I’ve been reading David Enoch’s great Taking Morality Seriously. Enoch defends Robust Realism according to which there are judgment-independent non-natural (causally inert) normative properties. One of the objections to Robust Realism briefly discussed in the book is the problem of semantic access. Enoch is explicitly very modest when he responds to this objection (he merely explains how a response might go). I still want to raise a question about this response as I think that we are getting here into very deep and interesting questions about normative concepts and properties.
The problem of semantic access is the question of how our talk and thought came to refer to the non-natural normative properties. The standard stories of how many other terms come to refer to ordinary objects and properties often make use of our causal interaction with those objects and properties. Obviously a Robust Realist cannot give a similar story of how the reference of our normative language gets fixed, as according to her there are no causal connections between our concepts and the denoted properties.
Enoch borrows the response to this challenge from Ralph Wedgwood and his version of Conceptual Role Semantics. Here’s a rough sketch of Wedgwood’s view as a starting point. It begins from logical connectives. Take conjunction. There are basic rules that govern the sentences of the type ‘A&B’. If you accept A and you accept B and the question arises, then you are committed to accepting also A&B. And, if you accept A&B, then you are committed to accepting A and accepting B. Furthermore, if you grasp the concept &, then you must also understand these basic rules that govern the relevant sentences. This also makes the previous basic rules rules of rationality; it is irrational to violate them if you understand the concept (and thus are aware of the rules).
On Wedgwood’s view, the conceptual role of a concept – the basic rules governing its use in inferences – also determines the reference/semantic value of the concept. And, in the case of logical connectives this is plausible. The assumption is that the reference/semantic value of a basic sentence is a truth-value, and that the reference/semantic values of the complex sentences are truth-values too but ones that are functions of the truth-values of the basic sentences. They also have to be the specific functions that make certain inferences in which the connective are used valid, namely ones that make best sense of the basic rules governing the connectives. And, this is how the conceptual role of the connectives fix their references to certain truth-functions.
Enoch and Wedgwood both want to use this view to explain how the reference of normative terms too gets fixed to the non-natural properties. There are basic rules that govern expressions such as ‘I have most reason to phi…’. One of them might be that if I accept the sentence, then I am committed to intending to phi. If I grasp the concept, I understand this basic rule and thus I will be irrational not to conform to it. Now, how does this get to the reference? Well, we must first ask what would make best sense of the basic inferential rules governing the concept? Some basic moves from ordinary beliefs to the reason-belief and from them to the intentions are correctness-preserving (this is analogical to truth-preservingness and based on whatever the goal of practical reasoning is). The solution to the reference-fixing problem for Enoch is then provided by the thought that, if the concept of reasons didn’t really refer to some non-natural sui generis causally inert normative reasons-relations, then the basic rules would not lead from correct input of practical reasoning to correct intentions (relative to whatever the goal of practical reasoning is). And, so, because that reference would be the best validation of the conceptual role of ‘reasons’, therefore it must be the reference. Thus, the conceptual role has again explained how the reference got determined.
Now, as you notice, this sketch is missing a lot details (many of which have been provided by Wedgwood). In the case of logical connectives, we know a lot about validity, truth and truth-preservingness, and so on. But, in the normative case, there’s less agreement about what the basic rules are, that it is what it is for the practical inferences to be ‘valid’ as in correctness-preserving (and what it is for the states to be correct too). This is because there is disagreement about the goals of practical reasoning. There are also questions of why and how Enoch's non-natural normative properties would be the best vindication for the inferential patterns.
I want to pursue another line of thought. With many concepts, it seems like the conceptual role is not enough to fix the reference of the concept. Consider water. There are basic transition rules that govern the use of the concept that anyone who understands the concept must grasp – this falls from the sky -> this is water, this is water -> this extinguishes thirst, and so on. The reason why these rules cannot determine the reference is that presumably the basic inferential rules that govern the concept on Twin Earth are the same as here. So, if the conceptual role determined the reference, our concept of water and theirs would have the same reference, and that’s just not true.
To deal with this problem, the conceptual role semanticists have to either admit that something other than the conceptual role fixes the reference or part of the conceptual role is our causal interaction with the substance (which creates problems elsewhere). This is to think that not only intralanguage rules determine the conceptual role but also extra-linguistic transitions.
So, now, the question I have is why the normative terms are supposed to be more like the logical connectives where the conceptual role is able to determine the reference without any causal connection and less like the term water where it isn’t (or where the role must be specified in terms of causal interaction)? I guess this is a question I would like to ask from those who are attracted to conceptual role semantics about normative terms.
For what it's worth, and I’m very uncertain about the rest, there seems to be one relevant difference between the conjunction case and the water case. In the conjunction case, given that we know the conceptual role is – which specific inferences the reference of the concept is supposed to make valid, there’s only one candidate for what the reference could be. There can be just one truth-function that could deliver the validity of those inferences. So, there’s no room for Twin Earth cases. Whatever the metaphysical realisation of the truth-function happens to be (and maybe there are many), we still essentially have the same function in terms of it making the same inferences valid. So, in a sense, what the reference is in the metaphysical sense just doesn’t matter as much as what the function does – spurns out certain truth-values from other truth-values. As Wedgwood notes, we can do this even without truth-values and truth-functions in terms of sets of possible worlds and set-functions (and of course there are many understandings of those).
So, what about the normative properties case? I can think of three options. Firstly, the Robust Realist could say that there could be in principle only one kind of non-natural causally inert properties, so therefore we need those properties as references to make ok our practical inferences. This is to say that differently constituted ‘normative XYZ’ that does all the same things as good old ‘normative H2O’ is impossible for some reason. Given how little we know of the constitution of the non-natural normative properties I find this option difficult to motivate. The second option is to say that (i) there are many possible non-natural normative properties in terms of their constitution and (ii) some magical non-causal mechanism fixes the references of our concepts to reasons, oughts, and good rather than to treasons, toughts, and tgood.
The final, and perhaps most plausible, model is to follow the logical connectives case again. What we really are after are certain functions – ones that take thoughts about circumstances and their correctness-values as arguments and give correctness-values of practical attitudes as outputs (as familiar, these inputs and outputs could be fact-plan worlds). Admittedly, just like in the case of truth-functions and truth-values, these correctness-values and correctness–functions could be perhaps realised in different ways on the metaphysical level.
But, with respect to making ok the practical inferences, it’s not that the metaphysical nature of the correctness-functions matters but rather only what correctness-values the functions give for the practical conclusions out of the correctness-values of their arguments. Perhaps even something natural (or abstract, or fictional, or) could realise those functions, but even this would not mean identifying the normative properties with natural properties given that the realiser of the function is not the same as the function itself. So, in the same way as the metaphysical nature of the truth-functions is irrelevant, the proper reference of the normative terms and thus the metaphysical nature of the normative properties drops out here again as something that doesn’t do much work and isn’t very interesting. Thus I’m starting to think that the conceptual role semantics undermines the metaphysical picture of Robust Realism which seems to be far more picky about the metaphysical nature of the relevant properties.
Hmm, I’m not sure I see the problem for the Robust Realist here. Either there’s a uniquely eligible candidate non-natural “realiser”, or there are multiple equally-eligible candidates. If there’s only one, no problem. If there are multiple, then can’t we just take their disjunction to be the referent of the normative concept in question?
We might allow for a kind of multiple (or disjunctive) realisability in this way whilst still being “picky” in taking the conceptual role to restrict the metaphysical nature of candidate “realisers” to only non-natural properties, no?
(I find this topic a little slippery, though, so do let me know if I’m missing something!)
Thanks Richard. I find this topic very slippery too. Here’s a couple of thoughts:
With the one uniquely eligible candidates, I think we can create a similar propblem even then. Presumably there are lots and lots of normative concepts if we include all the thick evaluative, aesthetic, moral and so on concepts. If these are to be different concepts, they need to have different conceptual roles – that is the basic rules for making inferences with these concepts are going to be only slightly different. So, consider the example of grace and delicacy. Presumably these are different concepts.
Now, we also need equally many normative properties on the Robust Realism picture. But, how do we then know that not one of these properties, say, gracefulness makes both ‘grace’ and ‘delicacy’ involving inferences ok. So, the question I guess how could the inferential roles of these concepts be so fine-grained that they fix a unique referent even of the existing normative properties?
I agree about the second point. There’s a worry there too, however. The conceptual role semanticist only cares about the reference in so far as it is needed to make the inferences correctness-preserving. I take it this is why according to Wedgwood the semantic value of ought is the ‘weakest’ property of a proposition that makes it correct to it as a part of one’s ideal plan. Given that this is the weakest property and we are non-committal about the correctness at this stage, you might think that the conceptual role per se might leave natural properties to play the realiser role. So, I guess that it’s hard to see how the conceptual role itself could limit the candidates to non-natural properties.
I agree that this topic is slippery, but I am sympathetic to the general Wedgwood/Enoch picture. Here is my own view of how this goes.
Consider the property of being valid, in the logical sense. Here is a klutzy way to reason: “All As are Bs; All Bs are Cs; the foregoing is an instance of the Barbara syllogism, which is valid; therefore All As are Cs.” It is much shorter to simply reason, “All As are Bs; All Bs are Cs; therefore All As are Cs.” In very complex situations (analyzing philosophy articles, for example) it might help to reason in the first, usually klutzy way. Or maybe if you are fighting wishful thinking tremendously hard. But in general, the concept of validity is introduced in attempting to talk about reasoning and evaluate it. It is, in a loose sense, metalanguage.
I think the same goes for normative properties like “ought”, “is a reason to”, “have most reason to”, “is morally required/permitted”, etc. The simple way to reason practically is to have one’s reasons as premises and one’s intentions as conclusions: “A, B, C, therefore I’ll do X.” The klutzy way is to introduce normative terms: “A, B, C; if A, B, C then I ought to do X; so I’ll do X.” On occasion it may be useful to introduce normative terms, but their basic home is in the attempt to talk about our practical reasoning and evaluate it.
As you rightly note, Jussi, I am very tentative about the whole thing, partly because I just don’t feel competent enough about this. Still, two points:
First, *some* indeterimnacy is not a problem for RR, right? I mean, it’s totally fine if the conceptual role for delicacy and grace aren’t fine-tuned enough. Waht would be threatening is much more by way of indeterminacy. And I don’t yet see that you’ve argued that case.
Second, you point out that on this picture hte metaphysical stuff drops out of the picture — the functional stuff does all the work. I’m not totally sure abut this, but I’m accepting this at least for the sake of argument (at least for now). Still, this is consistent with RR. It’s possible that we don’t need, as it were, the metaphysical stuff *here*, in order to supply a decent account of reference. But we do need the metaphysical stuff elsewhere (say, to give the right kind of objectivity, or as what is deliberatively indispensable). If so, we have these other reasons to believe that the only thing that can satisfy the functional role (which is the only thing directly doing semantic work) is the heavuer metaphysical stuff. And that’s ok for RR, isn’t it?
(Have I just rejected what I said I would accept for the sake of argument?)
Hi David,
thanks. I don’t feel competent enough about this yet either but I’m trying to investigate this more at the moment. I realise that my comments above are horribly sketchy but I really do want to write a paper on this. At the moment, I’m quite impressed by Hilary Putnam’s argument against conceptual role semantics in the case of logical connectives. I’m thinking that it’ll work for normative terms too.
Yes – I need to come up with better cases than that even if, for certain other theories, some people (like Roger Crisp) have taken that degree of indeterminacy to be decisive. I need to show that metaphysically very different kinds of referents can make the same practical inferences equally correctness-preserving.
With the second point, I think I see the requirements for RR’s semantic theory slightly differently. You seem to think that it’s ok if we can show here that the conceptual role *could* fix the reference to the non-natural properties, and then fight other metaethical views elsewhere.
I’m thinking that what we would need here is a stronger argument to the conclusion that the conceptual role *in fact fixes* reference to the non-natural properties. Otherwise, it could be the case that the non-natural properties exist but we in fact never talk about them because the reference of our concepts has been fixed to some other properties or even that they lack reference altogether.
Yes – it seems like you rejected what you said you would accept. Not that that is necessarily a bad thing. Anyway, I’ve got another paper to write for now but this will be probably in line after that. I hope I’ll be able to soup up my argument and I’ll probably discuss Wedgwood more given that this is his theory. I might use some help though at some point.