[wheels]

Can we construct a situation where that would be appropriate? Well, let's get as close as we can, factoring out as many moral variables as possible, to make the question more concrete.

Let's imagine that it's the 1980s, during the Cold War. And let's create an inverse Dr. Strangelove setup.

Let's say that the United States, based on erroneous intelligence, has become convinced that a nuclear attack from the USSR is imminent. To prevent annihilation of the population of the United States, authorization is given to promptly and preemptively attack the USSR and the protocol is started.

Now, let's say that there are a number of persons who are able to disarm these attacks, and that one of them has fallen into KGB hands. The KGB has just one hour to prevent a nuclear holocaust.

To factor out another variable, let's assume that the general being held by the KGB is located somewhere that will be within the blast range of the mutual destruction, such that should the reciprocal attacks take place, the general will in the coming days die of a mix of radiation poisoning and starvation, no doubt a gruesome way to die.

How far is the KGB ethically entitled to go to try to extract the cancelation codes?

Now, "utilitarianism" was mentioned below, and the simple utilitarian would answer, as noted, that if there was even any chance of a small discomfort being removed for a great many, than it would be worth incredible pain for a single person. Such is a common attack on utilitarianism -- that it allows for persecuting the innocent.

However, that's a very superficial view of utilitarianism. Utilitarianism also considers that in trying to maximize the happiness (or some proxy for it) within society, that people care very much about some basic presumption of justice within that society. For instance, I would not expect a society which tortured suspected jaywalkers to be optimized for happiness. It gives us a sense of security and comfort to know that there are certain guarantees in our social systems.

There's also the argument from the perspective of justice (i.e. an argument from rights rather than outcomes). The notion there is that you can construct a set of rights by supposing a "veil of ignorance" in which you do not know where you would fall in a society or situation. For instance, if nobody knew if they were the torturer or torturee, they might both agree that torture is wrong and that it should not be practiced. The "veil of ignorance" is basically what the grandparent's question on transposing nationalities was trying to get at.

The utilitarian viewpoint doesn't lend itself towards as clearly defined of a set of rules for right and wrong. On the other hand, a strict set of rights can draw questions at the extreme ends of the ethical spectrum.

[jacquesm]

In the abstract that's all fine and good but we're talking about some very concrete situations here that go nowhere near those extremes. Extremes are nice to come down on something 'in principle' but barring extraordinary concocted up scenarios to expose the weaknesses of having a pre-set mind about anything at all it is still very useful to come down to some ground rules that everybody lives by, for instance, laws and in this case the Geneva convention.

The idea is that then if someone decides to cross those lines that you try them in court to see if a judge sees it the same way, if not off to the slammer you go.

[wheels]

I agree. But sometimes thought experiments can be useful in sussing out why we hold certain values. Torture I don't actually consider to be one of those that's particularly opaque, but asking "Why do I think this is wrong?" and looking at the edge cases is often an enlightening process. (And I've been reading a good bit of political philosophy lately trying to resolve some dissonance on other topics where I found my own views inconsistent.)

[jacquesm]

It's the engineering mindset at work: reduction to extremes can give you a good idea of whether or not something has a discrete solution or if it is multi-valued. In this case it seems to me that it is likely to be multi-valued but only in non-real-world scenarios and for all intent and purposes you might as well treat it as discrete: torture == bad.

[e12e]

If you're inclined to go down that path: the Japanese treated US POWs terribly, and the US still dropped two nukes as a weapons test (there was credible intelligence Japan was in the process of surrendering because of the "regular" fire-bombing campaigns - but dropping nukes was seen as a great deterrent against Sovjet. And as a great opportunity for a field test).

Note: I'm mentioning this because you're scenario isn't quite as hypothetical as one might hope - for neither those bombed nor for the abused POWs.