Shared Agency on Gilbert and deep continuity

In a number of texts and now this outstanding book, Michael Bratman postulates shared intention as an explanation of shared action. ^[1] Margaret Gilbert, also in a number of important books and articles, postulates joint commitment as an explanation of acting together. ^[2] Here, I try to bring into sharper focus the similarities and differences between Bratmanianshared intention (BSI) and Gilbertianjoint commitment (GJC). To do this in the space allowed, I must paint with a broad brush. I hope both authors forgive me for glossing over some of the details of their views.

Bratman and Gilbert appear to have more or less the same explanandum in mind. They use the same introductory examples of dancing, building, walking, talking etc. together, as well as others that make it clear that the ‘togetherness’ in question is not, and does not entail proximity. Bratman adds to his pre-theoretical sketch a contrast with strategic interaction, of which actions productive of a Nash equilibrium are perhaps a paradigm case. ^[3] Our individual actions may be mutually responsive; in addition, there may be a readily available description of what we collectively accomplish (e.g. the painting of a house) that, in other circumstances, describes an action; still, for Bratman (and I take it Gilbert also) this is not sufficient for the target phenomenon. What is necessary and sufficient is that we act in partnership or collaboration in a sense that does not rule out our having diverse and perhaps egoistic and competitive motives and goals, but which does imply that we regard each other, well…, as partners, hence as more than elements of a background of circumstances. ^[4] It is this that Bratman and Gilbert set out to explain.

Bratman and Gilbert agree also on several features of the explanans of this phenomenon.

First, that it is, to put it broadly, a shared practical commitment which is, at least in part, a creature of its parties’ wills, and which is characterized by its psychological and action-guiding role, and norms of practical rationality, which, by and large, are accepted and on that basis conformed to.

Second, that this is a plural analogue of a state of individual practical commitment. In particular it has an analogous psychological role, and is guided by analogous norms. Hence, for Bratman, just as individual intention is the answer to Wittgenstein’s question as to the difference between one’s arm rising and one’s raising of one’s arm, so “shared intention [is] central to our answer to our social analogue of Wittgenstein’s question.” ^[5] And likewise, for Gilbert, a joint commitment is “the collective analogue of a personal commitment”, ^[6] where, as she sees it:

A personal decision, intention, and so on, gives rise to what may be called a personal commitment… I unilaterally create my personal commitments. I decide, intend… thereby committing myself. ^[7]

Third, that while the postulated shared state is one that some human agents are in, or have, hence, somewhat trivially, a state that has them as its subject or subjects, it does not in general depend for its existence on any psychological subject or subject of experience, further to these individual human agents. Nor does it modify any mind, other than their minds. (I here endorse Bratman’s reading of Gilbert’s ‘plural subject’ as meaning a plurality of jointly committed subjects, rather than any further subject.) ^[8] Does this weaken the analogy between individual and shared practical commitments? If it does, Bratman at least can explain why. For on his view, while an individual practical commitment (intention or plan) is embedded in a complex mental web that marks the boundaries of its subject, a shared practical commitment is not generally so embedded, so there is no evidence that it is proper to any plural or composite psychological subject. ^[9]

Fourth and finally, that the (generally observed) norms of BSI and GJC defeasibly and non-morally require of the parties that they find a way to work with (not just around) each other to realize their common goal, and where this is not possible, to modify or abandon it. In particular, they require, within limits, persistence through difficulty (e.g. flagging motivation in oneself or others), flexibility regarding one’s contribution or role, and a disposition to help the other parties.

Granted, Bratman thinks of these norms as ‘demands’ or ‘requirements’ whereas Gilbert thinks of them as ‘obligations’. But she is not firmly tied to the word. ^[10] Granted too, they give divergent explanations of these norms. For Bratman, the norms of stability, agglomeration, consistency and means-end coherence applying to individual intentions, and the relations between such intentions within a BSI, render a BSI subject to analogous emergent social norms. For Gilbert, it is simply that the parties to a GJC, qua its joint creator, owe themselves and hence each other its fulfilment (or failing that, its recision). ^[11] For her it follows that each has a special standing to demand norm-compliance from the others, to rebuke them for non-compliance, and a special vulnerability to betrayal by non-compliance. There are no such special standings in Bratman’s theory (although he allows that the parties to a BSI may be vulnerable to the frustratation of expectations it induces). ^[12]

Despite these theoretical differences, the impact of a GJC obligation is more or less the same as that of a BSI demand. It will be felt as a defeasible non-moral practical requirement to, as I have said, find a way of working with others. In particular, while it may appear that the GJC norms are more stringent, the real difference regards their escapability, and this is not, at bottom, a normative difference, but rather a difference in the persistence conditions of BSIs and GJCs. For while a BSI may have a default inertia, any party can annihilate it at will, simply by ceasing to hold their component intention. By contrast, agents can only annihilate a GJC by agreement (or else by fulfilling it). Hence, as the Gilbertian state is harder to wipe out, so its demands are too. Still, this is not at bottom a distinction between norms applying to BSIs and GJCs, but between their persistence conditions. ^[13]

In the light of this discussion of the similarities, I want now to assess Bratman’s main charge against Gilbert. It is that she fails to see the “deep continuity” between individual and shared agency. ^[14] By that he means, not that shared agency isn’t a step up for agents, onto- and phylo-genetically, but that it shouldn’t be a step up for theorists. In particular, we theorists need not introduce fundamentally new conceptual, metaphysical and normative practical elements to understand it. ^[15] By this I take it he means that we need not posit new practical concepts, kinds or norms. Gilbert, in his view, commits the triple sin of positing new elements of all three varieties. ^[16]

My discussion of the “new norm” charge will be brief. Begin by distinguishing two senses of ‘norm’ (or, more clearly, ‘requirement’ or ‘demand’) viz. the thing that is required and the thing that requires it. In the former sense, if Gilbert introduces a new norm, then Bratman is vulnerable to a tu quoque. For, as argued above, a BSI requires essentially the same things of its agents as a GJC. In the latter sense of ‘norm’, we may grant that Gilbert introduces norms that are discontinuous with those of individual agency as theorized byBratman. But they are not discontinuous with those of individual agency as theorized by Gilbert. For her, any creator(s) of a practical commitment, be they singular or plural, owes themselves its fulfilment (or failing that its recision) and hence has a special standing to demand norm-compliance of themselves, to rebuke themselves for non-compliance, and has a special vulnerability to betrayal by non-compliance. ^[17] There is then the deepest of continuities here.

The “new concept” and “new kind” charges also threaten a tu quoque. For Bratman introduces a new practical concept and a new practical kind: shared intention. I will focus here on the putative kind, not the concept, but I expect that most of what I say carries over. In particular, I expect that my talk of kind exemplification is roughly translatable into talk of concept satisfaction.

Now I am the first to grant that one could postulate shared intention without introducing a new kind. For intention is not a new kind: to postulate that it is jointly as well as singly exemplified by agents would then be to postulate shared intention without introducing any new kind. What would be new would be, not the kind, but one’s view as to the manner of its exemplification. ^[18]

But a moment’s reflection on what it would be to postulate that intention is jointly exemplified reveals that this is not Bratman’s postulation. For ‘intention’ is a folk-theoretic term, implicitly defined by its role in folk psychological theory. Now imagine an axiomatization of that theory. The axioms are likely to include generalizations (platitudes perhaps) ^[19] of the following forms,

• If an intention that p is had by s, then _is had by s

• If _is had by s, then an intention that p is had by s

where such gaps are filled by terms for psychological states (e.g. a belief that p is possible). The details do not matter. The point is just that ‘intention’ is defined via these sorts of generalizations concerning relations between psychological states (perhaps relations to stimuli and behaviour also). Now, this is a definition of intention as singly had by individuals, for ‘shared intention’ is a theorist’s term with no echo in folk platitudes or theory. Hence these axioms carry an implicit restriction on the value of s, viz. that it is singular, not plural. To postulate that intention is jointly exemplified would, then, be to lift that restriction. For that reason, this postulate would imply that a state such as belief is sometimes had by a plurality; indeed (as belief would also be defined by similar axioms, likewise desire etc.) it would imply that there is as complex a web of states at the jointly held level as there is as at the singly held level. But this is precisely what Bratman denies. So Bratman does not hold that an attribution of a shared intention entails that an intention in the ordinary folk-psychological sense is had by a plurality. But then he is introducing a new kind, shared intention.

Bratman may reply that, in his mouth, ‘intention’ is defined, not by folk theory, but by his planning theory of individual agency, and that it is intention so defined that he postulates to be jointly as well as singly exemplified. Were it so, his theory of individual agency would introduce a new kind, but his theory of shared agency would not introduce any additional kind. But I doubt that this is his view. His pioneering monograph on (individual) intention beings with the undertaking – quoted in Shared Agency– that it will “take seriously future-directed intentions” – the context being a then dominant philosophical tradition that did not. ^[20] That sounds a lot like an undertaking to make a study of an ­existing kind. No doubt Bratman sought, in that book, to make explicit, precisify and supplement the folk theory of intention. But it remains a study of intention, and not another thing.

By contrast, Shared Agency is not a study of shared intention, because it is not a study of anything that partakes of a holistic subject-grounded web of attitudes. Granted, the postulated phenomenon is intention-like: it settles practical issues, coordinates trans-temporal action and planning, etc. for agents taken jointly, much as intention does for agents taken singly. It may even be as intention-like as is nomically possible, given the lack of a minded subject. But it is not intention.

One option for Bratman, at this point, would be to maintain that he means to eliminate shared intention. His postulate is the condition that he describes as sufficient for shared intention: roughly, circumstances in which it is common knowledge among parties that each intends that they jointly do something by way of these intentions (which depend for their persistence on each other) and by way of mutual responsiveness of sub-intention and action. ^[21] This introduces no new kinds at all. His view may then be that it is useful to call such circumstances sufficient for “shared intention”, as this serves a heuristic purpose (it was a handy label for the target of our search for an explanans of shared action) and a presentational purpose (it draws the reader’s attention to the intention-like roles played by the phenomenon). But it serves no explanatory purpose: it explains nothing that is not already explained by the postulate itself. Hence shared intention is eliminable. This would respect the continuity constraint, but at a cost: shared intention as such becomes an idle wheel in the theory. If this is how Bratman respects his own continuity constraint, I’d like to hear him say so.

More likely, I suspect, is that Bratman will concede that shared intention is a new kind, but maintain that its introduction nevertheless respects the continuity constraint, as it is not “fundamentally new” and contains no new “elements”, because of the sufficient condition given.

This manoeuvre raises the question of whether the provision of a merely ­sufficient condition that is thus reductive shows that shared intention contains no new elements. Strictly speaking, it shows only that it is not necessary that it contain them, not that it is necessary, or even actual, that it not contain them. But Bratman’s Ockhamite idea is that if we can – as it were, from our armchair laboratories – construct shared intention using familiar, existing resources, we can presume that even if nature doesn’t or doesn’t always find that way, it won’t find a less ontologically economical way. ^[22]

The more serious dialectical problem faced by this manoeuvre is that Gilbert also offers a sufficient condition for her postulate, GJC, which contains no new elements. It is simply that the parties agree to do something. ^[23] (Bratman makes no mention of this: he notes that for Gilbert, agreement is sufficient for shared intention, also that open expressions of readiness for joint commitment are, for Gilbert, necessary and sufficient for GJC, but not that agreement is sufficient for GJC). Granted, agreement may be new to Bratman’s theory of individual agency. Granted, there may be good questions as to how to understand it. But it is scarcely a new postulate.

The point here is not that agreement is a typical or paradigmatic sufficient condition for GJC, only that, like Bratman’s sufficient condition for BSI, it is an ontologically conservative one. So to the extent that Bratman’s condition justifies a “no fundamentally new kinds” claim, Gilbert’s does too.

Granted, there is a difference between the sufficient conditions offered by these writers. In fact it is, I suspect, one that, at bottom, motivates Bratman’s charge of discontinuity. While Bratman offers a constitutively sufficient condition, Gilbert offers a causally sufficient one. That is, while he purports to tell us what BSI consists in or is realized by, she purports to tell us what triggers GJC.

This, I think, reflects an interesting and largely unacknowledged difference in these writers’ theoretical frameworks. For while BSI is not a psychological state or attitude, Bratman conceives of it as a persisting presence with a persisting power to guide action and planning: in this respect he has a quasi-psychological conception of it. By contrast, GJC appears to be an upshot state (like that of having promised) which consists in a prior act (of making a joint commitment, by openly expressing readinesses for joint commitment etc.) and a persisting absence (of both recision and fulfilment of this commitment). That’s not to deny that a GJC has causal consequences, it is just that, like a state of having promised, it effects them indirectly, as mediated by beliefs and memory regarding it.

Given these divergent theoretical frameworks, it is no objection to Gilbert that she does not give a constitutively sufficient condition for GJC. She is not working with an implicit ontology of roles and realizers, but one of undertakings and grounds. (No one has ever objected to the absence of a condition constitutively sufficient for having promised.) To be entitled to an Ockhamite claim akin to Bratman’s she needs a condition sufficient to ground an act of joint commitment, and to cause a state of joint commitment, and which introduces no new kinds. And this she gives: agreement.

It is easy to overlook the point, if one dwells as Bratman does on Gilbert’s assertion that “obligations with corresponding entitlements inhere in any joint commitment”. ^[24] That may suggest that GJCS are somehow “made of” (perhaps “new”) norms or demands. I do not think this is Gilbert’s view. GJCs are upshot states that owe their existence to a prior act, and a persisting absence: they aren’t “made of” anything. Demands “inhere” in them by being essential to them. Perhaps they inhere in a second way, one which marks a difference with BSI. Perhaps – and this is speculative: I can find no “smoking gun” passage in Gilbert – demands are inherent in that they are chosen as such (at least conditionally on others choosing likewise) at the point of joint commitment. They need not be chosen as a final end, being more commonly chosen as a means to the coordination of joint activity. Furthermore, the choice can be a reluctant one, as when one has no better option. ^[25] Still, they are never an unintended side effect of acting with others. By contrast, BSI demands need be no more than a foreseen or foreseeable consequence of one’s choices: they are not chosen as such. ^[26]

However that may be, BSI and GJC are on a par with respect to “deep continuity”. Both are new kinds (unless Bratman pursues the eliminativist option that I have described). Yet both can be accounted for by appeal only to existing kinds, as we can describe a sufficient condition that appeals only to existing kinds, and make an Ockhamite presumption that nature finds no more complex way. So it is not fair for Bratman to contrast BSI and GJC by calling the latter “irreducible”, “primitive” and “unanalyzable”. ^[27] And as the two postulates are on a par with respect to ontic parsimony, then the advantage in ideological parsimony that Gilbert’s sufficient condition has over Bratman’s is far from negligible.

I conclude by drawing attention to one theoretical advantage of GJC over BSI.

It is a drearily familiar fact that, even in “modest sociality” where institutional or asymmetric authority relations are absent, one can ϕ in partnership with others, freely ^[28] and rationally complying with, but not embracing a shared commitment to ϕ. In particular, one may strongly disapprove of the idea of our ϕ-ing, want, hope and expect that we will fail to ϕ (perhaps because one is confident that others will not in the end do their parts) and retain a private contingency plan (which perhaps one never has the opportunity to put into practice) to thwart the action if necessary. ^[29] Now, Bratman is careful to allow for intention that fails to conform to one’s evaluative rankings due to akrasia, lack of confidence in one’s rankings, or a need to coordinate with others. ^[30] I doubt, however, that he can allow for intention, still less shared intention, that we ϕ, in the circumstances described. But the quasi-contractual notion of GJC easily makes room for this sort of phenomenon. We may have agreed that we ϕ (which I might have had all sorts of incentives for doing) even though I strongly disapprove of the idea, want, hope and expect that the action will not come off, and have a contingency plan of the sort described. (Interestingly, and despite the tight connection between decision and intention, the same looks to be true of having jointly decided that we ϕ.) Simply put, one can sign up or volunteer for a plan of joint action, without ever embracing it.

Let me end by saying that I adore Bratman’s book, hugely admire his contribution to the theory of agency (individual and shared) and am delighted by this opportunity to discuss his work.