Group Agency,Responsibility,and Control

Introduction

The present discussion takes Christian List and Philip Pettit’s stimulating new book Group Agency: The Possibility, Design and Status of Corporate Agents as its point of departure. In the first part of that book, List and Pettit argue for the possibility and existence of group agents—that is, groups that have unity, exhibit rationality, and have a certain autonomy with respect to their individual members. In chapter 7, List and Pettit ask whether such group agents are fit to be held responsible for, and in particular whether and how they can be in control of, their actions. These are the questions I address in this article. While my discussion is presented in the context of List and Pettit’s nonreductive account of group agency, its purview is wider since it addresses the morphology of relations between an individual’s control in agency and that individual’s social and institutional context.

To ascribe responsibility for an action to a group, the group must be in control of the action in question. According to List and Pettit, there are three necessary—and jointly sufficient—conditions for a group agent being fit to be held responsible (List and Pettit 2011, 204).

Group agents, List and Pettit argue, can fulfill all these conditions, and consequently, they can be fit to be held responsible. As we will see, however, it is not clear how such corporate control at the group level can be combined with individual control of the same action. Indeed, List and Pettit formulate and address an argument to the effect that corporate and individual control cannot combine in this way. I call this the no combined control argument. The argument poses a challenge for their account of group agency, to which they respond by arguing that the same action can be under causal control of both a group agent and one or more individuals. They derive their suggested solution from established accounts of multilevel causation.

In part 1 of this article, I argue that List and Pettit’s solution works in only a restricted range of cases, leaving out interesting cases where group activity does not depend on specific individuals. In part 2, I discuss relevant notions of control and develop the notion of structural control. In part 3, I reject the no combined control argument since it equivocates between causal and structural control. I also argue for an act-dualism between corporate and individual agency, which complements my positive account of combined control.

1. The Combined Control Problem

List and Pettit (2011) start the discussion of the third requirement on group agents being fit to be held responsible, by claiming that

the notion of control needs analysis in any full theory of agency, but since the issue arises with individual agency as much as with group agency, we need not provide that analysis here. The challenge for us is not to explain what an agent’s control is but rather to show that there is no special reason why such control, whatever it involves, shouldn’t be instantiated in a group agent as much as in an individual. (207)

As will become clear, however, we need an analysis because the notion of corporate responsibility depends crucially on specific issues regarding the notion of control. It depends in particular on structural features of control; I discuss these in detail later. ¹ In a footnote, List and Pettit explicitly commit to causal control as the pertinent notion for present concerns:

One of us, Philip Pettit, believes that exercising control in the relevant way requires more than just exercising causal control. . . . As a group agent, we are only committed to the weaker view, that exercising control involves exercising causal control. (207n110; my italics)

Due to well-known problems of causal exclusion, it is not clear how a corporate agent and an individual agent can be in causal control of the very same action. ² How can corporate agents be in control without undermining that of the individuals and vice versa? This is the combined control problem. List and Pettit discuss and propose a solution to this problem, based on the assumption that the relevant notion of control is (or involves) causal control. This assumption makes it natural to argue for combined control by analogy to cases of multilevel causation. I look at the notion of causal control in section 1 of this article, postponing the discussion of structural control until section 2.1.

Let us take a look at List and Pettit’s helpful statement of the problem, which they provide in the form of an argument (List and Pettit 2011, 207–8):

No Combined Control Argument

Therefore:

This argument, if sound, effectively rules out the possibility of corporate responsibility, because group agents cannot be in control of what they do, which is a necessary requirement of responsibility.

In part 1 of this article I take issue with two lines of response to this problem in List and Pettit’s work. The first relies on Pettit and Frank Jackson’s account of program explanation, while the second is based on List and Peter Menzies’s difference-making account of higher-level causation. I argue that both accounts face problems, not least in their failure to address relevant cases of combined corporate and individual control.

1.1. Program Explanation and Control as Explanatory Relevance

In their book, List and Pettit (2011) take the no combined control argument to be valid and set out to reject the third premise. They rely on an analogy with cases of multilevel causation and especially on discussions of the so-called causal exclusion problem in philosophy of mind. List and Pettit have both made substantial contributions to these discussions. However, the primary impulse of Pettit’s contributions is the idea of programming explanation, which indeed is how the discussion is framed in List and Pettit (2011). The idea is that the group agent programs for the action, while the individual implements that program. The individual actually brings the effect about, while the presence of the group ensures the presence of some such causally efficacious individual. In this way, the effect depends on the group rather than on the specific individual, because the presence of the group makes the bringing about of the effect more robust. ³ As Pettit (2002, 160) explains in an earlier work,

the property of an agent that consists in his or her having certain beliefs and desires will programme for the performance of an appropriate action. In each case the presence of the higher-order state makes it highly probable, though in a non-causal way, that there will be a lower-order state present—the realizer of the higher-order—that produces the outcome to be explained.

This account of program explanation grounds the claim that both the programmer and the controller are causally relevant and both in causal control. List and Pettit (2011, 210) say,

Both programming and implementing are ways, intuitively, of being causally relevant and so it makes sense, depending on context, to invoke one or the other in causal explanation of the effect.

However, there are certain problems to this story.

In an explanatory context, it is surely sometimes apt to use a more abstract description and omit irrelevant details from the explanans. This does not in itself validate a metaphysical conclusion about higher-level causal control, however. To argue that programming entails causal control, one would need to argue that such explanations somehow track the causal facts and in so doing properly rebut the so-called causal exclusion argument. ⁴ It is not clear, however, what further premise would ground the transition from causal explanatory relevance to causal control. Also, the appeal to the intuitiveness of programming causes is argumentatively thin in the face of the vigorous controversy sparked by the issue of higher-level causation.

A further problem with their use of the programming account in the present context is that the intuitive plausibility of the examples given trades on the description sensitivity of programming explanations. “Description sensitivity” means that the explanatory relevance of programming causes depends on the description of the explanandum. The issue surfaces in the following extract, where List and Pettit discuss a case of heated water breaking a bottle:

The temperature of the water controls for the breaking of the flask so far as it ensures, more or less, that there will be some molecule, maybe this, maybe that, which has a momentum and position sufficient to trigger the breaking; the molecule itself controls for the breaking so far as it ensures that this particular crack materializes in the surface of the flask. (List and Pettit 2011, 211)

Notice how the description of the explanandum (the effect) changes significantly. First, the breaking is described as “the breaking,” a rough-grained description allowing for it to have occurred in different ways. This is important, because such a rough-grained description could have been made true by a slightly different breaking, one caused by a different implementation of the program. Second, the breaking is picked out by a fine-grained description: “this particular crack materializes in the surface of the flask.” This fine-grained description requires a more fine-grained explanation because it is not made true by the same wide range of events as the rough-grained “the breaking.” The intuitive appeal of the story as told in List and Pettit (2011) derives from the fact that the proper description of the explanans is relative to the given description of the explanandum. Explaining the materialization of a particular crack is therefore a different job from explaining that the bottle cracks some way or another. The needed metaphysical conclusion about causal control, however, does not follow from these features of explanations. ⁵

In developing their account of programming explanations, Jackson and Pettit (1990) distinguished between causal efficacy and causal relevance. Causal efficacy is not precisely defined but means something like being actively involved in the production of the effect. Causal relevance, however, roughly means causal-explanatorily relevant and does not imply causal efficacy:

The realization of the higher-order property did not produce the radiation in the manner of the lower-order. But it meant that there would be a suitably effiacious property available, perhaps that involving such and such particular atoms, perhaps one involving others. And so the property was causally relevant to the radiation, under a perfectly ordinary sense of relevance, though it was not efficacious. It did not do any work in producing the radiation—it was perfectly inert—but it had the relevance of ensuring that there would be some property there to exercise the efficacy required. (114)

The distinction between causal efficacy, to be found at the fundamental level, and causal relevance, understood as higher level programming, is inherently problematic. If there is no fundamental level, which should be an entirely open question, there is no causal efficacy. For interesting discussions of this so-called causal drainage problem, see Block (2003), Kim (2003), and Schaffer (2003). A better solution would be to cash out causal efficacy in terms of difference making (as List and Menzies do, see below) and argue directly that higher-level entities, or properties, can be causally efficacious in the fullest sense.

Even if we sidestep these problems, the program explanation account fails to cover interesting cases of combined corporate and individual control. These are cases where the presence of the group filters for certain ways an action can be done, thereby making the outcomes less robust, or less probable. Imagine a case where the presence of a group does not play any actual role for an individual doing a certain action (e.g., a Nazi bureaucrat saving a homosexual friend’s life) but where, if the individual had attempted to do the action (save his friend’s life) in a slightly different way, the group would have intervened and stopped him. It seems reasonable to say that groups can control for actions in this filtering way, without actually being causally involved; they are rather acting as potential preventers. The program explanation account does not cover these cases, because it is not the case that the group programs for the outcome, rather the opposite; that is, the group makes the outcome less robust or likely.

In light of these critical remarks about the program explanation account of corporate control, we should take a closer look at the List and Menzies article to which List and Pettit (2011) refer and which discusses the notion of higher-level causal efficacy directly in terms of difference making.

1.2. Causal Exclusion and Control as Difference Making

On the assumption that the relevant notion of control is causal control, an argument for combined causal efficacy should provide resources for solving the combined control problem. List and Menzies provide such an account in their “Non-reductive Physicalism and the Limits of the Exclusion Principle.” They do this by revising the so-called causal exclusion principle based on an account of causation as difference making. The exclusion principle, in a standard formulation by Jaegwon Kim (2005, 42), reads,

Exclusion. No single event can have more than one sufficient cause occurring at any given time—unless it is a genuine case of causal overdetermination.

Notice the similarity to (3) in the combined control problem. The alleged truth of this principle has often been taken to be an a priori matter. It has not been argued from empirical premises but rather been taken as some kind of regulative ideal on the metaphysics of causation; do not multiply causes beyond explanatory needs. That also explains the proviso in Kim’s formulation, since it rules out some atypical but empirically acceptable cases. ⁶ List and Menzies argue, correctly in my view, that the truth of the principle is a contingent matter. It depends on the specifics of every case. This is especially clear on a difference-making account of causation; if two cause candidates both bear the right difference-making relations to an effect, then they are both causes of that effect. Bear in mind that there might be alternative views on causation where the principle does not have this contingent status. It might be that the substantial discussion really boils down to what is the better account of causation. For present purposes, I shall leave that issue aside, noting that difference-making accounts have a strong standing in the contemporary causation literature. ⁷

Let us distill the main line of reasoning in List and Menzies’s revision of the exclusion principle. A naïve—but sufficient for our purposes—statement of the difference-making account is

Difference-Making: A is a cause of B if and only if (A B) and (~A ~B)

where denotes the counterfactual conditional: If the antecedent were to occur, the consequent would occur. Some details enter the picture in spelling out the truth conditions for the counterfactual conditionals. ⁸ The core idea, however, is simple enough; the truth of certain counterfactual conditionals is necessary and sufficient for causation. Assuming such a framework and giving up the a priori status of the exclusion principle, one can state necessary and sufficient conditions for causal competition and exclusion to occur.

First, however, we should note that it is cause candidates related by the supervenience relation that are of interest, both for the topic of higher-level causation and for corporate causal control. This is due to the shared assumption that supervenience obtains both between higher- and lower-level cause candidates and between corporate and individual agency. Assuming a difference-making account of causation, causal exclusion occurs only when the patterns of counterfactual dependence between the competing cause candidates and the effect differ in relevant ways. More precisely, causal competition does not occur in cases of realization sensitive causation (List and Menzies 2009, 493, 495–96):

Realization Sensitivity: F causes G, where F is realized by/supervenient on F*, is realization sensitive if and only if G is absent in all those F worlds that are closest non-F* worlds

where F and G are causal relata, F worlds are the possible worlds where F is present, and non-F* worlds are the possible worlds where F* is absent. Such cases are special, but if they occur, neither cause candidate is screened off because all the relevant counterfactuals:

F G

~F ~G

F* G

~F* ~G

will be true in such cases. In all other cases, a revised causal exclusion principle holds, stated here in two equivalent formulations (490):

Revised Exclusion Principle (Upwards Formulation): If a property F causes a property G, then no distinct property F* that supervenes on F causes G

Revised Exclusion Principle (Equivalent Downwards Formulation): If a property F causes a property G, then no distinct property F* that subvenes on or realizes F causes G

Importantly, exclusion occurs whenever realization sensitivity fails. So it is only in a very limited range of cases that causal efficacy combines among properties or events related by the supervenience relation.

Applied to the combined control problem, the List and Menzies account of higher-level causation therefore leaves very limited space for corporate control because we can only reject (3) in the no combined control argument on the assumption of realization sensitive agency, which is the relevant analogue to realization sensitive causation:

Realization Sensitive Agency: Group G doing E, where A is a particular individual’s activity in doing E, is realization sensitive if and only if E is absent in all those G worlds that are closest non-A worlds

where G worlds are the possible worlds where the group is present and non-A worlds are the possible worlds where the particular individual in question is not performing the relevant activity. An example would be one in which a given individual possesses certain skills without which the group will be unable to perform a certain action. If the group performs that action, the individual is in control, and the control of the group in such cases reduces to the control of that individual. By analogy to the multilevel causation case, corporate and individual causal control for actions combine in, and only in, cases of realization sensitive agency.

In cases of realization insensitive corporate agency, premise (3) of the combined control problem still holds. Some of the most interesting cases of corporate agency are realization insensitive in this sense. These are cases where a group does E, and A is a particular individual’s activity in doing E, but where, if A had not been done by that individual, E would still have been done. The group activity in such cases is insensitive to that particular individual’s activity. In cases where the group would have done E even without A, (3) still holds. As long as one accepts the other premises in the no combined control argument, one is required to reject corporate causal control in such cases. Notice, however, that the natural interpretation of such scenarios in the multilevel causation case is as an instance of downward exclusion—that is, the corporate causal control excludes the individual causal control. That interpretation, however, is in conflict with premise (2) of the no combined control argument, which says that individuals are in control of anything they do and so in control of anything they do in acting for a group.

1.3. Preliminary Conclusions

On the assumption that the relevant notion of control is causal control, proponents of corporate responsibility still have a pressing problem on their hands. Taking the lessons of List and Menzies’s revision of the causal exclusion principle, we can state a revised version of the no combined control argument.

Revised No Combined Control Argument

Therefore:

Is it sufficient for an interesting account of corporate responsibility that group agents can be in causal control in a highly limited range of cases? Arguably not. Cases of realization insensitive corporate agency still pose serious trouble. They are of great importance, because they seem to be paradigmatic cases where the group agent is in control at the expense of individual control. Moreover, they are cases where the corporate control does not reduce to the control of a particular individual, as is reasonably the case in cases of realization sensitive corporate agency. If such cases really involve corporate control, they pose a threat to the claim that individuals are in causal control of what they do in such cases.

Since we have assumed that causal control is the relevant notion so far, it is time to look into other notions of control and see whether they can give us a better solution to the combined control problem.

2. Other Relevant Notions of Control

For clarity, let us first restate the understandings of causal control in play so far. First, causal control assuming a difference-making account of causation.

Causal Control I: A causally controls for B if and only if (A B) and (~A ~B)

This parts with the understanding of causal efficacy implicit in the earlier discussion of programming explanation. On that account, being causally efficacious meant something like actually partaking in the production of the effect. If causal efficacy is required for causal control, there will only be causal control at the fundamental level: the level of individuals in the present context. On the other hand, if causal explanatory relevance is sufficient for causal control, an idea that I criticized earlier, and program explanation is a kind of causal explanation, appropriately figuring in a programming explanation will be sufficient for causal control. For the program explanation response to the combined control problem to get off the ground, causal control must be understood as

Causal Control II. A causally controls for B if and only if A produces B or A is causal explanatorily relevant for B

It is unfortunate that List and Pettit (2011) does not spell out the specific understanding of causal control, because it does actually matter for the combined control problem, and the two understandings discussed are not equivalent. Notice that due to the disjunctive nature of the definition of Causal Control II, it might be that the no combined control argument equivocates on the notion of control. I do not think, however, that this is best cashed out in terms of an ambiguity in the notion of causal control, but that might boil down to a terminological question. Even so, I think we should adopt an independently motivated account of causation, more in line with the List and Menzies approach.

In the chapter on the control desideratum—a desideratum on group agent design securing a certain amount of individual control—List and Pettit (2011, 170) also discuss control:

There is one sense in which members always have control over the group on our account: what the group thinks and does is always determined by what the members think and do. According to the supervenience thesis we have defended, the group’s attitudes and actions are a function of the members’ attitudes and actions. This, however, is not the sense of control we have in mind when we talk about protecting the members’ control in relation to the group agent.

This gives a third understanding of control relevant for the present discussion, namely supervenience. According to the supervenience thesis, which List and Pettit (2011) accept, the totality of individual preferences and attitudes fully determines the preferences and attitudes of the group agent. Fully determines means that fixing all the individual attitudes and preferences also fixes the attitudes and preferences of the group agent and that changing the attitudes or preferences of the group agent requires changing some attitudes or preferences of the individuals.

Control as Supervenience. The totality of individual preferences and attitudes fully determines the preferences and attitudes of the group agent.

Ascribing this kind of control, however, is not sufficient for individuals being in control of the group agent on particular matters of individual interest. This will require the individual to be individually decisive:

It is not enough for individuals to be able to contribute to what the group thinks or does on the issues that particularly matter to them. They must be individually decisive on those issues. (List and Pettit 2011, 171)

The authors then go on to define decisiveness by the following two conditions ⁹ (171):

Positive decisiveness. If the individual has a positive attitude on “p”—that is, he or she judges that p, or prefers that p, depending on whether we are dealing with judgments or preferences—then the group also has a positive attitude on “p”.

Negative decisiveness. If the individual has a negative attitude on “p”—that is, he or she does not judge that p, or does not prefer that p, again depending on whether we are dealing with judgments or preferences—then the group also has a negative attitude on “p”.

The decisiveness-based notion of control is expressed in the conditional:

Control as Decisiveness. If the individual has attitude Yes (or No) on “p”, then the group also has attitude Yes (or No) on “p”.

Finally, List and Pettit (2011) require that this conditional be true independently of content and context. That is, the conditional should hold true under variations of the content of “p” and under certain variations of the context—for example, under variations of the attitudes of other individuals. This relation of control holds between individuals and the attitudes of a group agent on matters of importance for those individuals. The propositions on these matters will compose the sphere of control for that individual. Insofar as these attitudes straightforwardly lead to actions, the individual will be in control of the group agent’s agency within his or her sphere of control.

This kind of individual control seems to be at odds with the group agent being in control on those very same matters. The control desideratum seems to act as a constraint on the autonomy of the corporate agent. Therefore, the corporate agent is seemingly not in control on matters within the individuals’ spheres of control. This gives a restriction on the relevant cases of combined control but does not directly concern the core issue, which is how control on matters outside the individuals’ spheres of control can be combined. It is time to probe into the notion of structural control alluded to several times already.

2.1. Structural Control

The notion of control relevant for corporate responsibility should include both causal control, understood in terms of difference making, and what I call structural control. ¹⁰ As we will soon see, incorporating structural control pays respect to the core insights of the discussion in List and Pettit (2011).

In an earlier article, Frank Jackson and Pettit (1992, 130) discussed the notion of structural explanation and in particular in terms of programming. Here is what they say it has to offer for collectivists:

While the program model does not undermine individual agent autonomy, neither does it particularly flatter the individual. It suggests that for many of the things that happen in social life—specifically, those that are subject to structural explanation—the particular attitudes and actions of particular individuals which led to those things were not necessary prerequisites. If those people had not done those things, other people would have wittingly or unwittingly steeped into the breach. Thus, if our model does not undermine agent autonomy, it does suggest that the initiatives of individuals may not be as important as is often thought. There are constancies in social affairs which no individual is in a position to alter.

The extract brings out the trade-off, or competition, that exists between individual and corporate control, a competition that the no combined control argument takes to the extreme. However, the way List and Pettit (2011, 211–12) deal with the combined control problem does not clearly bring this trade-off out. Rather, it creates the impression that corporate and individual control come in harmony:

This line of thought resolves the problem that seemed to prevent a group agent from satisfying the third condition for fitness to be held responsible. A group agent, so it now transpires, is as fit as any individual human being to be held responsible for what it does. The members of such an agent combine to form a single agent faced with normatively significant choices, capable of making a judgment on what is good and bad, right and wrong, and capable of ensuring that one or another option is chosen. The individuals who give life to such an agent have to answer, of course, for what they do in making corporate agency possible. But the entity they maintain also has to answer as a whole for what it does at the corporate level, drawing on the resources provided by its members. It has all the agential capacities to make this possible.

It is, I think, exactly on the point of control that we should be able to spell out how the trade-off between corporate and individual control and responsibility plays out. For this purpose, however, List and Pettit’s discussion in the book is suboptimal because it does not explicitly give room for doing this in terms of control.

On the view I suggest, having causal control for an action E is analyzed as being a difference maker for E. I will rely on some insights from James Woodward’s (2003) account of causation here, where causation requires invariance under certain interventionist counterfactuals. Having structural control for E is then analyzed as being a robustness enhancer or decreaser for the bringing about of E. ¹¹ In the context of group agency, being in control in at least one of these ways is a requirement for responsibility. Importantly, causal control alone is not a necessary requirement for responsibility.

Structural control works by ensuring the existence of mechanisms that, though not causally active in bringing the action about, control for the action by affecting the robustness of the bringing about of the action. This kind of control is exercised by a group agent when such mechanisms are ensured by the presence of the group agent. There are two principled ways in which such mechanisms can change the robustness of an outcome. (1) Preempted backup mechanisms enhance the robustness of the outcome by ensuring that it would have been brought about even if the actual cause had failed to do so. (2) Passive preventers decrease the robustness of the outcome by filtering for a limited set of ways in which the outcome could have been brought about. There will also be mixed cases where both kinds of mechanisms are at play. There is even the possibility of mixed cases where the mechanisms cancel each other out such that the robustness of the action appears to be independent of the structural context.

To get a more intuitive grasp of structural control by robustness enhancing, consider a standard case of late preemption, illustrated in Figure 1. Here, C₁ is an actual cause of E. However, there is no counterfactual dependence on the naïve difference-making view considered earlier, since C₂ would act as a backup and bring about E even in the absence of C₁. Such cases can be handled by evaluating dependencies when we hold variables fixed. Holding C₂ fixed at not-firing, E will depend on C₁, and the causal relation is established. ¹² C₂ is in structural control of E since the structure of the setup makes C₂ a robustness enhancer for the bringing about of E. That is, it can act as a backup bringing E about in the absence of C₁. If C₁ and C₂ composed a group agent, the group agent would be in structural control as long as the presence of the group agent ensured the presence of the structural setup illustrated in Figure 1.

OPEN IN VIEWER

Figure 1.

Late Preemption.

The difference in causal status between C₁ and C₂ is difficult to bring out on a difference-making view. In fact, it is perhaps the hardest problem case for the necessity of counterfactual dependence for causation. The judgment that C₁ is the actual cause of E in such cases appears to be very robust. However, the patterns of counterfactual dependence between E and the two cause candidates are relative to the individuation of E. Suppose that E is modally robust to the extent that it would have been the very same event even if it had been caused by C₂ instead of C₁. Then E does not depend counterfactually on C₁. If it is fine grained in ways that would make it into a different event if caused by C₂ rather than C₁, then it would indeed depend counterfactually on C₁. Depending on the fine graininess of the effect then, the patterns of counterfactual dependence will change.

Arguably, actions are fine grained in the sense that they are agent dependent. One and the numerically same action could not have been performed by a different agent. Or weaker; actions qua locus of responsibility are agent dependent, and that is the relevant notion in the present context of discussion. If we assume such agent relativity, the problematic preemption cases are easily handled. E depends on C₁ rather than C₂ because it could not have been brought about by any other agent than C₁ without losing its identity. This solution is not available for causal relata that are not actions, which introduces further complications; however, for present purposes we can sidestep them. ¹³ In cases like the one in Figure 1, then, C₂ controls for the instantiation of action of type E, while C₁ actually causes the particular token. C₁ is in causal control in the sense that the absence of C₁ would result in the absence of this particular action, while C₂ is in structural control in that it controls for the relevant act-type being instantiated. I will postpone the issue of who the proper agent is in the latter case to section 3.1.

Structural control as robustness decreasing can be illustrated with a case involving passive preventers. Even though actions qua loci of responsibility are fine grained in some ways, they are arguably rough grained in the way that they could have been performed in slightly different ways. If a structural context (e.g., the presence of a group agent) filters for certain alterations of the action by ensuring the presence of potential preventers, it controls for the action by making it less robust. Recall the case of the Nazi bureaucrat saving his friend. He actually saved his friend’s life, but had he attempted to do it in a slightly different way, the secret police would have stopped him. The structural context surely controls for the action happening exactly the way it did, but it does not make it more robust; rather, it makes it less robust since it filters out other ways in which it could have been performed. Arguably, a group agent exercising this kind of control by filtering can be held responsible for consequences of such actions, especially if the action would have had better consequences if performed differently.

3. The No Combined Control Argument; Alternative Rejection

We are now in position to reject the revised no combined control argument, but let me first restate it:

Revised No Combined Control Argument

Therefore:

We have seen that under a causal understanding of control as difference making, it gives only a very limited range of control for group agents. Group agents are in control in special cases of realization-sensitive agency. When accepting structural control as sufficient for satisfying the third requirement on responsibility, however, we can answer the combined control problem by pointing out that the argument equivocates on the notion of control. Actually there are two senses of control in play; individuals are typically in causal control, and groups are typically in structural control.

Even when we reject the argument, an important problem remains. If we assume that actions—qua loci of responsibility—have their agents essentially, all individuals will be in causal control of their actions. However, the group then cannot be in structural control of those actions. The reason is clear. If that specific individual had not performed that action, no one else could have. The individual is responsible for doing the action but not for that act-type being instantiated. This gives rise to a dilemma: Either actions, qua loci of responsibility, have their agents essentially, or they do not. If actions have their agents essentially, group agents are either not in control or there are actions that have group agents and not individual agents essentially. The first option rules out combined control, the second option undermines premise (1) of the no combined control argument. If actions do not have their agents essentially, there will be cases where the individual agent is not in control of the action, due to corporate structural control, and thus not responsible for that action. This undermines premise (2). Schematically, we have the following options:

Again it seems that either combined control is ruled out, except in cases of realization-sensitive agency, or premise (1) has to be rejected. The next section looks into an argument for, and some consequences of, rejecting (1).

3.1 Act-Dualism and the Modal Robustness of Corporate Agency

According to List and Pettit (2011, 44), groups are modally robust with respect to their individual members:

Other collections have an identity that can survive changes of membership. Examples are the collections of people constituting a nation, a university, or a purposive organization. We call the former “mere collections”, the latter “groups”.

I read the “can survive changes” as supporting both the predictive “will survive (some) changes” and the counterfactual “could have survived (some) changes.” The latter reading is the relevant one for the present concerns. ¹⁴

There are subtle issues concerning the individuation of actions that have significant consequences for the understanding of group agency. Invoking the idea that actions have their agents essentially—that is, that one and the same token action could not have been performed by a different agent, we can run the following argument.

Act-Dualism Argument

Therefore:

The group actions and the collections of individual actions cannot be identical because they have different modal properties—individual actions have their individual agents essentially, while the group actions do not. They do not because the group is modally robust with respect to its individual members, so the group could have performed the same action without the individuals that were actually involved (assuming that other necessary criteria for action-identity are satisfied).

An act-dualism of this kind—that is, that group actions supervene on but are not identical to collections of individual actions—seems to be in good harmony with List and Pettit’s theory of group agency. First, it adds to the solution to the combined control problem since we can also reject the first premise of the no combined control argument, allowing for a wider range of combined control. In particular, it resolves the cases of downward exclusion that were in conflict with premise (2) of the argument. Downward exclusion does not occur, because the fact that the group agent causally controls for the group action does not rule out that the individuals causally control for their individual actions. However, the individuals will still be in causal control of the group action in cases of realization-sensitive agency. Second, it provides an abstract understanding of how group agency and individual agency relate. Group actions are sometimes more modally robust than individual actions; they are necessitated by individual actions but do not necessitate them; and these features are brought about by structural features of the group agent.

The modal robustness of group agency is key for understanding the importance of an account of such agency. In contrast to philosophical debates about co-located objects differing only in modal properties (like the statue and the lump of clay out of which it is made), actions differing in modal properties have immediate practical consequences. The presence of a structure that makes a certain action more modally robust is straightforwardly practically relevant, something that is especially clear in a forward-looking perspective. Since group agency is more modally robust than individual agency in a wide range of cases, group agency should be a locus of great concern for improving corporate and individual agents, and their fitness to be responsible.

This insight about group agency also provides resources for distributing group responsibility among involved individuals and external agents (individuals, or other group agents, e.g., the state). The residue of responsibility at the corporate level, analyzed as structural features that act as robustness enhancers (or decreasers) for certain actions, is due to lack of revision of these structural features. Such revision is the responsibility of the involved individuals, varying with their role in the group, and external control. This might not be a surprising result. Rather the contrary. But insofar as it comes as a natural consequence of a theory of corporate responsibility and control, I take it as a consideration in favor of that theory.

The suggested act-dualism nicely illustrates how corporate and individual control interact in cases of realization-insensitive corporate agency. If one holds that the corporate action, qua locus of responsibility, essentially depends on the agency of the corporate agent, and that the individual action, qua locus of responsibility, essentially depends on the agency of that particular individual, the distribution of responsibility is straightforward. The individual is responsible for being the agent of the action but not for the act-type being instantiated. The residue of responsibility for the act-type being instantiated at the corporate level has to be distributed to the individuals, not as responsibility for the action directly, but as responsibility for the creating or upholding of corporate structure that controls for such actions.

There are several other notions of control that I have not discussed here, among them the control that the corporate agent might have over the epistemic status and the emotional repertoire of the individuals. ¹⁵ Such control might be due to a corporate structure that hides most of the relevant information (e.g., mafia structure) and that exploits individual emotional dispositions: loyalty, fear, crave for recognition, honor, and so on. Alleged examples are not hard to find. The discussion of such notions of control is no less important than the present one, but it has intentionally not been the topic of this article. Still, the present discussion highlights the framework within which these other kinds of control might operate.

Conclusion

The combined control problem poses a real threat to an interesting account of corporate responsibility when we assume a causal understanding of control. Proposed solutions in terms of control as explanatory relevance and, as difference making, are neither fully satisfactory. The difference-making account gives a restricted range of combined control, namely in cases of realization-sensitive corporate agency. Positively, I have argued that structural control is sufficient to fulfill List and Pettit’s control-requirement on responsibility. On that view, the no combined control argument is rejected due to an equivocation between causal and structural control. Having said that, problems remain, and I have suggested that an act-dualism is fruitful for giving a full-fledged account of the relation between individual and corporate control and agency.