The Good, The Bad, and The Empty

I've never understood Kant. His take on lying, for instance, always struck me as bizarre, and not even supported by the categorical imperative. Maybe it's just my interpretation of the categorical imperative that's nonstandard, but it seems right to me.

Consider the famous case (discussed by Kant in On a Supposed Right to Lie) of the murderer at the door, looking to kill a friend of yours (who you know to be inside your house). Kant claims that, if the murderer asks you if your friend is there, and you must "answer Yea or Nay", then you have a duty to tell the truth. Why do you have to give him good information? Because you could not universalize the maxim underlying the lie you want to tell, since no one would believe you if lying was a universal law. Kant goes on to say some weird things about responsibility, but we don't need to get into that.

Kant's mistake is clearest if we think about the communication between you and the murderer as a transmission of information. If you answer honestly, then you transmit what we might call positive information - information that points him in the direction of true beliefs about where your friend is. If you're dishonest, then you transmit what we can call negative information - information that makes his beliefs further from the truth than they were before. In this case, we can think of the transmission of information as altering the subjective probability that the murderer places on your friend being in your house. Because your friend is in your house, positive information is whatever makes this probability closer to 1 and negative information whatever makes it closer to 0. If you remain silent, then you transmit zero information - the murderer's subjective probability remains unchanged.

Now, efforts to transmit negative information tend to fail the categorical imperative, whether they are outright lies (false statements) or less direct attempts to mislead. Universalizing a maxim essentially means making your decision procedure transparent to the person who you are communicating with. Once he knows the relationship between your beliefs and your utterances, there he will not be deceived by the verbal content of your utterances, as this content is irrelevant. So if your communications in this context do not reliably track the truth, he can dismiss them as bad information.

However, efforts to transmit zero information withstand universalization. All that you have to do is give the same response, regardless of what is true. No matter where your friend is, say "No, he is not here." This response is just as good as silence, since even if the murderer knows your decision procedure, and every person in the world would follow the same maxim, he cannot learn anything from your statement. Kant seems to have had an inkling that transmissions of zero information were not defeated by the categorical imperative, and to have tried to exclude these responses, which is why he ruled out silence as a possible response. However, he failed to realize that any answer could be an empty communication, as long as it is given consistently.

The categorical imperative does rule out another pattern of responding, which intially seems cleverer than the "always say no" strategy: the "always lie" strategy. If you say no when your friend is there and yes when he is not, then, if your maxim were universalized, it would be just as useful to the murderer as honesty. Your responses distinguish the two states of the world, and even though you attempt to do so in a way that transmits negative information, a murderer who knew your decision procedure could see past your utterances to recognize that you did serve as a reliable indicator of the truth. You would be like a poker player with a stupid tell: you said your cards were good when you were bluffing, and claimed to bluff whenever you had a good hand. Or, you'd be like one of those knaves who always lie in logic puzzles, and the murderer would be in a position to discover the correct solution.

To truly be clever, you need a more flexible strategy than always lying: say whatever you think will minimize the subjective probability that the murderer has for whatever state of the world in fact obtains. This puts you in a battle of wits with the murderer. You could play it safe and just try to transmit zero information, or you could try to outsmart the murderer in a way that transmits negative information. You might try to size up your opponent: If I say this, he'll think that I think that ..." If you do try to transmit negative information, though, then you run the risk of losing the battle of wits and having the murderer see through your ploy (recognize your tell) and gain positive information. However, if you're smarter than your opponent, in the murderer scenario or any other, then you're likely to succeed in transmitting negative information. Indeed, if the murderer is going through the trouble of asking you whether your friend is there, then there's a good chance that he's gullible enough to at least put some credence in your claim that your friend is not there.

When people universalize a communications maxim, they tend to do so in such a way as to rule out the possibility of succeeding through your superior cleverness, or your opponent's foolishness. Thus, it simplifies the problem to one where all parties are rational, and makes morality into a game theory problem. The right action is whatever communications strategy, pure or mixed, maximizes your expected benefits. This leads to a zero information strategy in most cases that I can think of, including the murderer at the door scenario. However, the categorical imperative really has nothing to say against the strategy of communicating information that is as negative as possible, as long as you consider the limiting case, transmitting zero information, to be among the acceptable outcomes that do not defeat your communicative intention.

There are further wrinkles. One type of transmission of negative information that can withstand the categorical imperative, when used in moderation, is mimicry of honest communication. Say that you would ask for a loan and intend to pay it back in situations X, Y, and Z. If there is a situation, W, where you want to ask for a loan but do not intend to pay it back, and if situation W is indistinguishable from situations X, Y, and Z to the potential lender, then, even after making your decision procedure transparent (i.e. universalizing), you may still be able to get away with asking for the loan in situation W. The potential lender's subjective probability that you are lying is the frequency of W divided by the total frequency of W, X, Y, and Z, and if this probability is low enough for him to be willing to take the risk then you may have lied yourself into a loan, while remaining covered by the categorical imperative.

Similarly, using a "This house protected by ABC Security" sticker as your only home security system can be allowed by the categorical imperative, as long as the cases where you would have actually installed an ABC security system are common enough, and indistinguishable from the actual case to the eyes of a potential criminal, so that potential burglars would consider your house to be too great of a risk. So, for instance, you could save on your expected home security costs without running afoul of the categorical imperative by flipping a coin and buying an ABC security system and a sticker if it comes up heads, and just putting the sticker in your window if it comes up tails. But if you let the probability drop far below .5, then your security plan would not survive universalization, and, in practice, you would be free riding on other people who shell out for the real security system.

Of course there are other decision procedures besides coin flipping, such as looking at your budget to see if you can afford an ABC security system. However, you must be sure that you are being honest with yourelf about your decision procedure (are you setting the conditions for buying a real security system only after you know that you will not meet those conditions?), because if you are just coming up with an excuse to free ride then you fail the categorical imperative. You also must succeed at mimicry if you end up transmitting negative information, although here it is OK to "free ride" and to mimic the cases where other people actually buy home security systems, rather than the cases where you would have bought a home security system, since if you had bought a home security system under different conditions others would have been able to mimic those conditions.

The right way to decide what to do under deontological morality can often be more similar to consequentialist thinking than many deontologists realize.

Related: My incomplete series on truth, knights & knaves
Richard on lying