Why Sam Harris is just plain wrong about torture
In this blog post, Sam Harris once more defends an argument he made comparing torture and collateral damage in The End of Faith. This detailed defense is worth reading, and I think Harris is basically right to compare torture and collateral damage in war, both being willfully done evils which people defend on morally and factually implausible “greater good” grounds. But he’s still wrong about torture, even if he’s basically right about collateral damage — and exactly why he’s wrong is interesting and subtle. Here is the tricky bit that exposes exactly where Harris’ reasoning goes off the rails:
It is widely claimed that torture “does not work”—that it produces unreliable information, implicates innocent people, etc. As I argue in The End of Faith, this line of defense does not resolve the underlying ethical dilemma. Clearly, the claim that torture never works, or that it always produces bad information, is false. There are cases in which the mere threat of torture has worked. As I argue in The End of Faith, one can easily imagine situations in which even a very low probability of getting useful information through torture would seem to justify it—the looming threat of nuclear terrorism being the most obvious case. It is decidedly unhelpful that those who claim to know that torture is “always wrong” never seem to envision the circumstances in which good people would be tempted to use it. Critics of my collateral damage argument always ignore the hard case: where the person in custody is known to be involved in terrible acts of violence and where the threat of further atrocities is imminent.
Harris is simply wrong in thinking that “torture does not work” necessarily means that torture never works or always produces bad information, nor is a critic of his argument logically compelled to take any such extreme and probably unjustifiable position. Rather, the perfectly plausible and evidence-supported claim that torture is massively unreliable is sufficient, because there is no way for an interrogator to know at the time of interrogation whether the information extracted by torture is even remotely true, or in contrast is deliberately misleading. The problem here isn’t that torture always produces false information, but that it often does, so that  there is every reason to think that torture is as likely or more likely to produce bad information or no information than good information, and  there is no way of knowing in any particular circumstances whether this is an occasion where torture will produce good information that will help alleviate a threat, rather than bad information that will exacerbate a threat by wasting time and resources investigating phony leads or mobilizing forces in the wrong fashion.
That torture has on some past occasions produced good intelligence is something that was determined after the fact, just as the fact that torture has resulted in false confessions and other sorts of harmful misinformation on many, many occasions was determined after the fact. The problem, however, is that we don’t have access to that after-the-fact knowledge until, well, after the fact: Interrogators are unable — and almost certainly always will be unable — to determine whether any particular occasion where torture might be considered, including the most radical ticking time bomb “hard case” scenario conjured by Harris or the fevered imaginations of the writers of 24, will after the fact prove to be one of those occasions where torture provides good intelligence rather than harmful misinformation. We cannot know the future in this way except based on probabilities, and the probability of getting misleading and even positively harmful misinformation from torture is high, quite possibly higher than the probability of getting useful intelligence.
Harris not only fails to compare the chance of getting useful information to the chance of getting harmful misinformation, he also fails to compare torture to the obvious alternative: When Harris talks about a scenario in which “even a very low probability of getting useful information through torture would seem to justify it,” the appearance of justification is only preserved by failing to compare torture to non-torture interrogation. It simply isn’t sufficient that torture has some probability, however low, of getting information that could save many innocent lives: To justify its use, torture must have a demonstrably higher probability of getting useful information than non-torture interrogation methods — and it does not.
We know for certain that torture is unreliable. We also know for certain that non-torture interrogation techniques are also unreliable. But there is no good reason to think non-torture interrogation is less reliable than torture, and many reasons to think that it torture is in fact the more unreliable technique by far. What we do not and probably cannot know is whether any particular interrogation situation which confronts us right now, in a given moment, is one where torture would more reliably provide useful intelligence than non-torture interrogation.
If the justification for torture is to commit a moral evil to prevent a greater evil, we must have a reasonable basis for believing that the evil we commit is genuinely worth it. Given our imperfect knowledge of the future, we can only make such a calculation based on probabilities, and Harris utterly fails to consider the appropriate comparisons between probable outcomes. We cannot consider the possibility that torture will net us information that can be used to prevent great harm in isolation, we must compare that possibility to the possibility that torture will net us information that exacerbates the harm or makes it more difficult to prevent the harm, and we must also compare it to the possibility that we could get equally reliable or even more reliable harm-averting information by using well-established interrogation techniques that do not involve torture. Harris’ argument fails to make either comparison, and fails to recognize the need to do so.
In summary, we cannot possibly be justified in committing one moral atrocity to prevent another without having good reasons to think that the moral atrocity we commit has both  a better chance of preventing than exacerbating the atrocity we seek to prevent, and  a better chance to prevent the atrocity than any other action we could take that does not involve committing an atrocity of our own. I think it is highly dubious that we have good reason to think , that torture is more likely generate useful information than misleading misinformation: Certainly, Harris has not made this argument convincingly, nor even perceived the need to make it. I think that the preponderance of the evidence already shows that  is false, although I am open to evidence to the contrary if anyone can produce it: Again, Harris has not provided evidence for , nor even perceived the need to make a case for it. With good prima facie reasons to think that  and  are false, and in the absence of any convincing arguments I am aware of that both  and  are true, universal and unqualified moral opposition to torture is wholly justified. It is justified without relying on any absolutist assertion that torture never produces reliable intelligence, but only on the theoretically coherent and empirically supportable (and I think already quite adequately supported) claim that torture produces less reliable intelligence than well-established interrogation techniques which don’t involve torture.
The error Sam Harris has made here is very much akin to recommending a drug by citing only data comparing the efficacy of the drug to a placebo, and ignoring data comparing the efficacy of the drug to other drugs which are already established as being effective against the same condition and have fewer side effects. This is an error so basic and obvious that someone who is scientifically trained should not miss it — and I think Harris would not miss it, if it weren’t his own flawed reasoning he is defending.