In his Foundations of the Metaphysics of Morals1 Kant argued that the moral law (assuming there is one) must be unconditional and universal. As part of that argument he made a famous distinction between categorical and hypothetical imperatives. Imperatives are “ought” (or “should”) statements, such as “you ought to tell the truth”. The difference between the two kinds of imperatives is that hypothetical imperatives depend on a specific kind of condition, namely a desire, while categorical imperatives are universal, unconditional, and absolute. Thus, “if you want human civilization to survive the 21st century, you ought to eat the rich” is a hypothetical imperative, while the previous example, “you ought to tell the truth” is categorical.

An obvious problem with this distinction is that it doesn’t seem to be exhaustive: What about conditional imperatives in which the condition is not a desire? As far as I can see, Kant didn’t consider that option, but I must admit that I find Kant hideously obscure, so I might be wrong. However, the whole point of his appeal to categorical imperatives is that those are the kind of “oughts” he needed for his absolute and universal moral law. That is, any kind of conditional imperatives would – by definition – not be absolute and universal, but relative to their conditions. Hence, it seems that the category of “hypothetical imperatives” coincides with that of “conditional imperatives”, and the category of “categorical imperatives” with “unconditional imperatives”.

Because I’m not perfectly confident that this is right – due to the aforementioned obscurity of Kant, but also due to an obvious incoherence that will be mentioned below – I’ll speak of conditional and unconditional rules rather than of hypothetical and categorical imperatives in the following, however. Contrary to the latter, these notions are quite unambiguous. A conditional rule is a rule of the form “if C, then you should (not) do A” and an unconditional rule is just “you should (not) do A”, in which C stands for some kind of condition, and A for some kind of action (and “not” is either part of the sentence or it is not).

Part of the reason why especially the term “categorical imperative” is so ambiguous is that the same term is also used for Kant’s criterion (or criteria?) to infer absolute and unconditional moral rules. The most famous version of that criterion is the following: “Act only according to that maxim whereby you can, at the same time, will that it should become a universal law”.2 Or in other words, only do A if you want everyone to be allowed to do A. To avoid unnecessary confusion, let’s call this “Kant’s criterion” (rather than “the categorical imperative”).

Almost immediately after publication of Kant’s Foundations it was objected that his moral theory implies that if the principle that one ought to speak the truth is absolutely unconditional, then one should speak the truth to a maniacal murderer, even if that is sure to have disastrous consequences. In a little booklet3 Kant affirmed that the rule against lying is unconditional indeed, but I’ll mostly ignore his arguments in the following.

Imagine the following scenario: Xiuying is home alone. Suddenly someone rings her door. She opens her front door to find her friend Jack standing there. Out of breath, he begs her to let him in and hide him because an insane chainsaw killer is following him, threatening to cut him to pieces. Xiuying let’s Jack in. A few minutes later, someone else rings her door. There’s someone standing outside with a bloodied chainsaw, demanding to know the whereabouts of Jack. What should Xiuying do?

According to Kant, she should speak the truth. His main argument follows from his aforementioned principle. That principle applies always, in every situation, so Xiuying should tell herself that she can only lie if she wants everyone to be allowed to lie. She cannot possibly want that, because that would make all language meaningless and destroy trust, and thereby make society (and language itself) impossible. So, she cannot lie, even to the murderer at her door.

Recall that Kant’s principle (in the simplified rephrasing) is: “only do A if you want everyone to be allowed to do A”. In Xiuying’s conundrum, “lying” is substituted for A (or “lie” for “do A”). But why would she have to make that substitution? Why couldn’t she substitute “lying to a murderer about the whereabouts of his victim”? Surely, she’d want everyone to lie to murderers about the whereabouts of their victims (unless she would be the murderer, of course), so she can lie in that case as well. This, however, would turn the unconditional rule “you should not lie” into a rule with an exception “you should not lie except X” in which X is the exception clause, in this case something like “if telling the truth would have disastrous consequences”. But exceptions are conditions – more specifically, a rule with an exception is two conditional rules combined into one. “A except B” means “A if not B and not A if B”. And because moral rules have to be unconditional – for Kant – this is unacceptable.

There are, however, at least three reasons why this makes no sense. Firstly, if moral rules have to be absolute and unconditional – and thus, exception-less – then there can be at most one rule. If someone finds themselves in a situation in which two rules apply, one of which forbids a certain action, while the other demands that same action, then that person needs to make a choice. But if rules are absolute and unconditional, then she cannot make such a choice – she has to follow both rules, and that’s impossible. Ranking rules cannot solve this, because that would introduce conditions to all lower-ranked ruled – those would all become “you should (not) do A except if that breaks a rule with higher priority”. The only solution is to reduce all of morality to a single rule.

But what if that one rule is just Kant’s principle? That wouldn’t work, because then there would be no way to block “lying to a murderer about the whereabouts of his victim” as a substitution for A in “only do A if you want everyone to be allowed to do A”. Recall that the Kantian argument against that substitution is that it would result in a conditional rule, but that argument now fails, because if Kant’s principle is the only rule, its application doesn’t result in a “rule” at all. Rather, the requirement of unconditionality then only applies to Kant’s principle itself.

Secondly, one and the same action can often – if not always – be described in different terms. If Xiuying speaks the truth to the chainsaw killer, she is also helping a murderer. “Not lying” and “helping a murderer” are not two different actions – they are the same action under two different descriptions. And the alternative action could be similarly described as “lying” and/or “not helping a murderer”. By Kant’s criterion, you shouldn’t lie. By the same criterion, you shouldn’t help a murderer either. Consequently, regardless of what Xiuying does, she is breaking one rule and following another in one and the same action. But the rules are absolute and unconditional, so she has to follow both rules, and that – again – is impossible.

Furthermore, whatever Xiuying does, the rightness or wrongness of her action is conditional on its description – if it is described in one way then she ought to do it; if it is described in another way then she ought not. In other words, unconditional rules make right (and wrong) itself conditional on description.

The third reason why the unconditionality requirement makes no sense also has to do with language, but in a rather different way. In English “lying” means roughly “telling something that is untrue (while knowing that it is untrue) with an intent to deceive” (the clause in parenthesis can be omitted, because it follows from the deceptive intent). Imagine another language in which “lying” just means “telling something that is untrue while knowing that it is untrue” – that is, without the deceptive intent clause. Or actually, you don’t really have to imagine that, because that is very close to what Japanese uso-wo tsuku 嘘をつく means. A Japanese child telling exaggerated or nonsensical stories, but not with the explicit intention of deceiving his audience into accepting them as true, would be telling uso, but would not be lying. A completely accurate translation of “lying” into Japanese would, therefore, have to include an additional clause capturing that intention to deceive that is inherent in the English notion – something like damasu-tame-ni uso-wo tsuku 騙すために嘘をつく, perhaps (literally: “telling uso to deceive”), although this sounds a bit peculiar. And this has an important implication: the unconditional rule “you should not lie” then becomes something in Japanese that literally means that you should not tell uso to deceive – or in other words: “if it is to deceive, you should not tell uso” – and that is a conditional rule.

There are about 7.500 languages in the world today. A couple of years ago, I estimated that the current total number of languages is about 5% of all human languages that have existed throughout history,4 Moreover, there is no reason to believe that the 5% of languages in existence today is a representative sample of all of those languages, and even less to assume that the few languages we are familiar with are representative (of anything, except themselves, perhaps). Given the vast number of languages and their extensive differences, for any unconditional rule in one language, there is probably another language in which a literal (and complete!) translation of that rule would be conditional. Hence, the unconditionality of an unconditional rule is conditional on the language in which it is phrased. But if something can only be conditionally unconditional, then it is not unconditional at all.

Furthermore, it is quite likely that there is or was a language in which the closest equivalent to English “lying” has a clause “except if telling the truth would almost certainly have terrible consequences”. Let’s call that language “Benglish” here, and let’s call that word “blying”. If a speaker of Benglish would apply Kant’s principle to “blying” she would conclude that she ought not to bly. But if Xiuying would speak Benglish, she could speak untruth to the chainsaw murderer at her door and point him in the direction of the Goldman-Sachs office around the corner, for example, and she would not be blying, and thus would not be breaking a rule, because blying is speaking an untruth with deceptive intent an not to prevent something terrible from happening. (Notice that a Kantian could not really object to this on the ground that “blying” has a condition relative to “lying”, because so does “lying” relative to “uso-wo tsuku”.5)

Xiuying wouldn’t even have to switch to Benglish to make this move, moreover. If in her idiolect – that is, her version of English – “lying” means “speaking an untruth with deceptive intent and not to prevent something terrible from happening”, then she could deceive the chainsaw killer without any worry that she would be lying (and thus, that she would be breaking a moral rule). She could even decide that from now on, or in this occasion, this is what “lying” means to her. A Kantian might object to this, but that Kantian then would have to come up with a good argument establishing that the meanings of words are fixed and that people are not allowed to use words in slightly idiosyncratic ways. It doesn’t take much knowledge of how language works to realize that this is an illusion and that no such argument is forthcoming.

The idea of unconditional (moral) rules, then, is nonsensical. Rules could only be unconditional if languages would only differ in the words they use, but not in what those words mean,6 and if any action could be described in only one way. And even then, there could be at most one unconditional rule, unless some law of nature (or God) would prevent conflicts between rules in practice (which obviously isn’t the case either). Does this mean that Kant’s ethics is nonsensical as well? Perhaps, but as mentioned, I find Kant’s writings much to obscure to be sure what exactly he meant.7 One reason to doubt my understanding of Kant is that there appears to be an obvious contradiction in his moral theory.

Kant held that there is a categorical imperative against killing: “you should not kill”. Anyone who breaks that imperative apparently thinks that it is OK to kill, because that follows from the assumption that the killer applied Kant’s principle before deciding to kill, and Kant thinks that this is sufficient reason for society to kill the killer. (There’s some spurious reasoning in this argument, but we’ll ignore that.) Hence, it turns out that Kant’s rule is “you should not kill, except if you are an executioner and you are killing a killer” (or something like that). As mentioned above, an exception is a combination of two conditionals, and consequently, Kant’s insistence on capital punishment for killers turns the supposed categorical imperative against killing into a set of conditional rules. This is so obvious that I find it hard to believe that Kant didn’t notice this. There also is an obvious, albeit merely apparent, escape, however, and I assume that he tried to take that. Instead of applying the rule to killing, it could be applied to murder, and “murder” could be defined something like “killing, except in the capacity of executioner, or soldier, and following legal and proper orders”. That escape, however, doesn’t work – it doesn’t solve the conditionality, but merely pushes it back, much like conditionality in the case of “lying” could be pushed back (even over linguistic boundaries).8 Perhaps, it is less obvious that this escape doesn’t work – if one speaks only one language and believes that the meanings of the words in that language are somehow fixed and given, then it may even seem like a legitimate move.

Anyway, I’m not really interested in interpreting Kant – I gladly leave that to the Kant scholars. My aim here was merely to point out that there are no, and can be no, unconditional rules. Only in a fantasy world with one fixed language and with a god preventing practical conflicts between rules could there be unconditional rules. In the real world, the idea of an unconditional rule is incoherent.9

If you found this article and/or other articles in this blog useful or valuable, please consider making a small financial contribution to support this blog 𝐹=𝑚𝑎 and its author. You can find 𝐹=𝑚𝑎’s Patreon page here.


  1. Grundlegung zur Metaphysik der Sitten (1785).
  2. 4.421
  3. Über ein vermeintes Recht aus Menschenliebe zu lügen (1797).
  4. Lajos Brons (2014) “Language death and diversity: philosophical and linguistic implications” The Science of Mind 52: 243-260.
  5. Unless, of course, that Kantian believes that English or German is the natural language in which all moral rules are to be written.
  6. W.V.O. Quine once called this the “Myth of the Museum”. The (mythical) idea is that differences between languages are like different symbols on signs referring to the same museum exhibits, and thus that there are literal and exact one-to-one translations between words in different languages. Anyone who speaks two or more unrelated languages can tell you that this is a myth indeed, and even a closer look at related languages will reveal the same. — See: W.V.O. Quine (1969), “Ontological Relativity”, in: Ontological Relativity and Other Essays: 26-68, p. 27.
  7. On the other hand, there are other aspects of Kantian ethics that I find rather dubious as well. See, for example, Dao and Second-Order Consequentialism.
  8. If Kant can define “murder” in a way that’s convenient to him, them Xiuying can define “lying” in a way that’s convenient to her.
  9. And indeed, this is a conditional statement. (It’s not a rule, however.)