For the past few days I’ve been procrastinating by reading reddit’s askphilosophy, and moral realism has come up again and again. Moral realism has always seemed to me to be a strange, even quixotic, position, so I was very surprised to hear that it is apparently the majority position among academic philosophers. Even worse, a significant amount of philosophers also profess error theory - the idea that moral claims have truth values, and are “objective” in some sense, it is just that every such claim is false.

My natural leanings are towards the expressivist side of the debate, though I do not completely agree with the expressivist “party line”. Expressivism is  typicaly explained as the philosophy that moral statements don’t have truth values, but are simply expressions of attitudes. Essentially, under the expressivist view, a statement like “murder is bad” is simply expressing the proposition “boo murder!”.

A clear problem for expressivism, then, is the Frege-Geach problem, which is that, contra expressivism, propositions involving moral statements appear to be truth-apt. For instance, we can propose the following argument

1.) Murdering people is immoral

2.) Alice murdered Bob

3.) Therefore, Alice committed an immoral act.

We could clearly say this is a valid argument. How then, if moral statements are just expressions of attitude - as they are according to expressivism - can they behave in such a truth-apt manner, like standard logical predicates?

The Frege-geach problem becomes even more daunting for everyone involved when we use other predicates. For instance, consider the following arguments:

1.) People who tell jokes are funny

2.) Bob tells jokes

3.) Therefore Bob is funny


1.) Women with long hair are pretty

2.) Alice has long hair

3.) Therefore Alice is pretty


1.) hoop-skirts are unfashionable

2.) Alice is wearing a hoop-skirt

3.) Therefore Alice is/appears unfashionable.

In fact, pretty much any adjective describing some sort of personal quality or emotion works in such arguments, and therefore must be truth-apt. If the truth-aptness of moral statements is evidence towards moral realism, then so should the truth aptness of statements of humour be evidence towards humour-realism, or aesthetic statements towards aesthetic-realism, and fashion statements towards sartorial realism, an intuitively unappealing conclusion.

Or we could take the opposite tack, that the truth aptness of statements have no bearing on the metaphysical reality of their referent.  However, on the face of it, this is itself a philosophically fraught position. If such sentences are truth apt, then presumably moral propositions have a truth value. But from where can they obtain such a truth value. Under standard denotational semantics, we have a set of acts which includes the act of “murder”; we have another set of acts that are “wrong”. The sentence  “murder is wrong if and only if the act of “murder” is in the set of wrong actions. But then the obvious claim is, what is this set of wrong actions if not a definition for wrong? And this set is presumably global as truth is a global property - 1+1 =2 is not true for me and false for you. Thus we have deived Enoch’s “robust realism”. Moral facts are tue and objective and are actually-existing metaphysical constructs.

To me this seems to be an unsatisfactory conclusion. Generally, I think that trying to derive facts about the world from facts about how humans use language is a fundamentally suspicious enterprise. If this were the case then the universe would be a very weird place indeed. An incredibly strong version of the Sapir-whorf hypothesis would be true. Not only would our language affect our thought, but it could also affect the metaphysical underpinnings of our reality(!). //If we changed how we interpreted a joke about spinach, we would be conjuring up or abolishing a timeless metaphysical entity.

The problem here is the direct reference theory of semantics. Such theories smuggle in a kind of covert Platonism in the way they deal with abstract noun. When we say “X is good”, it must denote the set of things judged (by whom?) to be good. And what of the “meaning”?, the function that maps from entities or situations to judgements of goodness. Where can this function reside except with the other heavenly forms?

This is true of any predicate. I can say “x is blib” and lo and behold I have conjured up an objective, mind-independent quality of blibness. I mean, “blib” is certainly a truth-apt predicate, as can be seen by the following argument:

“If x is blib, then y is not blib

x is blib

therefore y is not blib”

Therefore blib-expressivism fails so there is evidence towards blib-realism, by the argument above.

In general I think direct reference theories of sematntics have two main problems. 1.) the direct reference, and 2.)  the implicitly global theory of truth. I will deal with the second problem first.

I think we need to distinguish between predicates which support global truth values, and those which support “subjective truth values”. A subjective truth value would be one which is agent-specific. These kinds of truth values can be seen, hopefully, incontroversially in sentences like “I like spinach.” The sentenec as it stands is true, or false, only for a single agent - the utterer. To talk about the global truth value of the sentence is meaningless. Nevertheless, such a sentence also passes the Frege-Geach test: it is clearly truth-apt.

The subjective truth value can, of course, be unrolled into a global truth value  by replacing the pronoun “I” with an agent-neutral descriptor - such as “agent 3002432”. The propositoin (“agent 3002432 like spinach”) then has a global truth value. My argument is that moral propositions like “murder is wrong” are implicitly similarly agent-specific propositions. The global proposition corresponding to such is something like “agent x believes that murder is wrong”, or “murder is wrong for agent x”.

The second part of my argument is to do with what the proposition “wrong” is doing in such sentences. Under a direct reference theory of semantics, this could be taken to imply that there is some definite set of “wrong things” which murder is a member of, and thus we again reach realism. I want to argue that this interpretation, and therefore direct reference semantics more generally, is incorrect (I’m planning to write a proper post on this at some point). Basically I think that the meaning of words does not refer to external objects, except indirectly. Meanings are clusters of representations in the mind of an agent. When we say a sentence like “snow is white”, we are referring to mental representations of snow, and whether, in that representation, whiteness is present (and prototypical). This is in direct contavention of the Russelian/Fregean approach to semantics in which a sentence like “snow is white” must refer to actually existing snow, and actually existing white light being reflected from it.

This is because the Russelian/Fregean view takes the truth value of the sentence to be the meaning of the sentence. I don’t. The meaning of the sentence to me is the composition of the various mental repreentations in the mind of an agent who utters or perceives that sentence. The truth value is entirely incidental.

(This has the effect of making meaning agent-relative, but I don’t think this is a cost. In reality, this makes more sense than an implied universal meaning of every sentence. If that were the case, then how could people misunderstand one another?)

These two arguments combine to form the core of my rebuttal to the Frege-Geach problem. The meaning of the sentences are agent-relative, and specific to a cetain agent. The agents tend to assign moral claims internal truth values, so they act as if they are truth apt, however it is not necessarily the case that moral claims possess are global truth values, of the sort required by robust realism. More generally, we can consider each moral claim such as “murder is wrong” as having attached to it an implicit qualifier - “This agent believes”, or “for this agent.” The truth-aptness of the statements is preserved.

A second argument that Enoch brings up is that phenomenologically we feel that moral disagreement is like a disagreement about factual matters, and it feels objective, and that we should take this as evidence towards moral realism being true.

There are two points I want to make here. The first is that I don’t think that simply feeling like some belief is true is a good argument in its favour. That feeling is your prior. Counting your prior as a datapoint is cheating.

The second concerns the nature of objectivity. I’ve been thinking hard about this, and there is no way I can see moral facts, even if they existed, as being objective. I define something as objective if, two different agents, both ideal reasoners, but starting from different priors, are theoretically able to converge on the same answer, using empirical evidence.

Let’s consider a factual but controversial question: Whether anthropogenic global warming is true. One agent believes it is, the other believes it is not. Both are ideal bayesian reasoners. This question is objective, and the fact that is the answer to this question is objective because, given effectively infinite resources, it is possible to answer such a question such that both reasoners are forced to converge on the same answer. There are a whole number of empirical tests, which the agents can agree on beforehand when their outcome is unknown and then update their beliefs according to the outcome of teh experiment. In the limit it would be possible to set up a direct experiment in which two earth replicas areconstructed and one is inhabited by humans producing carbon and the other not, and the climate of both can be tracked over time. This would get close to a conclusive proof, even for the agent whose prior beliefs were wrong.

Then compare this to moral questions. Say we have two agents. One who believes that slavery is morally right, the other who believes it is morally wrong. What experiments can they run to get a definitive answer that both would accept? They can agree on all the facts - that slavery causes suffering, both physical and emotional, and so on, but this does not prove or disprove the hypothesis. The pro-slavery agent could simply assert that suffering is morally good, and it is impossible to disprove empirically.  The agents could run a survey but this merely reports what other agents think about the question, and the other agents could be just as wrong as them. Indeed, suppose that moral facts exist and the pro-slavery agent was correct. Slavery is objectively morally good. The supposed “moral progess” of the past few centuries is actually moral decline. How would we ever find out and, if were somehow proved incontrovertibly, why should we care?

This ties in more generally with the problem of moral causation. Suppose that moral facts exist and are objective. How do they causally interact with the world? Unlike physical laws they do not define physical constraints, nor are they amenable to deduction through empirical observation. Under a materialist worldview there is no obvious way for a moral fact to interfere at all with the causaliy of the world. They do not have the power to move atoms or energy. But if they cannot causally interact in any way, then Occam’s razor would suggest they are needless, and should be eliminated.