other example in that battle goes, the author of the comment doesn’t even know whether torture was used and seems to think that there are no psychological tricks that you can play to get information in a short amount of time. Again an indication of not having read much about how interrogation works.
Here on Lesswrong we have AI players who get gatekeepers to let the AI go in two hours of text based communication. As far as I understand Eliezer did that feat without having professional grade training in interrogation. If you accept that’s possible in two hours, do you really think that a professional can’t get useful information from a prisioner in a few hours without using torture?
From what I heard, most of said psychological tricks relay on the person you’re interrogating not knowing that you’re not willing to torture them.
Not reliably. This worked on about half the people.
Depending on the prisoner. There are certainly many cases of prisoners who don’t talk. If the prisoners are say religious fanatics loyal to their cause, this is certainly very hard.
getting half your prisoners to capitulate is still pretty damn good.
Being able to read bodylanguage very well is also a road to information. You can use Barnum statements to give the subject the impression that you have more knowledge than you really have and then they aren’t doing anything wrong if they tell you what you know already.
In the case in the comment the example was an American soldier who probably doesn’t count as religious fanatic. The person who wrote it suggested that the fast transfer of information is evidence of there being torture involved.
It was further evidence for my claim that the person who wrote the supposedly insightful comment didn’t research this topic well.
I case wasn’t that there certain evidence that torture doesn’t work but that the person who wrote the comment isn’t familiar with the subject matter and as a result the comment doesn’t count as insightful.
Nothing works 100% reliably.
Similarly, basilisks would work as motivation to develop a certain kind of FAI but there’s a ban on discussing them here. Why? Isn’t it worth credibly threatening to torture people for 50 years to eventually save some large number of future people from dust specks (or worse) by more rapidly developing FAI?
It’s possible that the harm to society of knowing about and expecting torture is greater than the benefit of using torture. In that case, torturing in absolute secret seems to be the way to maximize utility. Not particularly comforting.
A) It’s not credible B)The basilisk only “works” on a very few people and as far as I can tell it only makes them upset and unhappy rather than working as hard as they can on FAI. C) Getting people on your side is pretty important. Telling people they will be tortured if they don’t get on your side is not a very good move for a small organization.
Um, the threat of torture only works if people know about the threat.