From the ALWS archives: A selection of papers from the International Wittgenstein Symposia in Kirchberg am Wechsel

1. The Proposal

The suggestion I want to put forward is that reflection on a correct account of rule-following in ethics invites to consider what I will call a ‘three fold model’ of conceptual application, that varies not just from the very basic cases but also from what has been called ‘modus ponens’ cases of rule following. Or at least may be considered a specific variation of such a model. The working out of this possibility may help us gain a better understanding of what is at stake in competent use of a given group of concepts and, maybe, to sort out competing interpretations of rule-following in a more case-specific and less general account.

If conceptual acquisition and understanding is to be adequately understood according to the rule-following model, this model must be capable of accommodating very different kinds of concepts (classificatory, relational, functional, evaluative, mathematical concepts). It must give account of the different ways concepts relate to, and characterize, experience. We may put aside the relevance of this question either per impossibility – we cannot really separate experience from conceptual understanding – or simply acknowledge without further interest the obvious existence of differing instructions we grasp by grasping rules, and otherwise go on giving a uniform account.

But if it should be possible, and I believe it is, to attend to differences at this level it could turn out that we require differing interpretations of the very idea of rule following for different cases.

That there should be a difference between very simple, basic cases, of rule following and more complex ones, is widely acknowledged and supported by Wittgenstein’s own writings. Basic cases are those where no further non-redundant linguistic specifications or reasons can be given, besides direct illustration of how concept application goes. In other cases, while acquisition may succeed without linguistic aid by the participation in further practices, some already linguistically trained subjects could on demand give some clearing explanations. And at higher levels of complexity in language acquisition, some language users at least may be able to articulate more or less sophisticated reasons to justify conceptual application. This need not amount to an exhaustive definitive definition but just sufficiently articulated necessary conditions – all these specifications resting surely at the end in basic concepts whose meanings cannot be put down in any fixed formula –. However, when rules are to some extent linguistically articulated in such a way, we get what has been characterized as the ‘modus ponens model’ of rule following.

These distinctions allude to the grade of specification of the rules given. Although this may not be unrelated to the topic, what I was questioning before was the different ways rules connect our concepts to experience. The focus is here on the specific conceptual contents involved. Some related proposals are made, for example, by Crispin Wright’s (1992), (2002) distinction between extension-determining and extension-reflecting concepts. Extension-reflecting concepts would register “self standing properties” and therefore the possibility of getting it wrong makes sense. Even if what it is to fall under the concept is epistemically constrained (weather something falls under the concept subject to human considerations) its existence is not constitutively dependent on human responses or considerations. By extension determining cases, though, there is no sense in which truth could transcend what we would ourselves say, as what seems to us, our own impressions or responses to some experience, are part of the content we are registering. And it is the very conceptual content we have to do with that demands it to be so. In Wright words: it is a priori that best opinion determines truth. Concepts of primary qualities would fall under the extension-reflecting and concepts of secondary qualities, under the second. This last distinction, however, is supposed to have a wider application and extend to avowals, for example, and, maybe, to moral concepts.

If we try, however, to figure out an adequate understanding of the rule-following considerations in ethics, trying to specify the rules governing conceptual application, we become (as a result of considerations of content) what appears to be a variation of the modus ponens model. Weather this model fits into the extension determining schema will depend on how this is exactly formulated, but some additional distinctions will be called for on its regard.

2. Rule-Following in Ethics

According to McDowell (1981) non-cognitivist disentangled explanations of thick ethical concepts could not explain their consistent rule guided use if it were not on the basis of some value neutral feature we should be capable of recognizing and to which we would be responding to. And the problem is that it should not be possible to sort out what feature this is, what all the members of the extension have in common, without the aid of corresponding evaluative considerations.

Carefully considered there are actually two different assertions in McDowell’s just cited claim:

• (i) It should not even make sense to pick up a value neutral class equivalent to the one thick ethical concepts sort out without taking into account evaluative considerations.
• (ii) It should not be possible, once the class is constituted, to see value neutral common features among the members of the extension of a thick ethical concept.

McDowell, I believe, wants to assert both (i) and (ii), but these two claims do not necessarily have to go together. Let’s call the first the ‘generation argument’ (GA) and the second the ‘application argument’ (AA).

McDowell illustrates his general claim arguing that given a list of items (individuals, actions, etc. let us suppose) that belong to the extension of a thick concept we most probably won’t be able to tell what such items have in common. The class consider in abstraction of evaluative aspects need have “no shape”, form no kind.

The argument would reach a non-cognitivist claiming that it is at some such recognizable level of appraisal that we are to find the features we respond to. And surely if what is at issue is a model of causal reactions to an independent world, we would situate ourselves at some such level. But this does not preclude us from finding some common factor at a higher order level. There is nothing in the class of ‘communists’ or ‘lawyers’ or items used ‘to hold the door’ that could be distinguished at such a level either. The argument, thus, does not necessarily refute disentangled explanations. We may perfectly well have some such (non-recognizable) morally neutral class to which a moral value is added and a thick ethical concept applied.

This would allow an answer to AA: there is no reason why it should not be plausible to assume that it is some specific class of behaviour, say – identifiable separately and morally neutral on the first place – that guides or application of a thick ethical concept. Once the class is identified, the value is added and the concept applied.

But McDowell could still counter that an explanation saying why we should at all pick up such a class as morally relevant is missing. The apprentice who is suppose to understand how to apply the concept may this way learn to apply the term on the basis of this independently discernable class, but would not understand why such behaviours or persons are to be called morally good. He would be like a child sticking red labels to all square things, without making any more sense of this than following orders. Moreover such behaviours in virtue of which the moral evaluation is done, may not be paid much attention to if it were not for our interest in the moral classification: and this, without having to embrace McDowell’s position, supports GA. In order to pick up some morally neutral class that is to be evaluated, we need some reason to fix special attention on it.

This seems right[] without [making] AA [true], that the relevant common characteristics are there all the way long. But it is not any more open to give a disentangled account insisting on the possibility of AA, this bringing us back to a modus ponens case –always when C(x), apply MV(x); application proceeding as in our child example. If this is not convincing, what kind of further specification would a disentangling supporter need in order to explain why some type of behaviour is to be morally evaluated such and so?

What fails is an explanation telling us why this or that behaviour is to be called ‘good’. But behaviours may be good for satisfying very different goals: to be healthy, becomes acceptance in some sect, favour the Gods, the clouds or whatever. What we are looking for is something specific, what a thick moral concept is expressing is that some behaviour is morally good or bad, so what we look for is not just good but morally good. What needs explanation is

• (i) What makes some kind of behaviour morally good?
• (ii) What do we mean first of all with morally?

A somehow standard explanation would be to say that morality has to do with the relation of men to each other (and their surroundings) we want to expect from all. Substituting we obtain:

• (iii) What kind of behaviour is good with respect to what we want to expect from all in their relation to one another and their surroundings?

Following Kantian proposals some will conclude, for example, that good relative to (iii) is what would equally protect the needs and interests of the affected. It is not my purpose to enforce this particular conclusion right now, the point is that whichever conclusion we may arrive at as an answer, it will deliver themeasure, call it FM, relative to whose fulfilment some morally neutral behaviour is to be called ‘morally good’.

• (iv) A behavioural type ismorally good if they fulfil FM

This binding engine is what would mediate between the antecedent and the consequent of the modus pones model: between the class of neutral behaviours on which basis we apply a moral value and the moral evaluation. We would actually have to do with a function that working on some descriptive level yields the evaluative as a result:

• (I.) Behavioural type input b → [FM] → Evaluative (+/-) output MV
• Conceptual application C

Applying this rule we obtain:

• (1) b₁ fulfils [FM]
• (2) MV+(b₁)
• (3) C₁(b₁)

However by our rule guided application of a moral concept what we follow is the derived more simple modus ponens rule:

• (II.) Behavioural input b₁ (assumed MV+) → C₁(b)

Because it has already been calculated that characteristics b₁ fulfil [FM], it is now a priori that whichever token falls under type b₁ it is MV. In this specific rule it remains implicit that FM is fulfilled and therefore MV+. This would account for the fact of children and traditional people finding no problem in blindly applying some such thick concepts on the basis of b₁, b₂, etc. without being capable of giving further explanations of why this behaviour is morally blameable, for example. The concept includes the explanation on itself by having established as a priori the relation between b₁ and MV by means of [FM].

3. The Three Fold Model and its Implications

The presented model is what I call the three fold model. It is obtained by trying to give a more explicit and satisfactory disentangled account of our rule guided use of thick ethical concepts. The result being that there is some function such that, when it is fulfilled by some type behaviour, it qualifies it as morally good. So the content we have to do with is established by some operation that assigns by each ongoing input a given output. Thick ethical concepts result out of synthesizing some such result in a concept. Therefore, given some type of behaviour for which it has being established that it fulfils FM it is a priori that a given moral value applies to it. It is a priori determined, that whichever extension b₁ has, all its members (in virtue of some given operation) become a given value and fall as a result under some new class.

Do we have to do with extension-determined concepts then? The idea is here not that the decision of weather the concept applies depends on our best opinion because our own reactions or impressions to some behaviour should be directly decisive of the case. This would apply to response-dependent models. Here we have two different questions actually. a) Weather, in deciding if a thick concept applies, best opinion is all there is to truth. And in presuming that given b_n some value applies, the question is if we do or don’t have to do with b_n. b) Weather the calculation required to assign a value to some b_n depends on best opinion. This will depend lately on weather human needs and interests, for example, can be determinable independently of our own responses –. Both questions I shall leave open here.

On the considerations made, however, some other distinction appears to be relevant. Contrary to concepts such as ‘red’, ‘tiger’, ‘cup’ or ‘tree’ whose meaning is open to development on the way, so to speak.. Some other concepts are such that their extension is dependent upon prefixed operations and to this extent there is no development of meaning on application. Any change would require going backwards and proving the correctness of the calculations made in its establishment. If this is right, we may distinguish between open-ended and invariably prefixed rules. That this distinction is not to be put together with that between extension-determined and extension-reflecting concepts can be seen as ‘red’, for example, would be a extension-determined but open-ended (susceptible of refinement or development). The distinction does not depend on weather best opinion determines of truth, but on the determinateness of meaning itself. Three fold concepts would fall under the second category but the distinction is not necessarily restricted to them.