. 2
( 10)


tuitive act consequentialism. That is wrong. Rather, things go the other
way. A novel direct form of consequentialism, Multiple-Act Consequen-
tialism, in fact collapses comfortably into extensional equivalence with
the best form of indirect consequentialism, the indirect form that has the
most intuitive implications. And it carries its coherent and intuitive direct
consequentialist rationale along with it when it does so.
In the next chapter we will see in detail that Multiple-Act Conse-
quentialism can evade the three standard objections to familiar forms of
act consequentialism, objections rooted in normative intuitions. But it
is also relevant that it is true, independent of any standard objections or
normative intuitions about cases. That is the central topic of this chapter.


Two non-normative facts require Multiple-Act Consequentialism if con-
sequentialism is true and if its proper speci¬c form has a coherent norma-
tive rationale. In other words, those non-normative facts require MAC

19 Postow (1977); Jackson (1987); T¨ nnsj¨ (1998).
a o

if a basic consequentialist principle “ for instance, HMP or a traditional
utilitarian valuation principle “ is true. This and the next two sections
show why.
The ¬rst key non-normative fact is that there are group acts, performed
by group agents. I will begin by clarifying the speci¬c conceptions of
group acts and agents to be deployed here. Then I will argue that they
are the correct conceptions.
The basic cells of group agents in my sense are what I will call atomic
agents.20 These are more or less momentary periods of human agents,
which persist for but a short time. I might also say that atomic agents
are brief time slices of people, which have correspondingly short sets
of options, as long as such language isn™t taken to imply a controversial
metaphysical thesis about the nature of personal identity over time.
Of course, momentary agents are not the most familiar agents. Familiar
agents persist over whole lives. But my analytic convenience is not really
revisionary, nor does it invoke controversial metaphysics. And I don™t mean
to be ¬ghting over the folk word “agent”. Clearly, it makes familiar sense to
talk of how someone™s options change over time, or how they change their
preferences or choices among stable options. Familiar agents try different
things at different times. Clearly, quite brief periods of someone™s life
are suf¬cient to try for certain options over others, and even to succeed
at certain things, and hence can constitute agents with options in my
perhaps unusually weak sense. That is all that the notion of an atomic
agent requires.
But perhaps the notion of a group agent or of a group act seems more
problematic. One or both notions may seem mere metaphors.
Still, since atomic agents exist, we should also grant that temporally
persisting individual agents, the most familiar and obvious kind of agent,
are a type of group agent. Some intuitive individual actions require a tem-
porally extended series of steps, and this is one type of group action in my
sense, which requires extended temporal coordination among temporally
distinct atomic agents. In such a case there is a single persisting group
agent comprised of many cooperating atomic agents.21

20 Thanks to Mark van Roojen and Guy Rohrbaugh for the phrase.
21 Let me stress again that I do not take this to imply controversial metaphysical theses about
the nature of identity over time. The connection between group agents and temporally
extended individual agency is proposed in Jackson (1987). It is also a theme of Korsgaard
(1996b) and Rovane (1998). If options and actions that someone can perform include
conditions that obtain after their death, or after their agency ceases, then this conception

Common sense also recognizes group actions and agents of other kinds,
involving several persisting humans. Certainly there are circumstances in
which it makes vivid sense to talk of a group such as an army or a family
trying one thing or another, or achieving one thing or another. This gives
these groups alternative options in the crucial sense.
Even this very common speech may seem metaphoric. But in our
messy concrete world, even individual people are built up from messy
¬‚eshy bits in such a way that it is sometimes a somewhat soft or inde-
terminate fact whether they try one thing rather than another at even a
particular time. And certainly it is sometimes a somewhat soft or inde-
terminate fact whether someone tries something continuously over time.
And yet intuitive individual agency is no mere metaphor. So too with
group agents consisting of many persons. On the other hand, plausible
metaphysical presumptions do rule out some things that my denial that
group agency is a mere metaphor might seem to suggest or threaten. If
literal group agency were to require some strange nonreducible monstros-
ity on the order of absolute idealism, that would be out of bounds. But
literal group agency doesn™t require any such thing. There are a variety
of more or less reductive accounts of group agency now in play among
So let me re¬ne the notions of group agent and group act deployed
here by asking this question: What concrete collections of atomic agents
constitute intuitive group agents capable of intuitive group action in my
sense? And let me begin my answer with a concrete case of Margaret
Gilbert™s. She asks,
What is it for two people to go for a walk together? . . . Imagine that Sue . . . is out
for a walk. . . . Suddenly she realizes that . . . [Jack,] a man in a black cloak[,] . . . has
begun to walk alongside her. . . . His physical proximity is clearly not enough to
make [it] . . . that they are going for a walk together. . . . Is each one™s possession
of the goal that they continue walking alongside each other logically suf¬cient
for their going for a walk together? I would say not. Note that it is possible that
each one™s possession of the goal in question is not known by either one. Sue may
look worried and Jack may suspect that she would rather be alone. . . . [Assume
then] that is it common knowledge between Jack and Sue that each one has the goal

of a continuing agent would need to be augmented in certain ways. But this will not be
important in what follows.
22 Gilbert (1989, 1996, 2000). The essay “Walking Together” in Gilbert (1996: 177“194) is a
good introduction to this complex project. See also Bratman (1987, 1992, 1993); Tuomela

in question. By this I mean, roughly, that each one™s goal is completely out in the
open as far as the two of them are concerned. . . . [Still] a crucial feature of going
for a walk together will be lacking. . . . [S]uppose that Jack starts drawing ahead.23

Gilbert suggests that this is a crux, and I agree. If Jack draws ahead and
that is the end of it, then there is no joint or group action, no going for
a walk together. Their activity isn™t entwined in the requisite way.24
Still, what exactly is lacking? Gilbert focuses on what she calls a nor-
mative though not a moral fact, that Sue is entitled to rebuke Jack.25 For
our purposes, this is an unfortunate feature of Gilbert™s account, since
we want to rest normative evaluations on MAC noncircularly. But it is
also apparently unnecessary to the central spirit of her proposal. What
seems crucial is not the fact of a (normative but nonmoral) obligation,
but rather merely the mutual acceptance that there is such an obligation.
We can believe that there are dragons when there are no dragons, and
we can accept or believe that there are obligations when there are no
obligations. As I will put it, Jack and Sue must accept that there is a reason
to continue to coordinate activity. This will serve to stabilize the activ-
ity, but without any genuine normative facts being required. It will also
explain the normative criticism and discussion we might expect in such a
situation if Jack goes blithely on, and which Gilbert stresses as signi¬cant.
My modi¬cation of Gilbert™s proposal is also desirable on other than sys-
tematic and metaphysical grounds. It will allow us to respect the obvious
fact that there can be group action toward abhorrent ends, where people
share a hideous project and hence have no true normative obligations of
any sort to continue to coordinate on that project, whatever they may
So that is my rough proposal, a variant of Gilbert™s view: Group action
and agency exists when there is common action by a number of agents
rooted in common true belief that there is a shared goal, and in acceptance
by all the members of the group that there is a reason to continue to
coordinate activity until the goal is adequately accomplished, a reason

23 Gilbert (1996: 178“180).
24 More exactly, it isn™t entwined in the requisite way for the existence of joint action when
they are still in close proximity in this way and they don™t have a plan that requires that they
ignore one another.
25 Gilbert (1996: 180). This theme is also present in her central original discussion in Gilbert
(1989: Chapter IV, 146“236), though not so clearly marked as crucial. Gilbert also dis-
cusses the acceptance by the two cooperators of this obligation, but characteristically calls
it knowledge of the obligation, which implies the existence of the obligation.

whose acceptance we can expect to occasion criticism and the acceptance
of criticism for failure to continue coordination until that point, or until
the goal is mutually abandoned.
Despite the fact that I will leave much of the vagueness of that
rough proposal intact, as suf¬cient for our purposes, let me add two
The ¬rst is occasioned by Aristotle™s distinction between processes and
actions, where “processes”, like making a rudder, characteristically aim at
a goal that extinguishes the activity in question, while pleasant “actions”,
like sur¬ng, do not.26 This will matter later on. So let it rather be that all
the members of the relevant group accept that there is a reason to con-
tinue to coordinate activity until the goal is adequately accomplished, or
inde¬nitely if it is not the sort of goal that will be ¬nitely and de¬nitely
accomplished, a reason whose acceptance we can expect to occasion crit-
icism and the acceptance of criticism for failure to continue coordination
until the point (if any) at which the goal is accomplished or mutually
The second complication is further speci¬cation of what I mean by
“common action”, “coordinated activity”, and “sharing a goal”, which in
turn involves claims about the sorts of motives that underlie group action.
Let me begin with the last phrase. Participants in a group action must
share a goal. This means that they must each have some motive to pursue
a certain perceived good. There must be some description of the content
of that shared goal that captures the contents of the goals of each. And
the goal must be moralized in a certain weak way. It must be taken to be
a good by each of the members of the group, which minimally involves
accepting it as a positive reason for choice. It also involves a good-pursuant
sentiment, motive, or intention, which I will call “bene¬cence”, I hope
without too much violence to ordinary usage. But perhaps mere desire for
the goal is enough, when the agent accepts the goal as a positive reason,
to constitute bene¬cence in this sense. Even abhorrent group acts pursue
some apparent good.
But what about “common action” and “coordinated activity”? Com-
mon action results from coordinated activity in pursuit of a shared goal.
And “coordinated activity” is at least hypothetically cooperative. You may
not be in communication with, or even know in any very speci¬c way, all
the participants in the group actions of which you are part, but you must

26 Aristotle (1980: Book I, Chapter 1).

be prepared to cooperate with them in pursuit of the goal of the action
should you be in position to do so. You must be prepared to work together
to achieve the goal. And this disposition must be supported by accepted
reasons, and also by a second sort of sentiment, a second sort of normative
motivation that is entwined with the bene¬cence that supports the goal
of the joint project. This is a sentiment of reciprocity that supports group
action in pursuit of that goal.
If there is a genuine group act in which we participate, then we will
have two sorts of normative involvement in that group, both modu-
lated by accepted reasons, and this will be mutually recognized in cer-
tain at least hypothetical ways. And indeed, those two sorts of norma-
tive involvement will be entwined. A particular group act will have a
particular normative goal, which will occasion one sort of normative
involvement, involving bene¬cence. And it is a group act, which will occa-
sion another sort, involving reciprocity. And indeed, in group acts these
two sorts of normative involvement will be entwined. We should expect
criticism of defections from group acts of two different but entwined
sorts, as failures to pursue a good and as failures to reciprocate. In group
actions in general, bene¬cence and reciprocity are entwined. But remem-
ber that I suggested in Chapter 1 that benevolence and reciprocity are
both basic moral motives. And benevolence is a certain sort of expansive
There are more complications regarding the nature of group actions
in my sense to which we will return in section V, particularly regarding
what it is to accept a reason. And we will return in section VI to issues of
moral motivation. But let me now resume our consideration of possible
alternative conceptions of group action.
There is a related yet still weaker conception of group action than my
own that shares some of its advantages over Gilbert™s view. But it is not
really workable. Gilbert requires genuine normative entitlements where
I merely require acceptance of reasons. But we might merely believe that
reasons exist without really accepting them, since accepting a reason in
my sense does have some motivational implications.
However, belief in such reasons is not enough for plausible group
action. Of course, genuine group action requires that there be some action
of a group toward its goal. But even if, say, a group of travelers inadvertently
stumbles toward some goal and everyone in the group believes in the
existence of some relevant reason to continue cooperative activity toward
that goal but without accepting that reason, there is no intuitive group
action. If everyone cognitively admits the existence of a reason to criticize

the others and to accept such criticism for failure to continue cooperative
activity toward that goal, but isn™t at all motivated to make that criticism
or to accept it, then in fact the travelers aren™t trying.
That is the key issue, I think: what things literally try. We may talk
of a boat trying to right itself, but we don™t really mean it. But we do
really mean it when we say a person is trying something over an extended
period of time, is on a quest; and that is a group act in my sense. And
since groups of many people can also try, such groups too can act. As I
said before, sometimes it is a somewhat soft or indeterminate fact whether
even some individual is trying one thing or another, but it is still a literal
fact that they are trying to do something. So it is still a literal fact that
they are intentionally acting.
Belief in the existence of reasons is not enough for trying. And Gilbert
requires normative facts that are not necessary. But there are also other
competing conceptions of group action. Some require weaker conditions
than my own, but don™t suf¬ce for trying. Some require richer condi-
tions, but which are not necessary for trying. Begin with notable richer
Michael Bratman makes this proposal:

Shared cooperative activity (SCA) involves, of course, appropriate behav-
iors. . . . Given appropriate behaviors, what else is needed . . . ? Suppose that you
and I sing a duet together, and that this is an SCA. I will be trying to be responsive
to your intentions and actions, knowing that you will be trying to be responsive
to my intentions and actions. This mutual responsiveness will be in the pursuit
of the goal we each have, namely, our singing the duet. You may have this goal
for different reasons than I do; but at the least we will each have this as a goal.
Finally, I will not merely stand back and allow you to sing your part of the duet.
If I believe that you need my help I will provide it if I can.27

There is a pair of relatively subtle elements of this view that are slight
enrichments of the view I prefer, enrichments to which we will shortly
return. But Bratman™s analysis of the remaining conditions undergoes
some re¬nement until it becomes roughly this: Each of our paired singers
sings their part, intends that they sing a duet, intends that they so act
in accordance with and because of meshing subplans for action, and this
is common knowledge between them. The stability of that intention in
each individual, as suggested in Bratman™s other work on the notion of

27 Bratman (1992: 327“328).

intention as a plan, provides a certain stability of interaction and ensures
certain sorts of mutual support.28
The relationship between this element of Bratman™s account and my
model turns on the relationship between stable intentions and accepted
reasons. And it seems that in the central cases for humans, these things do
not come apart. There may be animals, or even some humans, who have
intentions but don™t accept reasons. But, as Bratman himself suggests, you
and I seem to treat our continuing stable intentions as reasons of at least
some weight.29 And if we were not so to treat them, then they would
be insuf¬ciently stable to constitute stable group activity. So while my
variant of Gilbert™s view deploys a slightly different basic notion than the
weakened Bratman view, there is at least a rough con¬‚uence between
those two accounts.
Nevertheless, there are the controversial enrichments that Bratman
incorporates. The elements developed in the two preceding paragraphs
are not enough, Bratman believes, for an SCA. Here™s the ¬rst reason:

You and I are singing the duet. I fully expect you to get your notes right, and so
I intend to coordinate my notes with yours so that we sing the duet. But I have
no disposition at all to help you should you stumble on your notes; for I would
prefer your failure to our success. Were you unexpectedly to stumble I would
gleefully allow you to be embarrassed in front of the audience. . . . And you have
a similar attitude. . . . [O]ur singing may be jointly intentional; but it is not a SCA.30

Bratman can do with the notion of an SCA what he wishes. But this
enrichment doesn™t seem necessary for genuine group action, for genuine
group trying. Perhaps I am just not as nice a guy as Bratman. But I can
easily imagine conscientiously engaging in a group action whose goal I
favor, and yet hoping secretly that it will all fall apart because you will let
down your end while I have kept mine up. Perhaps your embarrassment
of that sort is a more signi¬cant goal to me than the goal of the group
action, even though not something I will actively pursue, indeed not
something that I can actively pursue. And even without such a secret goal,
I can imagine being prepared to do my part in a group action, but not
being prepared to go beyond my part to bail you out should you stumble
in certain ways. I accept the reasons that underlie the group activity and
grant them signi¬cant weight, enough weight that I am part of the group

28 Bratman (1987).
29 Ibid., 51.
30 Bratman (1992: 336“337).

attempt, but not as much weight, or perhaps not the kind of weight, that
Bratman requires. I am prepared to cooperate, but only so far.
There is another enrichment present in Bratman™s account. He requires
that continuing action be mutually responsive.

Suppose, for example, you and I lay plans for you to go to San Francisco while I go
to New York. We might have a web of intentions concerning this joint activity,
a web which satis¬es . . . [the preceding conditions.] And our activity of prior
planning may itself be a SCA. But if when we each go our separate ways there is
no mutual responsiveness in action, our activity is prepackaged cooperation, not

But the prior planning Bratman notes “ which, for instance, might
have speci¬ed the separate trips as part of a clever plot “ is suf¬cient to
constitute group activity in this case, as long as the individuals cleave to
the plan. Yet it probably wouldn™t be enough if they were still in reach
of one another and failed to respond together in even minimal ways to
changes in conditions. But such response seems assured by the conditions
required for group agency in my sense.
Bratman™s notion of an SCA is slightly richer than the notion we need
in two ways, as is Gilbert™s in one. And both alternative accounts overlap
more or less in the notion I will deploy. Now consider one more contrast
with a richer and more restrictive account.
Tuomela™s account of group action is quite complicated. It is in some
ways quite close to my proposal. But one of the complexities of Tuomela™s
view is that it is formulated in such a way as to suggest that there are
more and less central cases of group agency, and the core cases are a bit
richer than those I treat as crucial. Social group action, Tuomela believes,
requires explicit or implicit agreement, and it requires that the group
have an authority system, which determines the way in which individual
wills determine group will.32 In the paradigm cases of social group action,
according to Tuomela, there is explicit agreement and also a relatively rich
authority “system”, a characteristic way in which group members transfer
their authority “over some issues to the group, and the group . . . use[s]
special operative members to form its will or . . . form[s] the group will
through negotiation, bargaining, or voting with all its members acting as
operative ones.”33 While an informal group carrying a table counts as a

31 Ibid., 339.
32 Tuomela (1995: 171“227).
33 Ibid., 177.

social group engaged in group action, on Tuomela™s view, it is only in a
peripheral and somewhat degenerate way. It is only because that group can
be conceived as having an implicit agreement and a degenerate authority
But these complexities are unhelpful, at least in our context. Many
group agents that literally try involve neither agreements nor authority
systems of any reasonably literal sort. Still, even the central group agents
and actions in Tuomela™s sense are simply enriched in a fourth direction
from the core that my notion speci¬es. When an authority system is not
accepted by its participants as generating reasons, when so-called implicit
agreements are not recognized as reasons by the parties involved, the
authority system is unstable and perhaps illegitimate, and there really is
no implicit agreement.
We have seen that my notion of group action and agency captures a
notion in which Tuomela™s, Bratman™s, and Gilbert™s conceptions overlap,
and which can be enriched in various different ways to get the more
speci¬c forms of group agency that they favor as an analytical focus. That
seems an advantage of my account. And yet it seems rich enough to deliver
genuine group trying. It seems to be a minimum account that captures
all group acts.
But there aren™t just richer notions in the rough vicinity of my notion
of group agency, but weaker notions as well, even beyond the case we
considered earlier. One way to see this is to consider packs of wild dogs
that engage in cooperative and mutually responsive activity that isn™t
stabilized by anything closely resembling accepted reasons. That may
seem to be group action in some recognizable sense. But it seems a form
that is uncommon in humans when accepted reasons pull in another
direction, and accepted reasons with us always seem to pull somewhere.
And when it does exist in humans, its obvious contrast with more robust
and ordinary forms shows that it isn™t really group action at all. While
even humans can engage in some recognizable group activities “ for
instance, certain sorts of crowd response at concerts, boat races, or jousts “
that are not stabilized by accepted reasons and are even indeed such that
all accepted reasons pull in another direction, they are then no more
intentional actions by the group than an individual™s trembling or fainting
is intentional. They do not literally involve intentional action or trying.
That is also the cut between group acts in my sense and other weaker
forms of collective activity that involve humans. For instance, Jackson
and T¨ nnsj¨ have proposed that there are group or collective actions that
a o
are conjunctions of individual actions independent of any sort of shared

coordination or goal.34 The intention behind the collective action is per-
haps merely the conjunction of all the individual intentions behind the
individual actions that make it up. But this does not suf¬ce for there to
be a uni¬ed trying of the whole collective, with a unitary content. We
might insist on further restrictions on the individual intentions that can
sum to form a collective intention. But even shared goals do not suf¬ce
for intuitive unitary trying, since we can share a goal but be indisposed
to cooperate together in pursuit of that goal. For instance, we can hate
one another suf¬ciently that although we share a goal, we refuse to work
together to accomplish it. We are speci¬cally averse to trying to do any-
thing together.
Group action in my sense is a fact. We have now located this phe-
nomenon by reference to four richer and three weaker phenomena, and
seen that it plausibly captures the minimum conditions required for literal
trying and intentional action by a group. You may have questions about
some details of this proposal “ for instance, what it is to accept a reason “
but we will return to such details in section V, after the preliminary sketch
of MAC is complete.


The second key fact that undergirds MAC is that there is an overlapping
multiplicity of agents and acts. Let me stress sequentially the multiplicity
and the overlapping whose conjunction constitutes this key fact.
Multiplicity of agency is immediately evident from the nature of group
agency: There are atomic agents, but they are not the only kind. Some acts
are parts of other acts, and not necessarily acts of the same agent. Some
agents are parts of other agents. This means that there is a multiplicity not
only of numbers but also of types of agents. Some agents “ indeed, the most
intuitive “ persist longer than atomic agents. There are various temporal
scales of agents in the world “ individuals in moments of their lives, during
longer periods of their lives, or over whole lives. These are perhaps only
somewhat unintuitively (though accurately) considered group agents. But
some agents are intuitive group agents, and of various types and scales.
There are group acts of families, friendships, corporations, departments,
and universities. All of these types of agents are real, and all perform real
actions from among real sets of optional alternatives.

34 Jackson (1987); T¨ nnsj¨ (1989).
a o

But this multiplicity is overlapping. Atomic agents in some sense con-
stitute all agents. And particular atomic agents constitute at once parts of
more than one group agent. You at this moment are at once part of a
continuing person, and also of various organizations, families, and friend-
ships. Indeed, even at a single moment you may be acting in some way
that is relevant either by omission or commission to the group acts of all
these groups. What you say now may at once support your career and an
important family project but undercut your department and the crucial
goal of your friendship with Y. In that way, it may at once violate your
momentary role in certain group acts and ful¬ll your momentary role in
All the multiple agents of which you are now part overlap in the atomic
agent that you are at this moment. And atomic agents characteristically
form parts of overlapping but real agents of different types. This over-
lapping multiplicity is re¬‚ected in the reasons accepted by atomic agents,
which help constitute them as parts of more than one group agent all at
once. You accept at this moment some reasons that help constitute the
you of the moment to be part of a persisting agent with persisting activi-
ties and plans, but also some reasons that help constitute you to be part of
a cooperating department, and others that help constitute you to be part
of a cooperating family. But we will come back to that complexity.
The key non-normative facts that support MAC are now before us.
We will also presume that there is one basic normative fact, the truth of a
consequentialist normative principle that stipulates an ordering of possible
options from worst to best. It might be HMP or a classical utilitarian prin-
ciple, but it need not be. This provides the basic normative information
that consequentialist evaluation requires, and also the familiar normative
rationale for direct consequentialist evaluation. I will argue in this and the
following section that these two classes of facts together require Multiple-
Act Consequentialism. I will begin by sketching the principal rationale
for the two key tenets of MAC. In the next section, I will discuss pressing
questions about the probity of this rationale.
The characteristic normative tenets of MAC are principles governing
defection from group action and for balancing con¬‚icting forms of group
action. Identifying and motivating these tenets is our immediate goal.
MAC speci¬es right acts for atomic agents, among other things. These
acts are constituted as right by direct consequentialist evaluation “ in other
words, by the application of our basic normative principle for the speci-
¬cation of acts as best among available options. But this direct evaluation

must be performed all at once on the options of the multiple genuine
agents who overlap in a given atomic agent. This creates the possibility
of con¬‚icts, and so we must work toward some mechanism to adjudicate
those con¬‚icts. A form of consequentialism that does not provide some
mechanism to resolve these con¬‚icts does not provide a coherent crite-
rion of right action. And such a mechanism is what the characteristic
normative tenets of MAC provide.
MAC is a form of consequentialism. So its basic principles for adju-
dicating con¬‚icting forms of agency that overlap in a given atomic agent
should be closely rooted in our basic consequentialist normative principle.
That is the method or class of methods with a straightforward consequen-
tialist rationale. The two crucial normative tenets of MAC rest on this
One key and characteristic situation is this. Imagine that an atomic
agent may, as an individual, defect from a group agent with a good end,
a group project that will not be undercut by that single defection. And
imagine that in defecting, that atomic agent can grab some extra posi-
tive consequences on the side. We are all trying to save the passengers
of a sinking ship, but I notice that the rest of you don™t really need
my help to do it. And I can defect to do some relatively minor posi-
tive good instead “ say, polish the treasure box. This is a case of surplus
There are a variety of possible consequentialist responses to this sort
of case. One response is that of the traditional act consequentialist: The
atomic agent should defect. Another is that of a rigid group consequen-
tialist: The atomic agent should not defect.
Those are perhaps the natural responses, respectively, of someone who
thinks that only actions, options, and agents at the scale of atomic agency
are real, and of someone who thinks that only actions, options, or agents
at the scale of groups of atomic agents (say, those constituting intuitively
persisting individual agents with ¬xed intentions) are real. But the position
of MAC is that all these acts are real and overlapping. Neither defecting nor
cooperating has a more direct consequentialist rationale, because neither
the group act nor the defecting individual act has a more direct rationale.
So we have a real con¬‚ict here.
Still, it might be thought that the con¬‚ict is to be automatically resolved
in cases of this sort in one way or the other. Perhaps the possible defect-
ing act always dominates, or perhaps the group act always dominates,
irrespective of the relative normative weight of their consequences. Of
course, it wouldn™t be appropriate simply to ignore either the group act or

the individual defecting act, since both sorts of acts are real. The cases in
which an atomic agent should defect from bene¬cent group acts and the
cases in which it should not defect ought to re¬‚ect in some way the rel-
ative normative weight of the independent atomic agent and project, on
the one hand, and the group agent and project, on the other, as assessed
by our basic normative principle. But the question is, how should it
re¬‚ect that relative weight? If the background of cooperative activity is
preserved when ¬guring the options relevant to assessing defection, then
that defection gains a little on the side and loses nothing. And so it seems
that consequences are served if we defect. But if instead the action of the
atomic agent is taken as ¬xed, with the cooperative activity of the group
as the only variable item with normative weight, we get a predictably
different answer.
The crucial question is what to properly leave ¬xed and what to take as
properly variable when assessing the situation. The truth is that we must
properly let both relevant factors vary, the group act and the individual
act. It is those two alternatives that the atomic agent will pursue or shun
by its choice, since to pursue a group act is to perform one™s role in
that act and to defect from a group act is to fail to perform it. And it is
only by letting both acts vary that we can assess their relative normative
weights. Properly motivated direct consequentialism requires that both
acts be ceded relevant weight.
Here is a ¬rst pass at how to do that: Compare two situations. In
the ¬rst situation, the form of cooperative agency in question does not
exist, but the atomic agent pursues its alternative defecting project. In
the second, the form of cooperative agency exists, but the atomic agent
does not defect. The basic consequentialist normative principle will tell
us which is better. If it is the ¬rst, then MAC says to defect. If it is the
second, then MAC says not to defect.
That is approximately right, but it must be only a ¬rst pass, because
there is too much indeterminacy about what would be the case if the form
of cooperative agency in question did not exist. Should we consider the
possibility that none of the other atomic agents that constitute the group
exist, or that they exist but fail to constitute a group agent, or that they
constitute a group agent but adopt some alternative cooperative activity,
and, if so, which alternative cooperative activity? A second natural pass
is the suggestion that all of the possibilities matter in which either the
atomic agents exist but fail to constitute a group agent, or alternatively in
which they constitute a group agent but perform some alternative project.
A defector is in essence defecting from two group actions, that which

constitutes the group agent, and also the particular action of the group.
To the ¬rst defection, the existence of the group agent is the relevant
alternative; to the second, the options of the group agent are relevant.
But this second pass is unworkable. There are a number of alternative
options that the group agent might have taken, with different values. And
at least many of the options that the group agent would not have plausibly
taken seem irrelevant to any balancing judgment. And there may not be
a fact about which particular option it would have taken if it hadn™t taken
the one that it did, nor even determinate probabilities that it would have
variously taken various alternatives. And, in any case, the complexity of the
suggested comparisons is quite daunting, and indeed apparently beyond
the capacity of familiar consequentialist valuation principles, including
So we need a third pass, which heeds the ambiguities of the ¬rst pass
but avoids the unworkable complexity of the second. We need a single sit-
uation that can stand in for defection. Since in a case of genuine normative
con¬‚ict between defection and adherence, the group agent in question
can be presumed to have a properly bene¬cent project, and since the only
group project with special salience is the one actually adopted, the situa-
tion relevant to defection is either that in which the group agent doesn™t
exist, though the atomic agents that make it up do of course exist, or
that in which it exists but does nothing. But as far as I can see, these are
equivalent. So here is a third and ¬nal pass:
To properly assess the con¬‚ict presented earlier, we should compare a
¬rst situation in which the atomic agent achieves what it can by defection
but in which the various other atomic agents that in fact constitute the
group agent do not constitute such an agent, to a second situation in which
the group agent acts as it does and the atomic agent does not defect. If the
¬rst situation is better, then MAC says to defect. If the second situation is
better, MAC says not to defect. And our basic consequentialist principle
will determine which situations are better than which. I will call this
conception of how to handle these con¬‚icts the principle of Very Little
Defection, or VLD. As I put it earlier, one should defect from a group act
with good consequences only if one can achieve better consequences by the defecting
act alone than the entire group act achieves. That is the ¬rst characteristic
normative tenet of MAC.35

35 There may still be indeterminacies about what would be true on each of the counterfac-
tual conditions deployed in the test. These will create genuine normative indeterminacies
according to MAC.

Some indirect forms of consequentialism violate the natural norma-
tive rationale of consequentialism because they appeal to the merely
hypothetical consequences of other people doing what in fact they will
not do. But all consequentialist assessment depends on hypothetical claims
of some sort “ for instance, regarding what would or would not ensue if
a particular act were performed or not performed. According to VLD,
we assess the weight of both the group and the individual act in question
by reference to the effects of its particular presence and absence, in the
traditional direct way. Greater normative weight of that straightforward,
well-motivated sort determines which act is properly dominant according
to MAC.
In the case under consideration, defection will not disable the group
project, and will yield extra positive consequences. So it is natural to worry
that defection has a more direct consequentialist rationale, that properly
motivated direct consequentialism does not require that we assess the
relative weights of the group act and the defection in the way I have
But recall that individual acts that require a series of temporal steps
are group acts in my sense. And remember Casta˜ eda™s “paradox” of act
consequentialism. The temporal conjunction of two acts of one individ-
ual can have excellent consequences even when each conjunct performed
alone would have horrible consequences. But commitment to direct con-
sequentialist rationales surely does not forbid bene¬cent individual acts
that are temporal conjunctions of acts that are not individually bene¬-
cent; and in fact, one popular response to Casta˜ eda™s point has been to
conclude that such conjunctive acts are the only proper locus of direct
consequentialist evaluation. Likewise, commitment to direct consequen-
tialist rationales does not forbid participation in larger-scale group acts in
which one™s particular role is not individually bene¬cent. Of course, our
case involves what Gibbard calls surplus cooperation, in which the group
act will not be disabled by one™s individual defection. But that may be
true for all individual participants in a very weighty group act. The very
existence of many important group acts, including temporally extended
acts of persisting individual agents, requires that atomic agents not defect
in cases of the sort under consideration.
The smallest agent is surely not always the dominant locus of direct
consequentialist evaluation. You can try to do distinct things with vari-
ous limbs and ¬ngers. Perhaps, at least with practice, you might become
capable of enough dexterity that we could literally speak of your vari-
ous limbs and ¬ngers trying to do various things. But it would not be

appropriate to require that each of your limbs and ¬ngers act individually,
as much as possible, in direct accord with the basic consequentialist nor-
mative principle, so that you become incapable of uni¬ed action even in
a moment. Direct consequentialist evaluation of all our actions of various
types instead plausibly requires the sort of balancing that I have pro-
posed. It is just that we aren™t used to seeing that this rationale extends
beyond cases of intuitively individual action such as those invoked by
Casta˜ eda™s paradox, that it also encompasses group actions incorporating
many people.
There is another characteristic normative tenet of MAC. We also
need to balance con¬‚icting forms of cooperative agency. And here the
considerations we have already surveyed support the following proposal.
We can assess the relative importance of two forms of group agency by
ranking two situations: In the ¬rst situation, the ¬rst group agent doesn™t
exist because the atomic agents in question fail to properly constitute such
an agent, but the second group agent has its actual form. In the second
situation, the second group agent doesn™t exist, but the ¬rst has its actual
form. If the ¬rst situation is better according to our basic consequentialist principle,
then the second form of agency is more normatively signi¬cant, and MAC tells
us not to defect from that group agent. I will call this the principle of Defect
to the Dominant, or DD. It is the second characteristic normative tenet
of MAC.


The last section sketched the main rationale for the key normative tenets
of MAC. It rests on the facts of overlapping multiplicity of group agency
and the presumed truth of some basic consequentialist normative princi-
ple. But those facts may seem to provide even better rationales for various
alternatives. So, to buttress my argument, let me consider a few key con-
trasts and objections.
I™ve mentioned two already. One way of responding to the kinds of
con¬‚ict cases presented in the last section is to claim that the con¬‚icting
acts of both agents are to be evaluated in regard to their own consequences.
So an atomic agent acts rightly if it produces the most good it can as
an atomic agent, and a group agent acts rightly if it produces the most
good it can as a group agent. This means that there are situations in
which not all agents can act rightly, in which if an atomic agent is to
produce the most good it can, then the group agents of which it is part

cannot.36 But that, it may seem, is just the way it is sometimes. The
situation may seem analogous to a moral dilemma.
But, I reply, any practical ethics must resolve these con¬‚icts in one
way or another, whether they count as moral dilemmas or not. So we are
forced to some other treatment of these cases.
Another key contrast bears repeating. All agents overlap in atomic
agents. So, in a sense, consequentialism addresses its demands to atomic
agents in particular. Hence it may seem that in the con¬‚icts I have pre-
sented, atomic agents should be concerned ¬rst and foremost with the
good that they can individually produce. In the ¬rst con¬‚ict case I pre-
sented, they should always defect, contrary to VLD. And there is another
natural motivation for this alternative. That ¬rst con¬‚ict is one of surplus
cooperation. There is extra good on the side that defection will achieve,
and the group act will still achieve its goal. So consequences overall will
be better served by defection.
But, I again reply, it is important to remember that you and I aren™t
single atomic agents. Since continuing individuals are sequences of atomic
agents, and constitute group agents in my sense, and since there are con¬‚ict
cases of the sort in question in which each atomic agent that makes
up a continuing person with a continuing project might have equally
bene¬cent motives to defect while presuming on the continued pursuit
by the other atomic agents of the continuing goal, this proposal undercuts
the conditions required for the very possibility of effective individual
action over time. Since you and I are not single atomic agents, to free
atomic agents for individual maximization of the good in the way this
alternative suggests is to make our intuitive continuing individual pursuit
of the good often impossible. This is not merely an argument from analogy.
The conditions required for pursuit of individual projects over time that
are normatively protected by MAC and that rationalize that protection
are the same as those conditions that protect group acts that involve more
than one person.
Still, we should consider other ways to resolve the con¬‚icts in question.
It may seem that MAC provides the dominant group act with too much
protection. Consider not just VLD but DD. It may be that one™s choice
is between defections from two group acts. Imagine that the ¬rst group
act has better consequences, but that those consequences will also be

36 Postow (1977) takes this line.

achieved if I defect. But the second act also has good consequences, and
those consequences will not be achieved if I defect. And since MAC
tells me to defect from the second group, that may sometimes seem
inappropriate.37 Consider this particular case:38 Two disasters threaten
humanity, Big Disaster, which is that we will all be boiled in oil for
the rest of our lives, and Small Disaster, a small but nasty war. If a cer-
tain group agent in which I participate takes action, Big Disaster will be
averted, though in the normal course of things Small Disaster will occur.
But if I secretly defect from my role in that group, I can see to it that
Small Disaster is avoided, through an individual project that takes time
and hence is a smallish group act, and the big group will go ahead and
achieve its goal without me. At least in some cases of surplus cooperation,
it seems, we should defect from the weightiest group act.
MAC deploys different replies for different cases of this general sort. In
the main and simple case, where what is achieved by defection is minimal
and clearly forbidden by conditions required for being a nondefecting
participant in the big group act, the motivational and other psychological
conditions required for being a participant in the group action will gen-
erate criticism and acceptance of criticism of such defections. And then
MAC will demand nondefection. Since consequentialist assessment of
actions requires actions, this is not suf¬cient grounds for complaint against
MAC. But there are other cases of the general sort under consideration
in which defection clearly seems appropriate. And I think that those are
cases, like that of the two disasters, in which we cannot imagine anyone
even remotely like us who participates in the big group agent complaining
about the alleged failure of reciprocity, or even having the suppressed dis-
position to do so. The participants in such a group agent are not disposed
either to provide or to accept criticism for that sort of so-called defection.
So it isn™t really defection at all. The project of the group involves an at least
implicit exception for the prevention of unnecessary disasters.39 Actual
group agents involve plentiful implicit exceptions of this sort, I think.
That in turn may generate another worry. Why can™t we say in any
case of bene¬cent defection that the group is operating to take advan-
tage of unexpected opportunities, so that in fact the group action is
preserved and generates better consequences than would otherwise be

37 Tom Carson ¬rst brought these cases to my attention.
38 I owe this case to Torbj¨ rn T¨ nnsj¨ .
o a o
39 Hooker (2000) develops this Brandtian move in a rule utilitarian context.

Again, the test is the nature of the criticism that the participants are
disposed to give and accept, which helps determine the exact nature of
the group action involved. When apparent defections are criticized, then
there is no implicit exception.
It may seem troubling that MAC makes issues of right action for an
individual turn not only on the consequences of their individual actions
but also on the group acts that they are inside, and hence on the details
of their motives and the reasons they accept. That delving into motives
may seem hard to square with any straightforward consequentialist
rationale. MAC treats the following two cases asymmetrically. In the
¬rst case, defection from a bene¬cent group agent will yield some
extra bene¬t on the side. In the second case, defection from a merely
bene¬cent group practice, which does not meet the motivational and
other psychological tests I have proposed for group action, will yield
some extra bene¬t on the side. Even if the consequences at stake are
the same, in the ¬rst case MAC will forbid defection, and in the second
it will require it. And the difference turns on the reasons accepted by
the individuals in question and their motives. It seems not to turn on
But, I again reply, direct consequentialist assessment of actions requires
that actions exist. And group actions require the extra conditions in
question. So the difference in question is normatively relevant.
It may also seem troubling that MAC makes issues of right action
for an individual turn on details of what other people do or don™t do,
and even on details of their motivations and accepted reasons. But the
point is that you and I are literally part of group agents that include
other people. Just as there are group agents that consist of many temporal
periods of one person™s life, so too there are group agents that consist of
many people. So agents are not as separate as this worry presumes.
MAC, as so far characterized, requires signi¬cant respect for the
demands of bene¬cent group actions of which one is already part. How
does it tell us to determine which group acts to join? By straightforward
consequentialist assessment of the effects of joining, as opposed to
available alternatives, constrained of course in the usual way by the tenets
of MAC. So MAC treats group acts of which we are already part and
other group acts, which are merely possible or which actually exist but
of which we are not part, in an asymmetrical way. And this may seem
grounds for complaint.
The exact nature of this asymmetry according to MAC can be misun-
derstood, since future cooperators can be relevant parts of existing group

agents according to MAC, as for instance the treatment of continuing
individual projects that I have proposed requires. “Existing” is to be read
timelessly.40 But still there is an asymmetry. If A and B are engaging in
a group action that produces a good result that would also be produced
if C joined, but C can produce a little bene¬t instead of joining, MAC
tells C not to join. But if C joins, then MAC will tell C not to defect,
even if C can defect in time to produce that little bene¬t on the side.
But again, the integrity of the actions in question is what requires
this. And it seems that the particular psychological conditions that group
actions in my sense require allow MAC to track our intuitive reactions
to such cases quite well. We react somewhat differently to defectors
from ongoing bene¬cent group acts than to people who never join.
There are different sorts of failures of reciprocity involved. But to see this
properly, we will need to consider further details of MAC and its ethical
applications. And we will need to turn more explicitly to issues about


The most characteristic features of MAC are now in place. You should
defect from bene¬cent group acts only when you can create more value
by the defection alone than the entire group act creates, and when such
group acts con¬‚ict you should defect from the act that creates less value.
And group action exists when there is common action by a number of
agents rooted in common true belief that there is a shared goal, and in
acceptance by all the members of the group that there is a reason to con-
tinue to coordinate activity until the goal is adequately accomplished, or
inde¬nitely if it is not the sort of goal that will be ¬nitely and de¬nitely
accomplished, a reason whose acceptance we can expect to occasion crit-
icism and the acceptance of criticism for failure to continue coordination
until the point (if any) at which the goal is accomplished or mutually
abandoned, and which is supported by entwined motives of bene¬cence
and reciprocity.
That is the minimum mechanism that is necessary to understand the
main thrust of my application of MAC in the next chapter. But certain
details of those applications, and some of our later discussions, will turn

40 The fact that future atomic agents who are a part of your future are inside a group act does
not imply that you at this moment are already inside it. It does not imply this even if you
at this moment and those future atomic agents are in fact parts of other group acts.

on two details of MAC that are the focus of this section. And you may
have some questions about those details anyway.
First detail: Group action in my sense is rooted in common true belief
about shared goals and also in shared accepted reasons, and that common
true belief is also at least largely about accepted reasons. So you may
wonder what it is to accept a reason. The answer is that it is something
very like Allan Gibbard™s conception of what it is to accept a norm,
developed in Chapter 4 of Wise Choices, Apt Feelings.41 But not quite.
Gibbard distinguishes three notions, those of accepting a norm, inter-
nalizing a norm, and being in the grip of a norm. While none of these
three is exactly what we need, all are close. Consider two cases of moti-
vational con¬‚ict that involve different forms of weakness of will. In the
¬rst, you think you ought to stop eating nuts at a banquet, but go on
eating them. The reasons and norms you accept are in con¬‚ict with an
appetite for nuts. In the second, in a Milgram-type obedience experi-
ment, you administer the lash to someone at the direction of some suit-
ably impressive authority though you feel some con¬‚ict about that, and
yet outside of such an experimental situation you would readily assent
to norms that forbid you to do such a thing.42 In the ¬rst case, there is
a con¬‚ict between reasons or norms you accept and other motivations.
Gibbard suggests that in the second case, there is con¬‚ict between the
norms you accept and the norms of politeness and submission to impres-
sive authority that you are in the grip of during the experiment, and that
both norms that you accept and norms that you are in the grip of count
as internalized norms. But he also says that nonlinguistic animals like dogs
can internalize norms, and that accepting a norm involves paradigmatically
human linguistic phenomena, that it is a kind of governance by norms that
one would express in spontaneous and sincere normative discussion and
that would then be subject to conversational demands for consistency and
My notion of an accepted reason is the notion of something that is
linguistically mediated like Gibbard™s accepted norms, and yet something
that you can be merely in the grip of in something like his sense. Dis-
tinguish two sorts of reaction in a Milgram situation. In the ¬rst case,
you obey the authority just as a dog would obey. In that case, you accept
no reasons supporting obedience in my sense. But in the second and
more realistic case, various linguistic pronouncements involving the need

41 Gibbard (1990b: 55“82).
42 Ibid., 56“61; Milgram (1974).

to obey authority and to be polite bubble through your sensibility or
unconscious in causally important ways, and you obey, even though you
would not re¬‚ectively endorse those reasons in ordinary circumstances.
In that case, you accept reasons supporting obedience in my sense, but do
not accept corresponding norms in Gibbard™s sense.
These minor analytical differences answer to the different demands of
our theoretical projects. Gibbard is after an analysis of what it is to call
or believe something rational. That is a rather stately and all-in notion.
But here we need to identify various forms of perhaps con¬‚icting group
agency in which an atomic agent can participate, and which may have
different degrees of motivational ef¬cacy even in a single person at a single
moment. Hence Gibbard™s notion of accepting a norm ought to be more
stately and all-in than my notion of accepting a reason.
There is another contrast worth noting. Reasons that are accepted
needn™t even be held as prima facie reasons in any very robust motivational
sense. We have some tendency to act on reasons that are accepted, but it
may not be a very strong tendency. It may be easily swamped by other
There are various ways in which the acceptance of reasons mediates our
acts, and indeed in some sense it mediates all our acts. First, when human
action is psychologically subject to guidance by normative principles in
any full and familiar sense, it is governed not just by external sanctions but
also by accepted reasons, and adult human action almost always is subject
to normative principles in this way. Second, acceptance of certain sorts
of reasons helps to constitute group activity of various sorts, as we have
seen. Third, even in the absence of group action or obviously normative
governance, accepted reasons play a role in our action, as deontologists in
the tradition of Kant have long recognized. A human action in any full
sense, an action that is intuitively fair game (unlike tics or re¬‚exes) for
normative evaluation, is governed by the reasons accepted by the subject,
which help to constitute the action as intentional.
This is not to say that there are no limits to the ef¬cacy of accepted
reasons in us. There can be con¬‚ict in the reasons we accept, and different
degrees to which we accept certain reasons. And sometimes we are torn
by non-normative passions like hunger, and behave in a way contrary to
all the reasons we accept. And sometimes we act out of habit or re¬‚ex
and not intention. Indeed, it may be that it is only after a certain degree
of relatively contingent training and maturity that we come to accept
reasons at all. But accepted reasons are important in our actual lives as
human agents.

And this is not to say that hungry ¬sh or sharks, who accept no reasons,
do not act. I think that wherever there is desire, there is action, and
that sharks often have desires. But when animals capable of language
accept reasons, then I claim that their agency is centrally and dominantly
constituted by that “higher” capacity (which of course in turn plausibly
depends on the existence of the capacity for mere animal desire). Our
human intentions can be betrayed by our fear or hunger. Not so the
attempts of a shark or a gold¬sh, I think. And even if you think that
a long pursuit undertaken by a shark can be betrayed by a momentary
passion without being abandoned, still the way in which that shark project
is betrayed is not the same as the way in which the continuing individual
project of a reason-accepting animal can be betrayed. At the moment the
shark defects, it isn™t obvious which is the dominant project. And while
there are even forms of roughly intuitive group action that do not involve
accepted reasons “ which, for instance, dogs and some ¬sh exhibit “ still
for animals who accept reasons, that “higher” capacity is where genuine
group trying is crucially rooted.
And in any case, ethics can be practically relevant to action only when
that action is either directly or indirectly governed by the acceptance of
reasons. That is where ethical discussion can bring pressure to bear.
The second detail of the mechanism of MAC that we need to consider
is related to the ¬rst. There are different roles in MAC for the reasons
that agents accept. Some of the reasons that atomic agents accept help
to constitute the existence of effective group agents with one speci¬c
project or another. I will call these “agent-constituting” reasons. Other
reasons that they accept specify abstract forms of normative governance to
which they assent. These are “agent-governing” reasons. Clearly also they
must accept reasons (or have some other method) to resolve the con¬‚icting
demands of the various agents of which they are part, con¬‚icting demands
that are re¬‚ected in the con¬‚icts of their agent-constituting reasons. These
are “agent-balancing” reasons when suitably speci¬c. Note that there
are plausible agent-governing reasons that specify proper forms for both
agent-balancing reasons and agent-constituting reasons. They are higher-
order reasons. And note also that agent-governing reasons must provide
some direction about what group acts to join.
MAC speci¬es all of these kinds of reasons at once in their proper form
by direct application of a basic consequentialist valuation principle, and
in a manner re¬‚ecting the two characteristic normative tenets of MAC.
But let me say a little more about the three kinds that will be relevant to
some details of our following discussions.

The principal role of agent-constituting reasons in MAC is the consti-
tution of agents and actions. It is largely in the accepted reasons of atomic
agents that group agency is rooted in the world. An atomic agent must
not act merely from among its momentary options on the basis of the
basic consequentialist normative principle. MAC will demand that it also
act on the basis of various principles that constitute appropriate forms of
group agency. Direct consequentialist choice by a group agent involves its
atomic agents in acceptance of agent-constituting reasons that are different
than the basic consequentialist normative principle. As we will eventually
see, agent-constituting reasons correspond quite closely to familiar and
intuitive individual virtues and also to familiar and intuitive deontological
restrictions on action. They are a way in which MAC allows a role for
recognizably Kantian respect among cooperators.
There are, at least in the case of some group agents, two kinds of
agent-constituting reasons. In these agents, the existence of a particular
group agent “ say, a corporation “ is distinct from its possessing at least
certain group projects. Acceptance of reasons that give such a group a
particular project will constitute a more re¬ned and speci¬c sort of group
agent, so they are still agent-constituting. But I will call them in particular
“project-constituting” reasons.
There are limits to the appropriate role of agent-constituting reasons
according to MAC. Not all of the group actions and group agents of
which an atomic agent is part represent genuine normative demands on
that atomic agent according to MAC. That depends on the propriety and
weight of the projects of those group agents. Agent-constituting reasons
characteristically demand a kind of Kantian respect for fellow cooperators.
But Kantian respect even for cooperators is limited according to MAC. For
instance, horri¬c group projects of which we are part should not engage
our Kantian respect. And various bene¬cent group projects of which we
are part should engage more of our Kantian respect than others. The core
of an account of proper projects and their weight is provided by the basic
consequentialist valuation principle that we now assume and by our two
key normative tenets, VLD and DD. Those principles are re¬‚ected in
certain proper agent-governing reasons. But, as we will see, these are not
the only appropriate agent-governing reasons.
Agent-governing reasons specify proper choices in a suitably abstract
way. They govern appropriate agent-constituting reasons from above, and
specify from above proper mechanisms for balancing forms of agency.
There are complexities involving proper projects of agents. For one
thing, very good choice from among a set of options may not be choice

of the best of those options. Not all options other than the best have the
same evaluative status. Some are intuitively close enough to being best
to count as pretty ¬ne. In general, we need a rule that tells us when a
choice or project is “proper” from among its options, proper in what-
ever normatively salient ways there are that can be rooted in our general
normative principle. But we already have enough complexity regarding
proper projects for our purposes, because of a detail I will note in the
next section.
Another important complication is that agent-governing reasons must
provide some direction about what group acts to join. But I™ve already
noted that atomic agents (among others) should join group acts on direct
consequentialist grounds, just as a direct consequentialist rationale would
suggest. An atomic agent should join a group act when its individual
contribution to the group act will generate the best consequences available
to that atomic agent from among its various individual options, presuming
that it has no countermanding obligations according to MAC.
The third kind of reasons deployed by MAC are agent-balancing rea-
sons. The actual behavior of atomic agents who accept reasons that con-
stitute con¬‚icting and overlapping group agents will depend in part on
the degree to which they accept the various reasons involved, and in part
on the degree to which those reasons engage their motivation. But MAC
must provide normative guidance regarding the proper way to resolve
con¬‚icts among con¬‚icting agent-constituting reasons. These are agent-
governing reasons when suitably abstract and higher-order. But when
the acceptance of reasons constitutes an actual balancing by an atomic
agent of speci¬c forms of group action in which it partakes, the higher-
order agent-governing reasons that prescribe a certain abstract form for
appropriate balancing have a ¬rst-order echo in accepted action-balancing


There is one remaining crucial detail before we turn to the application of
MAC. As we will see in the next chapter, the actual existence of certain
sorts of group agents is the key to MAC™s response to standing objections
to act consequentialism. It is uncontroversial that there are normatively
weighty forms of cooperative practice with the nature I will deploy. For
instance, the general practice of most people is to refrain from murder,
and that serves a mutually bene¬cial goal. But the crucial fact for MAC
is that there be corresponding group acts in my sense.

There are. For instance, refraining from murder is a negative group
act. And yet it is an act, which we can try together to do. Such group
cooperation is also what I will call a “one-off group act”. The particular
form of cooperation adopted by just those atomic agents simply exists or
it doesn™t; it involves only those two relevant options. And it is suitably
good-producing, since if it does exist, that is better than if it doesn™t. While
at least some armies have some recognizable existence as group agents
independent of shared commitment to at least some particular campaigns
they are undertaking, the crucial group agent that refrains from murder is
constituted by commitment to that speci¬c project. There is no distinction
between project-constituting and other sorts of agent-constituting reasons
for such a group. For cases like these, as long as it is clear that the existence
of such a group agent is better than its nonexistence, we needn™t worry
about the ¬ne details of proper projects according to MAC.
I believe that many cooperative moral practices are one-off group acts
that are in fact bene¬cent, which in fact have good consequences. These
general normative practices are analogous to Aristotelian processes, and
have no ¬nitely achievable and distinct goals. But they have goals, and
indeed normative goals, though they are sometimes negative goals. They
have goals in a more robust sense than coordinated dog activity, since
like all full-blooded human actions they are modulated by the acceptance
of reasons. But they are also undergirded by motives of bene¬cence and
If I am right that these normative practices are group acts, then we will
have two sorts of motivational involvement in such practices, modulated
by and supporting accepted reasons, and this will be mutually recog-
nized in certain at least hypothetical ways. And indeed, those two sorts
of involvement will be entwined. Such a particular group act, a group
act that is a kind of normative practice, will have in particular a norma-
tive goal, an even more obviously normative goal than other group acts
that aren™t normative practices, which will occasion one sort of involve-
ment. And it is a cooperative group act, which all by itself will occasion
another sort. And indeed, in group acts these two sorts of involvement are
entwined. We should expect criticism of defections from such group acts
of two different but entwined sorts, supported by motives of bene¬cence
and reciprocity.
As I suggested in Chapter 1, benevolence and reciprocity are both basic
moral motives. And benevolence is a certain sort of bene¬cence. In our
broadly bene¬cent standing moral practices, benevolence and reciprocity
are entwined in the way characteristic of group action. Many weighty

normative practices meet the motivational conditions required to consti-
tute group acts.
A case may make my meaning clearer. A murderer kills people, and that
is a very bad result, which violates our sense of what proper benevolence,
or at least bene¬cent nonmalevolence, requires “ a proper goal that we all
ought to share, and that at least most murderers often share. The murderer
fails to properly acknowledge how bad it is for someone to meet a violent
death. But the murderer also defaults from our important general coop-
erative practice of not murdering innocent people in peacetime, and to
do that seems unjust and unfair. It violates our sense of proper reciprocity,
and even makes some think he deserves himself to be killed. He defaults
from his fair and ordinary role in a group agent. And if he is never inside
that group practice “ well, then he is merely a vicious animal, some feel.
We will return to that point in a minute.
Notice that our motive of reciprocity is entwined with our motive of
nonmalevolence. In such a case, we will criticize the defecting murderer
out of both sentiments, which are entangled and mutually reinforcing.
This case also reveals the mutual recognition of cooperation that group
action requires. Our murderer knew that people don™t go around killing
people, at least outside of situations like war, and that we all aim at the
good of avoiding the violent death of humans. It would be absurd for the
murderer to deny knowledge of this cooperative practice.
The normative practices that I will deploy in the next chapter to dispose
of intuitive objections to act consequentialism are plausibly all of this sort.
Moral motivation, or at least our moral motivation, is generally rooted in
entwined reciprocity and benevolence, or at least nonmalevolence. And
there is the necessary mutual recognition of these forms of cooperation,
supported by criticism of defectors and acceptance of criticism. So the
common moral practices that I will deploy in the next chapter are in fact
group agents in my sense, and indeed usually one-off group agents.
But consider this objection: It may seem that our moral practice of
refraining from murder involves no mutual responsiveness. Each thinks
that they have reason not to murder no matter what others do, and will
continue not to murder even if others do murder. So, it seems, the non-
murderers aren™t a group agent.
But, I reply, there is some ambiguity in what it is to think one has a
reason not to murder no matter what others do. Surely if some people
murder, and even fall outside of the group act of refraining from murder,
we do not conclude that we have no reason not to murder. But that is not
contrary to MAC. Surely there is reason not to murder in most particular

cases even independent of the existence of a group act of refraining from
murder, since murder characteristically is not productive of the good. But
the key question is whether or not there is a cooperative action of not
murdering that involves a sentiment of reciprocity, so that we would be
somewhat less motivated to refrain from murder if no one else in the
world so refrained. And so we would. Another possible test, depending
on the details of our project of nonmurdering and whom it is intended to
protect, is whether we are somewhat less motivated to refrain from killing
those who are outside of that very bene¬cent cooperative practice. And
so, it seems, we are.
There is another worry about this treatment of our moral practices as
group agents. In a one-off group agent, there is no difference between the
existence of the agent and the pursuit of a particular project. But then it is
hard to see how a particular atomic agent could be criticized, on grounds
of the kind of reciprocity I am invoking, for failing to hold up their end
of a group practice. If they murder, then they are outside of the group act
of refraining from murder. Certainly a different past period of your life
might be inside the relevant group agent, but if you at the moment are
not, then you at the moment do not seem subject to criticism of that sort.
While it is possible to be part of a team while not furthering a particular
contingent goal of the team, it seems impossible to be part of the group
practice of not murdering and yet murder. And indeed, even in the case
of the team, you could not properly be criticized for defecting from the
team itself simply because you defect from a particular project of the team.
My reply has three parts, and will involve a little more precision about
the nature of defection from group acts, and hence about reciprocity.
There are at least two relevant kinds of defection from group acts that
involve different sorts of failures of reciprocity, and there is a third kind of
failure of reciprocity that doesn™t involve defection.
In the ¬rst case, you retain all the motivational and psychological con-
ditions characteristic of participation in the group act, but you defect for
the moment from the common action of the group “ say, because of a
con¬‚ict. When there is no con¬‚ict or analogous problem, you return to
common action. And if there had been no such con¬‚ict, you would have
acted in concert with the group. In this case, there is defection that yet
leaves you inside the group agent at least to a degree, and criticism and
acceptance of criticism on grounds of that sort of reciprocity seems appro-
priate. Humans are capable of the degree of motivational complexity such
that some murderers are yet obviously involved in the one-off group act
of refraining from murder, are trying not to murder in some recognizable

sense, even as they murder. This is re¬‚ected in some complexity in our
response to such murderers, so that in one sense they are still inside the
reciprocal practice even when they violate it. And even when we are not
quite prepared to say that they were trying not to murder even as they
murdered, they can still retain a standing af¬nity to the group practice
that generates criticism and acceptance of criticism on grounds of this
¬rst sort of reciprocity.
But there are other sorts of defection. Sometimes defection from a
group act will put one outside of the motivational and psychological
conditions characteristic of that act. One leaves the group. Even in these
cases, one is likely to be criticized on grounds of reciprocity. But one will
no longer accept that criticism as appropriate when it presumes continued
membership in the group. The atomic agent that one now is is outside
the group. But still, the agents who participate in the group act of not
murdering are not merely atomic agents in realistic cases. They include
intuitive continuing individuals who are suf¬ciently uni¬ed to have at
least some persisting project. And they too are not only subject to but
sensitive to criticism of at least a closely related sort, involving a somewhat
different sort of failure of reciprocity. A continuing individual uni¬ed by
one practice can be inside a second practice at one time and yet violate it
at another.
There is yet a third kind of failure of reciprocity that we are inclined
to criticize, one that would apply to those humans, if any, who are always
completely outside the group act of refraining from murder. They fail
to join this group act that they are characteristically obligated to join on
strict consequentialist grounds, even though they bene¬t from it. And in
the case at hand, this failure is so severe that they may seem more like
sharks than like ordinary humans.
Consider one last objection. Because I will regard our common moral
practices, such as telling the truth, or refraining from murder, as a series of
one-off group agents, it is plausible to wonder why we shouldn™t regard
all our everyday moral practices taken together as constituting another
one-off group agent. Given the plausibly massively bene¬cent effects of
this agent, MAC would then almost never countenance defection from
common moral practices, and so would be more conservative than seems
There are two replies to this objection, so to speak from above and from
below. First from above: “Our common moral practices” may be taken
widely, to cover the consensus of all reasonable people over all times and
places, or more narrowly “ say, merely to cover our particular society with

its idiosyncratic current commitments. The dangers of conservatism ¬‚ow
mostly from parochial commitments. But MAC will customarily cede
dominance to the broader and hence weightier group agent. Note also
that the timeless conception of group agency that MAC deploys means
that we may need to consider competing moral practices that span all
time, so that the sorts of conservatism in question must be understood
in the properly nontemporal sense.
But there is also a second reply, from below: In the case where some
component norm of our common moral practice has bad effects, and
there is a group agent that encompasses the full practice, still there is
another group agent that includes the same atomic agents but eschews
that component in its project and has relevantly better consequences on
its own. There are two concentric one-off agents, such that the inner
one with the less determinate project would have better consequences
alone. So the inner agent is dominant according to MAC. Notice that
this agent involves the same atomic agents, so this isn™t a case in which some
individual is defecting in order to gain some more positive consequences
on the side. Rather, the same atomic agents participate in two one-off
group agents, one nested in the other because its project is nested in that
of the other. On the other hand, where each component of commonsense
morality has positive effects in the relevant sense, this reply is not available.
So MAC is basically conservative regarding elements of our common
moral practice that have good effects, whose presence in the common
practice is better than their absence would be.
This reply relies on a distinction between individual defection from
a group agent in pursuit of surplus good, and reforming defection from
harmful elements of a group project. And it may seem that such a distinc-
tion cannot be drawn. Reforming defection involves con¬‚icting group
acts, but so too would individual defection that takes some time. A
reformer may be criticized on grounds of reciprocity by the unreformed.
And individual defection in pursuit of surplus good may be rationalized
by a suitable adjustment in the general project of a group. The project of
the group can be modi¬ed to allow an exception.
But still, there is a relevant difference. There are constraints on what
can plausibly be considered a concentric group agent that are set by con-
straints on what can plausibly be considered a component norm of the
larger group project in question, which rest in turn on what can plausi-
bly be considered the nature of that larger group project. Extensionally
equivalent group projects are not always the same. Tryings involve inten-
sionality and not just intentionality. So not all bene¬cent reforms in the

project of a group agent can be achieved by movement to a concentric
group agent, which simply eschews component norms in the project of
the ¬rst. Individual exceptions are a frequent case in point. And there
is a second relevant consideration. The propriety of the elimination of
a component norm, its being consequentially helpful, will characteris-
tically depend on the cooperative and to some degree general effects of
the group action in question. But an individual pursues surplus good by
a relevantly particular and individual defection, though it may involve a
different sort of group action that constitutes an intuitive individual agent
pursuing an individual project over time. To summarize, there is a differ-
ence between a rule change and an exception for an individual participant,
and not all rule changes are simply the elimination of a component rule.
And so the case of concentric group agents is a rather special case.
With the necessary mechanism in place, and with various immediate
worries defused, we can now return to the three standard objections to act
consequentialism and to a more concrete discussion of how MAC works.

Three Objections

I have given a more or less direct argument for the truth of MAC, which
depends on the presumption that some basic consequentialist normative
principle for ranking outcomes is true. But we must also see if MAC
can evade the three standard intuitive objections to familiar forms of act
consequentialism. And we need a better and more concrete sense of how
it works.


One characteristic objection to act consequentialism is that it is too per-
missive. It sometimes allows or even requires the violation of very intuitive
general moral prohibitions against killing, injuring, lying, and the like.
There are two different aspects of this objection. First, it underlines the
moral signi¬cance of certain sorts of goods “ for instance, life, health, and
true belief “ that may be treated as basic by conceivable consequentialist
normative principles but that characteristically aren™t so treated. Second,
it suggests that the maximization even of such goods as life and health by
murder and injury “ indeed, even of such goods as the relative absence of
murders and injuries “ is often immoral.
The gilded city is a gaudy reef of predators and pain. A wave of your
wand, and it would disappear. Many lives would be lost today. But the
world would be a happier place. Indeed, over time it would suffer fewer
murders, and include more people, since innocent visitors would not


. 2
( 10)