Bradley's Regress and a Problem in Action Theory

Abstract

There is a problem which often arises during the course of various discussions in action theory and related fields about how exactly we are to characterise the relation which obtains between an agent and her (token) actions. An agent is a particular individual; it is often assumed that any token action of hers must be another.¹ But what is the relation between these two particulars, when the agent is the agent of the action in question? Obviously, one asymmetric relation between them is this: the agent, S, is the agent of the action, A. But is there a further relation between agent and action in virtue of which it is correct to say that A is S's action? The idea that agency must be reducible to something assumed to be more basic, such as causation, has sometimes tempted philosophers to think so – and there are also fairly common locutions which can make it seem as though S's being the agent of an action, A might hold in virtue of another, perhaps more basic relationship – ‘execution’ or ‘performance’, or simply ‘doing’, for example. But as I shall try to show below, it is not at all easy to make any of these ideas work.

In order to have a handy label for the problem which attaches to this search for a relation to undergird propositions of the form ‘S is the agent of A', I am going to call it the ‘agent-action problem’. The problem has rarely been acknowledged as a general difficulty – although it gives rise to various sub-problems, which have been often enough remarked upon.² Moreover, even when the sub-problems are observed, they are sometimes noted merely as passing curiosities which perhaps constitute nothing more than minor linguistic inconveniences to the philosopher of action. In a way, then, neither the sub-problems nor the fundamental problem which in my view underlies the sub-problems has really received any serious, sustained scrutiny of a properly wide-ranging sort. In this paper, I want to suggest, however, that it deserves such scrutiny – and that a failure properly to get to grips with the general form of the problem is indicative of philosophy of action's failure to get a decent ontological understanding of its own subject matter. This failure, I believe, is connected to some of the puzzles in which philosophers find themselves embroiled, with respect to such issues as whether agents are causes of their actions³; whether the agent ‘disappears' in a problematic way from certain pictures of what action involves⁴; and which physical events, precisely, compose or constitute our actions.⁵ I want to argue that once we understand the true source of the agent-action problem, it can be seen that the problem is related in certain interesting ways to the philosophical difficulty which has come to be known as ‘Bradley's Regress'. The range of options for responding to it can, I think, therefore be usefully illuminated by reflecting on those that have been floated for dealing with that regress. Precisely parallel solutions to the two issues, I shall argue, are not necessarily available – because there are important differences between the two problems, differences relating to the fact that Bradley's Regress generates only a possibly tolerable infinity of universals, whereas parallel moves in the agent-action case generate more problematic-looking infinities of particulars. Nevertheless, some of the insights yielded by thinking about the source of Bradley's Regress seem highly relevant also to the agent-action case, and towards the end of the paper, I will make some suggestions about where the most hopeful resources for solving the agent-action problem may lie.

In the first section of the paper, I shall try to explain what the agent-action problem is, and why the difficulty cannot simply be regarded as a passing curiosity. In the second, I shall introduce Bradley's Regress and give some reasons for thinking that the agent-action problem is a version of an interestingly similar difficulty. Then, in the final section of the paper, I shall briefly outline three approaches that have been adopted to Bradley's Regress and consider what their analogues might be in the action case, arguing in favour of a version of the third solution. I shall also offer some brief reflections concerning possible lessons for philosophy of action.

What is the agent-action problem? The issue is that none of the ways in which we might initially be tempted to talk about the relation we bear to our own actions really seems to pass philosophical muster. Looking to natural language for guidance, a natural first recourse might be the generic verb ‘do’. Actions, it is often supposed, are things we do – and so they are things, one might think, to which we stand in the relation of doing – the relation that one might think could be expressed by the open sentence ‘S does/did A'. We certainly make frequent reference to people doing things and speak of the things we have done – and it might seem natural to suppose that these things that we do must be our actions. But as Hornsby forcefully argued in her (1980), these things we say that we do certainly cannot be token actions. For the purposes of talking about individual actions in English, philosophers usually have recourse to so-called ‘perfect nominals' – expressions of the form ‘my raising of my arm’; ‘Caesar's crossing of the Rubicon’; ‘Oswald's shooting of Kennedy’. These are derived from what I shall sweepingly (and unsatisfactorily) call ‘action sentences',⁶ (which, for present purposes, I shall take to be basic sentences of the form SUBJ + VERB ± OBJ – sentences such as ‘I raised my arm’, ‘Caesar crossed the Rubicon’, ‘Oswald shot Kennedy’) by means of a recipe which we might describe roughly thus: nominalise the verb using a gerundive construction, convert the subject term into an appropriate possessive adjective; and insert an appropriate preposition (usually ‘of') to connect with the object of the sentence, if there is one.⁷ By means of this recipe we can derive nominal expressions like those above, expressions of the sort which seem to refer to particular actions. But these sorts of expressions cannot be substituted for ‘A' in the ‘S does/did A' schema. kinds I did not do my raising of my arm; Caesar did not do his crossing of the Rubicon; and Oswald did not do his shooting of Kennedy. These expressions are clearly ungrammatical. As Hornsby notes, we ought to think of ‘do’, where we use it in such contexts as ‘What did you do?’; ‘She did something’, etc. as a schematic verb.⁸ What it stands in for, schematically, though, is never an act-token, but rather what one might call an‘act-type’ – what philosophers often call an act, for the purposes of disambiguation. ‘Raise one's arm’, ‘cross the Rubicon’, ‘cross a river, ‘shoot someone’, ‘shoot Kennedy’ are all acts, on this understanding of the word ‘act’. Hornsby has argued that it is these latter things, the acts and not the individual actions, which are the things we do.⁹ Some of the things I have done this morning, for instance, are the following: make my bed, eat my breakfast, catch the train. But I have not done my making of my bed, my eating of my breakfast and my catching of the train. If one ever offers any kind of singular term, rather than a whole proposition, as an answer to the question ‘What did you do?’, one would give an infinitival construction – ‘shut the gate’; ‘poison the chalice’; ‘light the fire’. Or consider the following example: ‘What Caesar was going to do before he died was alter his will’. (Wiggins, 1985: 285). Here it is obvious that what Caesar was going to do (the thing he was going to do) simply cannot be an individual action because the expression ‘What Caesar was going to do’ is not robbed of a reference by the fact that Caesar's assassination prevented any such individual will-rewriting action from occurring. There is still something that Caesar was going to do, even if he never did it. And if he had done it, that same something would have become a thing that Caesar had done. But that means that if we take grammatical structure seriously, the things people do have to be types of thing; they cannot be individual actions.

Moreover, because ‘do’ is a schematic verb in such contexts, it is not even true that we bear the doing relation to these acts, let alone that we stand in that relation to our token actions. For example ‘I do raise my arm’ is ungrammatical (unless the ‘do’ is read as an auxiliary, intended merely to give emphasis to the verb ‘raise’ on the habitual reading generally indicated by the present tense in English, e.g. ‘I do raise my arm when I have a question to ask!’). ‘Do’ rather stands in for these other verb-phrases, replacing them when we want to speak in a general way about acting (e.g. ‘I did lots of things this morning – I made my bed, cleaned the bathroom and prepared lunch’). Here, ‘did lots of things' is generic and ‘made my bed’, ‘cleaned the bathroom’ and ‘made lunch’ offer some of the specifications of the relevant ‘things done’. So even here, ‘do’ is not itself a relation which the agent bears to anything. It does not relate agent to act-type any more than it relates agent to act-token. It does not represent a relation, in fact, at all.

Another common recourse of the philosopher looking to speak about the relation between agents and actions is the multi-purpose philosophical workhorse, ‘perform’. Actions, it is said, are things we perform. Unlike ‘do’, ‘perform’ can indeed represent a relation – as it does in such sentences as these: ‘The choir performed Handel's Messiah’; ‘I performed the Fosbury Flop’; ‘Georgie performed the Heimlich manoeuvre’. But here, the second of the relata in each two-place relation (‘Handel's Messiah’, ‘the Fosbury Flop’, and so on) are not particular actions, but rather such things as pieces of music, or types of athletic or other kinds of feat. Moreover, the sentences which express these relations are themselves action sentences, and hence can produce the perfect nominals by means of the recipe according to which singular terms for individual actions are generally delivered; these particular sentences, for instance, deliver ‘the choir's performing/performance of the Messiah’; ‘my performing of the Fosbury Flop’ and ‘her performing of the Heimlich Manoeuvre’, once our recipe for forming purportedly referring expressions for individual actions is applied to the original sentences. But then, if all individual actions are performed by their agents, as is being suggested by our current proposal, these performings will have themselves to be things to which agents stand in a performance relation. The performings will have to be performed. And now, what are we to say in general about agents' performings of the performings of works, or feats, which constitute their individual actions? Are these performings of performings themselves actions? If they are, then surely we are off on a regress – since, being actions, they will themselves have to be things which are performed – and we now have performings of performings of performings to acknowledge. But if they are not, we need a principled account of why they are not – a principled account of why a sentence of the form “S performed X”, where ‘X' names some kind of feat/accomplishment/musical work, etc., is not precisely the type of sentence – personal subject, active verb in perfective aspect, etc. – that should yield reference to an individual action when nominalised according to our recipe. And it is not easy to see what the answer would be, short, that is, of admitting that actions ought not in fact to be regarded as things to which we stand in the relation of performance, which puts paid to the hopes of this particular strategy to elucidate the relation between agents and actions.

It might be pointed out that one can also use the indefinite article in connection with the verb ‘perform’ – for example, one can perform such things as ‘a cartwheel’ or ‘a three-point turn’. And a sentence such as ‘I performed a cartwheel’ might look at first grammatically similar to ‘I ate a cake’ and therefore amenable to similar logical treatment of the Ǝx(aRx) kind. But when I perform a cartwheel, it is evident that the cartwheel I perform doesn't exist independently of my performing it, in the way that the cake exists, independently of (and prior to) my eating it. My performing the cartwheel just is the cartwheel I perform; my three-point turn just is my performing of it. Once again, no genuine relation between performer and gymnastic or vehicular manoeuvre is in the offing here. To adapt a phrase used by John Hyman, in connection with the very similar, but non-action-involving constructions ‘live a (good) life’ and ‘die a peaceful death’, the construction ‘perform an F' is here “a syntactic expedient” and “… the verb does not express a genuine relation. On the contrary, it refers to one of the very things that it appears, syntactically, to relate” (2015: 56). ‘Perform’, then, does not look like a hopeful candidate to be the relation which holds between agents and their actions – ‘perform an action’, like ‘die a death’, is merely a syntactic variant on the simple use of an active verb (such as ‘cartwheeled’, for example) and the verb ‘perform’ in this kind of context in fact refers not to a relation borne by the agent to her action, but rather simply indicates the act itself.¹⁰

A third possibility which has been widely canvassed – and which indeed now constitutes a distinctive position on the libertarian side of the free will debate - is that actions are caused by their agents; that the relation between agents and their actions is causation. But this view has been subjected to fairly devastating criticism also, perhaps the most famous critique coming from Davidson (1971: 52), in the shape of a dilemma. If agents cause their actions, then we must answer the question how they do so – whether by acting, or not. If we say the former, then clearly a regress looms, one which looks rather similar to the regress we ran into in connection with the verb ‘perform’, above – we must cause our causings of our actions, and cause our causings of our causings of them, and so on. If we say the latter, though, we are left with the arguably quite unintelligible suggestion that agents cause their actions by some means other than the means in terms of which we always understand the intentional production of results by agents – viz., action itself. How on earth might an agent intentionally cause an action in such a way that her intentionally causing that action did not itself amount to an action?

So: we appear to have arrived at the following situation: we do not ‘do’ our actions, because the verb ‘do’ does not express a relation at all, once the contexts which initially suggested it are properly examined; and neither is ‘perform’, in at least some of its uses. ‘Perform’ can be a relation – but we cannot allow either that we perform particular actions or that we cause them, without generating dilemmas, one horn of which is an unappealing regress, and the other of which appears to be a baffling and unjustified refusal to embark on it. Perhaps there are other alternatives to the verbs ‘do’, ‘perform’ and ‘cause’ – but I suspect that, if there are, they are likely merely to be near-synonyms of one or other of the three we have so far canvassed. And so the question arises: if none of these natural-language verbs will do, what is the relation between an agent and her action? What is the ‘R' such that a subject, S R-s A' when she acts, and when A is her action? And if for some reason we are unable to return a sensible answer to this question, if there is no such ‘R', how is that inability to be diagnosed? Why does nothing seem to work? In the next section, I want briefly to outline the issue which has come to be known as ‘Bradley's Regress', after F.H. Bradley (Bradley, 1897: Chs. II and III) before going on to suggest that this regress will cast light on the source of our current difficulty.

So far, we have been asking the question what the relation is between an agent and her action. I want now to turn for a moment to a rather similar question that has received much more in the way of explicit philosophical attention, namely, what is the relation between a thing and its properties? Some possible answers might include the homely and everyday ‘having’ or ‘possessing’ – things have their properties; the more technical-sounding ‘instantiating’ or ‘exemplifying’, much beloved of philosophers; or the metaphysically-laden ‘supporting’ (as in the scholastic doctrine that ‘substance supports accidents'). But whichever of these answers we return, we will face Bradley's Regress.

One might think that although Bradley's Regress can be applied straightforwardly to many of the kinds of sentences which have been called ‘action sentences' (as just shown above with the example of ‘Brutus stabbed Caesar') it is not obvious that the regress bears directly on the problem I have been calling the agent-action problem at all. For the regress just generated develops in response to a relational answer to the question what the relation is between the subject (and object) of a sentence and the property or relation which its predicative portion purports to stand for. The issue it raises is therefore an issue about the relation between a thing, on the one hand, and some kind of universal (either a property or a relation) on the other, not an issue about the relation between those subjects and any particular entities, such as token actions). This is perfectly true. But my claim is that nevertheless, close parallels are discernible between the thinking that leads to Bradley's Regress, and that which leads to our problems about the relation between agent and action. The main parallel is that in both cases, the attempt to explicate what we take to be the metaphysical commitments of an ordinary sentence by introducing (i) some form of reification in connection with its predicative component and (ii) a relation-word (‘instantiates', on the one hand, or ‘performs' or ‘causes' on the other) to reconnect the subject of the sentence with the new entity which our reification has introduced, quickly lands us in regressive territory. We cannot say anything about the relation between the subject of a simple predicative sentence and the property ascribed by that sentence without inviting regress; likewise, we cannot say anything about the relation between the subject of an action sentence and either (i) the property/relation ascribed by its predicate to that subject or (ii) the token action whose existence might be supposed to be entailed by that action predication without inviting regress. This seems to me to constitute a striking parallel, worthy of further investigation.

It must be conceded that there are important differences between the two kinds of regress – the one involving universals and the one involving particulars. One difference is that at every stage of the agent-action regress, the tie that binds agent to action is the same one – that is, the tie represented by the verbs ‘perform’ or ‘cause’. In the pure Bradleian regress, on the other hand, we get a new relation at every stage, because at each stage, a new argument-place is generated, ensuring a hierarchical structure, with new relations constantly appearing as we proceed along the regressive pathway. All the additional complexity which is involved in Bradley's regress is packed instead, in the case of the action regress, into the proliferation of definite descriptions of the relevant actions – where we have not just action A, but Jess's performing of action A, and then Jess's performing of her performing of action A, and so on. But despite the fact that the regresses generated in these two cases are formally different, it seems clear that they are two versions of what is ultimately the same problem – the difficulty one always meets in trying to reify (in order to speak about) the connections between the elements of propositions. As soon as one introduces names for the items referred to or otherwise introduced by (e.g. quantified over by) the predicative parts of sentences, whether those are properties and relations, on the one hand, or particular actions, on the other, one loses what has sometimes been called the unity of the proposition, and what one is left with seems to be a mere list of names – which require to be reconnected by means of some further glue – and naturally, we find ourselves reaching for finite verbs in order to provide it. But the trouble is that these verbs can then always be nominalised in their turn, so that the predicate or relation they express can become an object of reference. The process thus described can be repeated to infinity.

What do I mean, exactly, when I say that the problems arise in the course of our making ‘an attempt to explicate what we take to be the metaphysical commitments of an ordinary sentence’? The thought that the metaphysical commitments of ordinary sentences need explicating in the first place is usually motivated by the widespread intuition that ‘truth depends on being’,¹¹ which often pushes us, in a philosophical context, to feel we cannot rest content with the mere fact that a particular predication is true; we feel as metaphysicians that we must explain what entities in the world serve to make it true. In the case of Bradley's Regress, for example, we might start (in the most simple monadic case) with a simple predicative sentence such as ‘This ball is blue’, say. Then, in trying to understand how truth might possibly depend on being, in the case of such a sentence, we might nominalise the predicative part, so that we have something to refer to – and say that what the sentence says is that the ball has the property of blueness, or instantiates blueness, or some such thing. This initial move is what now invites the regress. For our new sentence (e.g. ‘the ball has the property of blueness') is itself a predication – and if the metaphysical commitments of predications can always be cashed out by means of predicate-nominalisation, we end up with ‘the ball has the property of having the property of blueness'. And this sentence is in turn a predication …. and so on. We are off on the regress to infinity that Bradley observed.

In the case of action, a similar situation arises when we try to think about what might make an action sentence true, but the situation is more complicated here because there is more controversy about what the relevant ontological commitments should be taken to be.¹² One view would be that the ontological resources must at least include (though they may also outrun) the very same resources as required for the standard adjectival Bradley cases (such as ‘This ball is blue’) – that what a sentence such as e.g. ‘Jess turned on the light’ does is ascribe a property to Jess – a property that she is said by the sentence to possess at least one particular time - the property of turning on the light. Something like this is, I think, basically Kim's view about how to understand action sentences: for him, events (including actions) are exemplifications by a substance of a property at a time¹³ - and what an action sentence says is that such an exemplification exists. A second kind of view might be that we should think of ‘Jess turned on the light’ as instead being a relational sentence – one which says that Jess bore the ‘turning on’ relation to the light at some past time. But those very many philosophers who believe in the existence of individual, token actions may want to insist that even if there is nothing strictly wrong with either of the above construals, they need further analysis and that if, for example, Jess bore the ‘turning on’ relation to the light, that is because there was at some point in the past a particular action of a certain kind.¹⁴ Just as it is often assumed in philosophy of mind that where someone ‘believes that p’ that must be in virtue of the presence of a ‘token belief-state’, or some such entity, so it is likewise assumed in philosophy of action that particular actions are required for the understanding of the ontological grounding of action sentences. There is in my view somewhat more to be said for the second of these assumptions than the first (Steward 1997). But even the latter assumption needs to be treated with extreme caution.

Once we are at this point in the dialectic and have embraced the idea that we should embrace the category of particular actions, we will want to be able to refer to the action which is the preferred truthmaker(s) for our original sentence, and we might effect a kind of nominalisation in order to be able to do so. We might talk, for example, of ‘Jess's turning on of the light’ and say that in order for the sentence to be true, Jess has to have performed or caused this action. But this claim now generates a regress, just as the move to nominalisation of the predicate did in the Bradley case. For the new sentences we have now produced by this means (e.g. ‘Jess performed her turning on of the light’) are themselves action sentences, and so if the metaphysical commitments of action predications always require to be cashed out by means of the kind of nominalisation which yields referring expressions for actions, we end up with ‘Jess performed her performing of her turning on of the light’ – and we are off on a regress that is surely very similar to the regress to which Bradley drew our attention. Indeed, it is rather striking that the three verbs for talking about the relation between agents and their actions which have mostly recommended themselves in that case have analogues in the parallel choices that exist about how to speak of the relations between things and their properties. The homely and everyday ‘do’ is a little like the homely and everyday ‘have’, an everyday verb which we attempt to put here to very generic use; I do actions, I have properties. The quasi-technical ‘perform’ is, I suggest, rather like such philosophers' technical verbs as ‘instantiate’ and ‘exemplify’ – a term that is not altogether without its ordinary uses, but which is employed by philosophers in a rather non-ordinary way, for specifically philosophical purposes. And then, finally, the more metaphysical sounding ‘cause’ can be compared to the relation of ‘support’, as invoked in the doctrine that a substance ‘supports' its accidents, in that it is an attempt to give a metaphysical account of what the relationship is, as it were – a more robust account which purports to provide more information about what exactly the relation in question must be like. But whether or not the similarities in the linguistic resources we draw upon in each case to connect subject and nominalised verb strike one as suggestive, there is a clear parallel between the agent-action problem and the difficulties posed by Bradley's regress. In both cases, if the initial move to introduce (i) reification and (ii) a relation between subject and reified entity, has a sensible motivation in the first place, then it looks as though the same move will demand to be repeated on the very same motivational grounds as we began with – generating either an infinite hierarchy of instantiation relations or performings of performings and causings of causings, ad infinitum.

In the following, and final section of the paper, I want to turn to consider some of the responses people have made to Bradley's Regress, and to consider whether any promises us a satisfying answer to the agent-action problem.

What should we think of Bradley's Regress? In this final section of the paper, I shall survey three prominent responses to the regress and consider what there is to be said for their analogues in the agent-action case. One interesting thing, it seems to me, is that one response that seems to me potentially appealing in the case of the Bradley regress looks much less appealing as a response to the agent-action problem; it may be, therefore, that although the two issues are related, the differences between them demand different solutions.