EVOLUTIONARY TRANSITIONS AND ARTIFICIAL LIFE

EVOLUTIONARY TRANSITIONS AND ARTIFICIAL LIFE

John Stewart (jes999@tpg.com.au)
(A later version of this paper was published in the journal Artificial Life [1997] 3: 101-120.)

Abstract

A major challenge for artificial life is to synthesize the
evolutionary transitions that have repeatedly formed differentiated
higher-level entities from cooperative organizations of lower-level
entities, producing the nested hierarchical structure of living processes.
This article identifies the key elements and relationships that must be
incorporated or synthesized in an artificial life system if these
transitions are to emerge. The processes currently included in artificial
life systems are unable to provide an adequate basis for the emergence of
the complex cooperative organization that is essential to the transitions.
A new theory of the evolution of cooperative organization is developed
that points to the additional processes that must be included in
artificial life systems to underpin the emergence of the transitions.

I Introduction

A distinctive feature of living entities is that they are organized as
nested hierarchies: entities are composed of smaller units that are in
turn composed of still smaller units, and so on. For example, human social
systems are constituted by organisms that are in turn made up of cells
that in turn comprise molecular processes.

From an evolutionary perspective, this familiar structure appears to
result from the repeated formation of higher-level entities through the
evolution of differentiated cooperative organizations of lower-level
entities, for example, the formation of early cells from organizations of
molecular processes, the eukaryote cell from complex symbiotic
communities, multicellular organisms from organizations of cells, and
social systems from organizations of metazoans. This evolution has been
characterized by the establishment of an extensive cooperative division of
labor within the organizations of lower-level entities that is associated
with a high degree of cooperative differentiation and cooperative
specialization.

A central objective of the artificial life approach is to synthesize from
artificial components key biological phenomena. If this objective is to be
met, it will be necessary to synthesize entities that are organized as
nested hierarchies, and to synthesize entities that undergo the critical
evolutionary transitions to form differentiated higher-level entities. The
importance of this challenge is widely recognized among artificial life
researchers (e.g., see [17, 28, 32]).

In this article I set out (a) to demonstrate that the processes that are
currently explicitly included in artificial life will not meet this
challenge; and (b) to identify the specific features that need to be
incorporated in an artificial life (alife) system to encourage the
emergence of the transitions to higher levels of organization.

I begin in Section 2 by demonstrating that the processes that are
currently proposed by theory to explain the evolution of cooperation are
limited in their capacity to account for the formation of higher-level
entities through the evolution of differentiated cooperative organizations
of lower-level entities. Section 3 identifies a form of hierarchical
organization that can comprehensively overcome these limitations and that
has underpinned the transitions from molecular processes to cells, from
cells to metazoans, and from metazoans to human societies. This is
followed by consideration of the extent to which this form of organization
has also been significant in the emergence of living from nonliving
processes.

Section 4 notes that this form of hierarchical organization has not been
synthesized in alife systems to date. To assist in identifying how this
synthesis could be achieved, I analyze two illustrative examples of the
evolution of hierarchical organization at different levels of
organization. The article concludes in Section 5 by abstracting from the
examples the key structures and relationships that would need to be
incorporated or synthesized in an alife system for the hierarchical
organization and transitions to emerge.

2 Horizontal Self-Organization

2.1 Cooperative Horizontal Organization

Under what circumstances will cooperative organizations arise within a
population of living entities (e.g., a population of molecular processes,
or cells, or multicellular organisms)? I will first consider the evolution
of what will be referred to here as horizontal cooperative organization.
This is organization in which entities are at the same level of
organization and therefore do not have any capacity to control other
entities within the organization--entities mutually influence each other
in interactions and are unable to influence other entities unilaterally
[29]. This contrasts with what I will refer to as vertical organization,
in which a horizontal organization is controlled by one or more entities
that are in hierarchical relationship to the horizontal organization. The
hierarchical relationship means that the entities are able to influence
the horizontal organization without being influenced by it--this capacity
to influence unilaterally constitutes the ability of the entities to
control the horizontal organization [29]. The controlling entities
collectively comprise what will be referred to in this article as the
manager of the organization.

An organization of entities is constituted by a set of relationships
between the entities. The relationships are in turn constituted by
adaptations of entities. A cooperative organization will arise in a
population where the cooperative adaptations that constitute the
organization are selected and reproduced through time.

Adaptations that establish cooperative relationships between living
entities at the same level of organization can be reproduced through time
where the adaptations provide net advantage to the cooperators themselves.
The advantage may accrue as a direct result of involvement in the
cooperation itself (e.g., mutualism) or may depend on the initial
cooperator gaining the benefit of a further cooperative act that is
initiated by one or more other entities. In the case of reciprocal
altruism, the further cooperative act is initiated by the beneficiary of
the initial cooperation (e.g., see [3, 34]), and in the case of an
autocatalytic set, by some other member(s) of the organization who may not
have benefited directly from the initial cooperative act (e.g., see [11,
14]). At first glance, it may seem that kin selection and related
mechanisms should also be treated as processes of horizontal
self-organization. In general, these operate where the cooperation and its
benefits are disproportionately directed to entities whose propensity to
cooperate is similar to that of the initiator of the cooperation (e.g.,
due to relatedness, as in the genetic kinship theory of Hamilton [15]).
However, on closer examination it is evident that these mechanisms involve
vertical organization; as will be demonstrated in Section 3, kin selection
operates where the horizontal organization of individuals is constrained
and controlled by a lower-level manager comprising the genetic elements
that are common across individuals due to relatedness.

Where the conditions necessary for the operation of these horizontal
processes are appropriately met, cooperative organization will arise and
persist in the population; the entities that comprise the cooperative
organization and the organization itself are able to outcompete
individuals within the population. Where the conditions are met,
organizations can arise that exploit circumstances in which cooperation
provides net benefits, for example, where individuals can provide benefits
to others more efficiently than the others can produce the benefits
themselves (e.g., specialization and cooperative division of labor); and
where individuals refrain from actions that would otherwise benefit the
individual but harm others in the organization (e.g., restraint of
competition that would otherwise degrade resources [16] or that would
reduce overall profitability in an industry [25]).

Taken together with genetic kin selection, these horizontal processes may
appear able to account for the evolution of relatively simple cooperative
organizations such as most of those found among nonhuman metazoans.

2.2 Limitations of Horizontal Self-Organization

However, these processes alone are limited in their capacity to establish
organizations that fully exploit the potential benefits of cooperation:
The processes are unable to overcome fully the widely recognized
impediments to the evolution of cooperation. These impediments arise
because in most circumstances where selection operates at the level of
individual entities, adaptations must compete primarily on the basis of
their effects on the entity exhibiting the adaptation; the effects of an
adaptation on other entities will not usually contribute to the success of
the adaptation, no matter how beneficial its cooperative effects on others
may be, and irrespective of whether the resultant cooperative arrangement
is more competitive as a whole; and in most circumstances, selection will
favor "free riders" or "cheats" that undermine cooperation by taking any
benefits provided by other entities in the organization, without
cooperating in return.

These impediments are not restricted to the gene-based evolution of
cooperation between multicellular organisms. They also manifest at all
other levels of living processes: In relation to molecular processes see
Maynard Smith [23] and Bresch et al. [71; in relation to the cellular
level, see Buss [8]; and in relation to the human social level, see Olson
[25] and Williamson [35].

The processes relied upon by reciprocity theory and by genetic kinship
theory can overcome these impediments only to the extent that they can
ensure that the effects of a cooperative adaptation on others are taken
into account in determining the success of the adaptation: For example,
kin selection is effective only to the extent that the effects of a
cooperative adaptation benefit other individuals that also exhibit and
reproduce the adaptation (e.g., related individuals), and reciprocity is
effective only to the extent that the beneficial effects of a cooperative
adaptation on others are returned through reciprocation to the individual
exhibiting the adaptation. To the extent that the processes fail to ensure
that the effects of an adaptation on others are not captured by the
adaptation, cooperative arrangements that are more beneficial as a whole
will nonetheless fail to evolve.

Of these processes, reciprocity might appear to have the greatest
potential to account for the evolution of cooperation across the various
levels of biological organization: Unlike genetic kin selection,
reciprocity is not limited to circumstances of genetic similarity, and
unlike mutualism, it is not limited to cooperation that is intrinsically
advantageous to all participants. However, reciprocity is susceptible to
undermining by "cheats" (e.g., see [3]). This is particularly the case
where cheats cannot be identified and excluded from the benefits of future
exchanges. Cheating is especially undermining of reciprocal cooperation
where the benefits of a cooperative act are not localized to a few
identified recipients but instead spread to many others in the
organization, making the identification and exclusion of cheats extremely
difficult (e.g., "public goods" in the context of human systems of
exchange relations). This difficulty severely limits the capacity of
reciprocity to exploit fully the benefits of cooperation: Particularly in
complex differentiated organizations, cooperation that benefits many other
entities within the organization could be expected to play a significant
role; and processes that are unable to establish cooperation of this type
will be unable to achieve the evolution of such organizations.

Selection operating at the level of the group where each group is a
horizontal organization is also limited in its capacity to overcome these
impediments; within each group, the evolution of beneficial cooperation
will be impeded as it is in all other horizontal organizations.

In summary, these horizontal processes clearly fall far short of the ideal
of ensuring that all the effects of an adaptation on others (and
ultimately on the organization as a whole) are appropriately and
universally taken into account in determining the success of the
adaptation. Horizontal processes are therefore unable to exploit fully the
potential benefits of cooperative organization and are poor candidates to
account for the evolution of the more complex forms of differentiated
cooperative organizations that have characterized the major evolutionary
transitions that have given rise to new levels of biological organization.

3 Vertical Self-Organization

3.1 The Governance of Living Processes

3.1.1 Management

What arrangements could arise that would overcome the limitations of
horizontal organization and enable organizations to evolve the complex
cooperative relationships that underpin the formation of new levels of
biological organization?

From the analysis outlined above, it is evident that these limitations
would be overcome by new arrangements within the organization that ensure
that the success of cooperative adaptations is determined by the net
effects of the adaptations on others in the organization (and ultimately
their effects on the organization as a whole). To the extent that this
condition is met, cooperative arrangements that provide the greatest
benefit to the organization would prevail.

Stewart [31] has suggested that this could be achieved by the inclusion
within the organization of one or more entities that

· are in hierarchical relationship with the entities that comprise the
original horizontal organization and have the capacity to intervene in the
organization to promote cooperation, for instance, by intervening to
sustain or inhibit entities in the horizontal organization according to
the extent to which their net effect on others either benefits or harms
the organization; and

· are capable of evolving, and whose evolutionary success is dependent on
the success of the organization as a whole. This coincidence of
evolutionary interests between the intervening entities and the
organization as a whole would ensure that the entities evolve
interventions that realize their potential to promote beneficial
cooperation.

These entities that are in hierarchical relationship to the original
horizontal organization collectively constitute the manager of the
organization.

In principle, the manager could intervene in a horizontal organization to
support co-operators who provide benefits to others without benefit to
themselves, and who would otherwise be outcompeted in the horizontal
organization. Interventions of this kind could underpin the evolution of
division of labor between entities in the organization, allowing the
extensive cooperative specialization and differentiation that
characterizes the major evolutionary transitions under consideration here.
Interventions could also inhibit free riders who would otherwise undermine
cooperation arising among other entities. The manager could also produce
net benefits for the organization as a whole by supporting adaptations
that produce only longer-term benefits and that would otherwise be
outcompeted in the short term within the organization.

The manager could vary in the extent to which it overrides the adaptive
capacity of entities in the horizontal organization. At one extreme, the
manager would tightly control the horizontal organization, with all
heritable adaptation originated by the manager (e.g., the genome's
management of molecular processes within the eukaryote cell, and extreme
examples of top-down management in human hierarchical organization). At
the other, the manager would feed back general rewards and punishment to
entities in the horizontal organization to reflect the effects of their
adaptations on the organization as a whole, with the entities taking
account of this feedback as they adapt (e.g., some modern, flexible forms
of human organization). An ideal manager of this kind would cause entities
to adapt as if their effects on others were effects on self, enabling
cooperative possibilities to be explored fully.

Significantly, this vertical organization would not have to rely on
fortuitous synergy between the interests of the organization and the
interests of its constituent entities for the interests of the
organization to be maximized: Instead, an ideal manager would be able to
construct whatever synergy of interests is needed to overcome any initial
conflict between the interests of constituents and the interests of the
organization to enable the organization to adapt optimally as a whole; the
manager would do this by intervening to ensure that whatever adaptations
of entities are needed to meet the interests of the organization are also
in the interests of the entities. Under ideal arrangements, this would
ensure that entities that are pursuing their own interests are also
pursuing the interests of the organization. Once this synergy is achieved,
cooperative relationships that maximize the interests of the organization
would emerge as a consequence merely of the pursuit by entities of their
own interests.

In this way, vertical organization could comprehensively overcome the
limitation in the capacity of group selection to evolve cooperative
arrangements within groups that are each a horizontal organization: An
ideal manager could, in principle, intervene in a horizontal organization
to construct any possible set of relationships between entities and
support any possible types of entities. Group selection operating on a
population of ideal vertical organizations would therefore be unlimited in
its capacity to search the space of possible organizations. It would not
be restricted to searching that subset of the space of organizational
types that contains only organizations limited to the restricted forms of
cooperation that can arise and persist in horizontal organization.

3.1.2 The Hierarchical Relationship

The requirement that the intervenor(s) be in hierarchical relationship to
the original horizontal organization is essential. It is not sufficient
that there be entities within the horizontal organization that have the
capacity to intervene in the way outlined to promote cooperation: An
entity that is a typical member of the horizontal organization and that
uses resources to sustain or inhibit other members of the organization
without any benefit to itself is itself likely to be outcompeted in the
organization (this is the "second-order problem" of Axelrod [1]).

How does the hierarchical relationship overcome the second-order problem?
As we have seen, a hierarchical relationship exists between two sets of
entities or processes when one set influences or constrains the other
without being influenced by it [29]. This capacity to modify without in
turn being modified constitutes the essence of the ability of one set of
processes to regulate or manage another, by, for example, causing the
other set of processes to act or adapt in ways it would not in the absence
of the regulation. The hierarchical relationship that constitutes vertical
organization is fundamentally asymmetrical. This contrasts with purely
horizontal organization in which entities interact dynamically, mutually
influencing each other without dominance or control. n and manage a
horizontal organization without in turn being influenced by it enables the
manager to unilaterally appropriate for its own reproduction and
maintenance resources and services from the horizontal organization. And
it is able to obtain these benefits without having to participate in the
competitive interactions and cooperative exchanges of the horizontal
organization. This enables the manager to stand outside and act across the
dynamical interactions of the horizontal organization, managing them for
its own benefit.

This capacity to obtain resources and services unilaterally is critical
because the capacity assists in ensuring that the evolutionary success of
the manager is advanced by its ability to produce beneficial cooperative
arrangements in the horizontal organization. The capacity to appropriate
resources does this because it enables the manager to benefit from any
beneficial cooperative arrangements supported by its interventions: It can
harvest benefits and have them utilized for its own purposes. The
coincidence of interests established in this way between the manager and
the organization as a whole will be complete when the manager is fully
dependent on the reproduction of the organization for its own reproduction
and when the only way in which the manager can pursue its success is by
enhancing the success of the organization as a whole.

This contrasts with the situation of a member of the horizontal
organization that encounters the second-order problem: The member can
sustainably engage in interactions that promote cooperation only to the
extent that it benefits from these interactions; if the interactions
themselves provide insufficient benefit to the member, then, unlike the
manager, it has no capacity to sustain its involvement in the interactions
by unilaterally harvesting from across the organization some of the wider
benefits that may flow to the organization as a whole from its promotion
of beneficial cooperation.

The manager that constrains the horizontal organization to produce
beneficial cooperative arrangements may be either an upper-level manager
that is external to the controlled entities, or a lower-level manager that
is internal to the controlled entities.

3.1.3 Upper-Level Management

The constraints provided by an upper-level manager are termed boundary
conditions by Salthe [29]. Key examples of an upper-level manager that
manages a horizontal organization by producing boundary conditions that
promote cooperation are an early cell that includes an RNA manager that
establishes beneficial cooperative arrangements in a protein-based
autocatalytic set (the horizontal organization). It can do this by, for
example, intervening to catalyze the formation of a protein that is
beneficial to the autocatalytic set but that would not otherwise be
reproduced within the set; and a human manager comprising a chieftain,
ruler, government, or committee that promotes cooperation in a horizontal
organization of humans by, for example, punishing individuals who
undermine cooperation within the organization because they steal the
products of cooperative arrangements or because they fail to reciprocate
in exchange relations.

The evolution of these instances of upper-level management will be
considered in detail in Section 4 to assist in identifying how the
evolution of these forms of organization can be encouraged in artificial
life systems.

3.1.4 Lower-Level Management

A lower-level manager comprises evolvable entities that are at lower
levels in the nested hierarchies that constitute each of the entities of
the horizontal organization; that is, a lower-level manager is composed of
internal constituents of the entities of the horizontal organization, in
contrast to an upper-level manager whose entities are external to the
entities of the horizontal organization. Examples of these evolvable
lower-level internal constituents include the genome in relation to a cell
or a multicellular organism, and both the genome and clusters of
socialized behavior patterns (e.g., norms) in relation to a human. These
internal constituents influence the entities and organizations of which
they are a part through lower-level constraints (termed initiating
conditions by Salthe [29]). The constraints manifest in the entities of
the horizontal organization as intrinsic properties of the entities that
predispose them toward particular behaviors and other characteristics. It
is worth noting here that genetic arrangements can comprise both an
upper-level manager of molecular processes within a cell (the genetic
elements are external to the processes being managed) and a lower-level
manager of, for example, a society of organisms (in this case the genetic
arrangements are internal constituents of the organisms being managed).

Identifying examples of lower-level managers, and understanding how they
can control and constrain horizontal organization in ways that promote
cooperation, is not so clear cut and intuitively obvious as it is for
upper-level managers. It will be necessary to present a number of specific
examples. The nature of lower-level management is probably best
illustrated by the consideration of examples of human organization in
which a horizontal organization can be controlled and constrained by both
upper-level and lower-level management.

First, consider a level of organization in a hierarchical company or firm:
the behavior of individuals at this level can be controlled and managed
both (a) by the establishment by a higher level in the hierarchy of an
appropriate pattern of rewards and punishments (i.e., boundary conditions)
for individuals; and (b) by assuring that these individuals have
particular intrinsic properties, such as diligence, honesty, and
conscientiousness. These intrinsic properties arise from lower-level
constituents of the individual such as genes or socialized behavior
patterns.

Second, consider a human family: The behavior of children can be
constrained and managed by both (a) the establishment by parents (the
upper-level manager) of appropriate patterns of rewards and punishments;
and (b) by the inculcation in the children of particular behavior patterns
(e.g., norms) that will form intrinsic, lower-level constituents of the
children that constrain their behavior even in the absence of upper-level
constraints such as the possibility of rewards and punishment.

Finally, consider a human social group such as a tribe: The group could be
controlled to produce egalitarian behavior either (a) by a powerful ruler
who rewards egalitarian behavior and punishes alternative behavior; or (b)
by assuring that the group of individuals are constrained genetically to
interact in an egalitarian way or are inculcated with behavior patterns
that also constrain them to behave in this way.

The capacity of a lower-level manager to constrain and manage a horizontal
organization gives it the potential to, for example, establish cooperative
arrangements by constraining individuals to provide resources to
specialists who would not otherwise be sustainable in the horizontal
organization. And a lower-level manager has the same capacity as an
upper-level manager to use its control of the horizontal organization to
have the benefits of cooperation deployed to enhance the success of the
manager, for instance, by directing resources to the reproduction of the
genetic elements or behavioral patterns that collectively make up the
manager. As is the case for an upper-level manager, if a lower-level
manager is to realize fully its potential to promote cooperation, it must
be evolvable, and its evolutionary success must be dependent on the
success of the organization as a whole. If these conditions are met, the
lower-level manager will evolve constraints that will produce beneficial
cooperation in the horizontal organization.

Examples of organizations that are managed in this way by a lower-level
manager composed of evolvable internal constituents of the entities in the
horizontal organization are (a) a multicellular organism that is a
horizontal organization of cells, with each cell constrained by a
lower-level constituent, the genome. The genome is identical in all cells,
and collectively these genomes across all cells constitute the lower-level
manager that controls the organization of cells; (b) an insect society
that is a horizontal organization of organisms managed by a genome that is
reproduced across the society as lower-level constituents of the
organisms. Collectively the genomes constitute the lower-level manager;
and (c) egalitarian groups of human hunter-gatherers that are composed of
a horizontal organization of humans constrained by a cluster of socialized
behavior patterns (e.g., norms) and probably also by some common genetic
elements. The cluster of socialized behavior patterns is a lower-level
constituent reproduced in individuals across the organization, which
collectively constitute a lower-level manager. The cluster of behavior
patterns can control the group to advance the interests of the manager by,
for example, including behavior patterns that actuate individuals to
reproduce the cluster by inculcating it in others, including in their
progeny, and by actuating them to punish individuals (including by
expulsion) in whom the cluster has not been reproduced.

3.1.5 Management Constituted by a Horizontal Organization

It has been implicit in the discussion to this point that the manager
(whether upper or lower level) reproduces and responds to selection as a
coherent unit. If this is the case, and if the success of the manager
depends on the success of the organization as a whole, the management
instituted by the manager will be in the interests of the organization.
However, if the manager itself is composed of a number of entities, and is
therefore itself a horizontal organization, competition among the entities
will impede the ability of the manager to adapt optimally as a cooperative
whole, in the same way that competition limits any other horizontal
organization; and to the extent that the manager is unable to adapt
optimally as a whole, it will fail to manage optimally the original
horizontal organization in which it intervenes. Thus, for example, a
management entity may establish hierarchical controls that serve its
competitive interests at the expense of the interests of the manager as a
whole, and a management entity that can establish a beneficial
intervention in the initial horizontal organization may be outcompeted
within the managing horizontal organization.

This is particularly a problem for lower-level management: A lower-level
manager is necessarily composed of internal constituents within each of
the entities of the original horizontal organization--the potential for
competition among these numerous constituents is considerable. If the
competition is not constrained in any way, a lower-level manager will not
be constituted: The lower-level constituents will not reproduce or respond
to selection as a coherent unit, and there will not be any capacity to
modify outcomes across the horizontal organization at all. It will be an
unmanaged horizontal organization. The establishment of arrangements that
prevent differential success among its constituent entities have therefore
been critical to the evolution of organizations managed by a lower-level
manager.

This impediment to the evolution of the manager as a unit can be overcome
in the same way that it is for the original horizontal organization, that
is, by the emergence of a new level of management that intervenes in the
original managing horizontal organization to promote beneficial
cooperation. In this way, multi-level management may evolve. However, if
the new level of management is itself a horizontal organization, this is
not a final solution: The impediment is simply exported to the new level.

Of course, the impediment will not arise when the manager is composed of a single entity, for instance, by a single RNA structure in the case of the
molecular example of upper-level hierarchical control considered above, or
by a chieftain in the example of human organization managed by an
upper-level manager. This suggests that the impediment can also be
overcome in relation to multi-level management by heading the management
with a single entity that successfully controls lower levels of
management. Many modern human hierarchical organizations are managed in this way.

However, arrangements of this sort can overcome the impediment only when the manager is composed of or headed by a single entity. The difficulty will resurface whenever the potential for competition among a number of entities arises, for instance, when a chieftain is to be replaced, or when the single RNA structure reproduces.

This problem is particularly significant when the manager is composed of
an entity such as an RNA structure that discovers adaptations through a
process that involves differential reproductive success between entities:
In these circumstances, reproduction of the entity may result in
competition between its progeny. This is less a problem in the case of a
human ruler who tests alternative adaptations against internal models and
against internal proxies for differential reproductive success, rather
than by actual differential reproductive success among rulers.

3.1.6 Recursive Management of Competition

The difficulties that arise because of competition between entities that
constitute the manager can, however, be overcome recursively without the
emergence of new levels of management. These arrangements are recursive in the sense that they are established by adaptations of entities within the
managing horizontal organization itself. Ideally, the arrangements will
operate to suppress only competition that does not result in the success
of heritable variation that maximizes the success of the organization
("heritable variation" is used broadly in this article to refer to all
variation, genetic or otherwise, that can provide a basis for evolutionary
change. It includes, for example, variation in ideas and beliefs that are
transmittable between human individuals). Examples of organizations that
can internally select heritable variation on the basis of its benefit to
the organization (e.g., by testing the effects of alternatives on internal
proxies for organizational success) are humans, and modern hierarchical
organizations of humans. The advantage of internal testing is that it
enables the organization to discover adaptations during its life, rather
than having to rely on differential reproductive success between
organizations to test variation [31].

However, all competition involving heritable variation must be suppressed
within organizations that do not have internal arrangements that can
differentiate between variation that is likely to benefit the organization
and variation that is not. These organizations must rely on a
between-group selection process involving the differential reproductive
success of organizations to select variation that maximizes the fitness of
organizations. If selection operating at the level of the group is to be
fully effective, competition between entities within the organization must
be suppressed, thereby concentrating competition and natural selection at
the between-organization level [36]. This ensures that there is no heritable differential success within the organisation, and that the only way in which entities can achieve heritable relative success is through their contribution to the differential success of organisations.

At first it may seem that a manager which is a horizontal organisation could have no greater capacity to recursively overcome internal competition than could the original horizontal organisation. Alternatively, it may be suggested that if the manager is able to recursively suppress competition, why couldn't the original horizontal organisation also do so, rendering the manager redundant and unnecessary?

The reason why the original horizontal organization and the manager have
fundamentally different capacities in this respect is that the manager
controls a horizontal organization, and it can use this control to
construct structures and processes that can act across the organization to
suppress competition. Only a manager has the capacity to control and
constrain the organization on a sufficient scale to suppress competition
across the organization.

However, this raises a further issue: How can adaptations that suppress
competition become established within the manager so that they can achieve
the necessary hierarchical control across the organization? how will they
overcome competition from alternatives within the manager that don't
invest resources in the suppression of competition? This further instance
of the second-order problem can be overcome in the following way:
Suppressors will not be outcompeted if the competition they suppress
within the organization also includes the competition they would otherwise
encounter from alternatives. That is, successful suppressors must also
suppress competition from alternatives who do not suppress.

A series of examples will illustrate how a manager is able to suppress
competition by using its capacity to control a horizontal organization and
how the controls can escape the second-order problem and avoid being
outcompeted within the manager. Consider a horizontal organization of
organisms that is managed by a lower-level manager that is composed of
genetic arrangements: Genetic elements that arise in the manager may
actuate individuals to direct their cooperation preferentially toward
closer relatives who are more likely to include and to reproduce the
manager, and who are also more likely to include and reproduce these
particular genetic elements (i.e, the kin selection processes of Hamilton
[15]); genetic elements that arise in such a manager also may actuate
individuals to punish other individuals who do not exhibit the actions of
individuals controlled by the manager as well as individuals who do not
act as if they include the particular genetic elements that actuate
punishers; that is, nonpunishers are also punished (this example is
explored in detail by Boyd and Richerson [6], but without the hierarchical
perspective developed here); and finally, genetic elements may arise that
actuate individuals to direct their cooperation toward supporting the
reproduction of only a single individual within the horizontal
organization, thereby preventing the reproduction of individuals that
might not include the manager and that also might not include these
particular genetic elements (e.g., some eusocial insect colonies).

Arrangements that suppress competition at various levels of organization
and that have been studied in some detail are surveyed by Jablonka [19].
Additional examples to those already considered above, described from the
hierarchical perspective, include the organization of genes on single
chromosomes, which reduces competition among genes within the upper-level
manager that manages molecular processes within cells [9]; meiosis, which
also limits competition among genes and chromosomes within the upper-level
manager of cells [13, 24]; and sequestration of the germ line together
with reproduction through a single cell, which reduces competition between
the genomes that constitute the lower-level manager controlling
organizations of cells [8].

3.1.7 The Significance of Vertical Organization

However, the successful suppression of competition within the organization
and its concentration at the between-group level is not sufficient in
itself to ensure that group selection will be able to establish the
extensive level of cooperative differentiation that characterizes the key
evolutionary transitions. For this to be achieved, the variation that
arises between organizations must include the production of organizations
within which the necessary division of labor is able to be
sustained--selection will be unable to select these forms of organization
if the variation presented for selection does not include them. The
vertical arrangements discussed here are therefore essential to the key
evolutionary transitions not only because they allow the comprehensive
management of competition, but also because, as we have seen, they can
control horizontal organization to produce a wide range of alternative
organizations that would not otherwise be available for selection. For
instance, a manager can underpin comprehensive differentiation by
intervening to redirect resources to support specialists that could not
otherwise reproduce or even persist in a horizontal organization.

The significance of vertical organization is somewhat obscured in the
instances of the evolution of cooperative organization commonly studied by
biologists. This is because these instances involve organizations of
entities that already include evolvable lower-level constituents (i.e.,
genetic arrangements), and a manager can be readily constituted merely
through the suppression of competition between these preexisting
lower-level constituents across the organization. The significant role of
vertical organization is more clearly seen by studying evolutionary
sequences in which the evolvable lower-or higher-level entities are
initially absent or poorly developed, as in the sequences discussed in
Section 4.

In summary, the process of vertical self-organization described here is
essential for the evolutionary transitions in which higher-level entities
have been formed through the evolution of highly differentiated
cooperative organizations of lower-level entities. The familiar nested
hierarchies of living processes arise through the repeated formation of
organizations of entities that are managed by hierarchical arrangements
that ensure the entities adapt and act to serve the interests of the
organization as a whole.

3.2 The Governance of Matter

3.2.1 The Hierarchical Perspective

To what extent can the concepts and processes that underpin this account
of horizontal and vertical self-organization in living processes also
provide an understanding of the emergence of living processes from
inanimate matter?

Horizontal organization among nonliving entities is widespread: It is
evident that interactions among entities at the same level of organization
can give rise to organizations of entities. Such organizations form and
persist to the extent that the relationships between entities that
constitute the organizations are reproduced to some extent through time,
because, for instance, the relationships represent stable or dynamic
equilibria. However, in contrast to horizontal organization among living
processes, mere physical persistence is a sufficient condition for
nonliving processes: In the case of living processes, the adaptations that
underpin relationships must be not only physically realizable, but also
competitive, for instance, by maximizing the fitness of participants. As
for vertical organization in living processes, hierarchical relationships
among nonliving entities will be constituted where one set of processes or
entities is able to influence another set without in turn being
influenced.

It is evident from consideration of the material world that the
asymmetrical functional relationships that characterize hierarchical
separation can arise when there is a difference in scale between
interacting entities or processes: For example, where entities differ
sufficiently in scale, a larger-scale entity may influence the dynamical
behavior of a set of smaller-scale entities without itself being
influenced by the interactions; because of its larger scale, the
hierarchical entity does not participate in the lower-level processes
dynamically: It stands outside and acts across the dynamic of
smaller-scale entities.

The difference in scale is often reflected in the duration of time of
phenomena (longer for higher-scale entities) or of the length of periods
between events (longer for events coming out of processes of larger
scale). When an asymmetrical functional relationship is constituted in
this way, processes that constitute boundary conditions would operate on a
much slower time scale than the dynamical interactions of the level below;
from the perspective of an entity participating in the lower-level
dynamic, boundary conditions of this type are typically seen as relatively
unchanging features that are not influenced by the individual entity and
the interactions in which it is involved [29].

Because the nonliving world is separated into components and processes
that differ widely in scale and that often are also organized as nested
hierarchies (e.g., quarks, protons, atoms, molecules, oceans, planets,
galaxies, etc.), hierarchical interactions are pervasive. As a
consequence, the provision of an adequate account of processes at any
particular focal level of nonliving processes will usually require the
inclusion of relevant processes at both lower and higher levels of
hierarchy that influence and constrain the focal level but do not
participate in the interactions of the focal level dynamic [29]. For
example, chemical systems with identical initiating conditions and
identical focal-level processes can unfold into entirely different systems
under different boundary conditions (e.g., under differences in
temperature, pressure, and the location and form of any structures of
greater scale and stability that interact with the system, such as any
structure that contains the system); and processes at the quantum level
may unfold as waves or particles depending on the boundary conditions they
encounter. The hierarchical perspective is essential to provide an
adequate account of the evolution of physical systems in the material
world where boundary conditions vary in space and time.

3.2.2 Management that Produces New Organizations

This capacity for boundary conditions to influence the nature of the
organizations of entities that will form and persist in an interacting
dynamic of entities is critical to our discussion here: Just as living
entities that are in a hierarchical relationship with a dynamic of
interacting living entities may produce forms of organization that would
not have arisen otherwise, nonliving entities may similarly constrain a
material dynamic.

Thus at a given focal level of nonliving organization, an interacting
dynamic of entities may give rise to horizontal organizations of entities
that themselves form entities of larger scale. These larger-scale entities
may interact in hierarchical fashion with the parent dynamic of
smaller-scale entities, providing new boundary conditions that constrain
the dynamic to produce new forms of organizations of entities that would
not have arisen otherwise in the dynamic.

For example, at the molecular level, an interacting dynamic of
smaller-scale atoms and molecules may give rise to molecules of larger
scale (formed as horizontal organizations of smaller-scale atoms and
molecules) that in turn provide new boundary conditions for the dynamic of
smaller-scale entities; the larger scale enables the molecules to stand
outside and act across the dynamical interactions of the smaller-scale
entities to produce outcomes that would be improbable in the unconstrained
dynamic. Thus a larger-scale molecule could cause the formation of
molecules that are unlikely to arise in the unconstrained dynamic because
their formation requires, for example, a coming together and particular
positioning of a number of smaller-scale molecules that is highly unlikely
to occur spontaneously in the interactions of the dynamic: The capacity of
the larger-scale molecule to stand outside and act across the dynamic in
both space and time enables it to collect together over time the outcomes
of a number of different events that are highly unlikely to occur
simultaneously but that are likely to occur sequentially over time, and to
put together particular spatial arrangements and positionings of
smaller-scale entities that would otherwise be improbable. This process
is, of course, from another perspective, chemical catalysis. The new
organizations formed in this way may themselves constitute higher-scale
entities that in turn provide new boundary conditions for the dynamic,
resulting in the formation of further organizations that would not arise
spontaneously in the unconstrained dynamic, and so on.

In this way, a new space of possible arrangements of matter can be opened
up for exploration. Organization is no longer limited to what can come
together fortuitously through the unconstrained interactions of entities
in a horizontal dynamic; vertical organization opens the way to the
formation of more complex organizations by a process of construction.

However, the search of this new space will generally be undirected: There
will not necessarily be any pattern to the organizations produced. On the
basis outlined, vertical organization can manage matter to form
organizations that would not otherwise arise, but this management would
not necessarily have any particular objective or direction; it does not
include any overriding mechanism that, for example, would ensure that only
management that achieved particular outcomes would persist.

3.2.3 Self-Replicating Management

Such an overriding mechanism may arise, however, if and when these
processes of horizontal and vertical self-organization produce
self-replicating management, for instance, through the production of
larger-scale entities that manage the parental dynamic to produce copies
of themselves. These entities may be self-replicating as individuals, or
as a collection of entities (e.g., the autocatalytic sets of Eigen and
Schuster [11]; and Farmer et al. [14]).

A key feature of self-replicating management is that away from equilibrium
it produces a larger-scale organization (the population) whose growth,
until it is otherwise limited, is subject to positive feedback--every
increase in population size in turn increases the capacity of the
population to grow. When the population encounters resource limits, the
result is competition between members of the population, which drives the
familiar process of natural selection. Natural selection in turn gives the
management of matter direction and pattern--only the most competitive
management will persist. In contrast, the population growth of managing
entities that are produced by other types of managing entities is not
driven in this way by positive feedback: An increase in the population of
one type of manager does not in turn have any effect on the extent to
which additional managers of that type are produced.

Although populations of self-replicating managers may achieve
substantially higher scales than individual managing entities, initially
this does not mean that management will be organized or that matter will
be managed on these larger scales--as we have seen, without vertical
organization there is limited capacity to evolve large-scale cooperative
adaptations that coordinate the activities of individuals across the
greater scales: Selection founded on competition precludes the less fit,
irrespective of whether they are participating in a beneficial cooperative
arrangement that is more competitive as a whole.

3.2.4 Management of Management

As we have also seen, this limitation can be overcome through the
formation of horizontal organizations of self-replicating managers that
are in turn managed by arrangements in hierarchical relationship to the
managers. Repetition of this process will produce living processes
organized as nested hierarchies and will progressively extend the
management of living processes across space and time. In this way, as
managed living processes increase in scale, they gain the capacity to
manage the material world at greater and greater scales to reproduce
themselves. Whatever level of nonliving organization living processes
originate in, life will tend to become organized at, and manage matter at,
increasingly larger scales.

When applied at the molecular level of organization, this account
parallels in many respects the standard theories of the origin of life
through the evolution of self-replicating molecular systems (e.g., see
[21]). However, these accounts have not identified the essential
hierarchical relationship between the self-replicating entities and the
dynamic they manage, and they have not recognized that this vertical
organization arises for similar reasons to the emergence of vertical
organization in living processes: The capacity to influence without in
turn being influenced enables a manager to govern the level below to
produce advantageous forms of organization that would not arise
spontaneously in that level.

4 The Emergence of Evolutionary Transitions in Artificial Life

4.1 Cooperation in Artificial Life

It will be necessary for artificial life to include the processes of
horizontal and vertical self-organization if it is to synthesize living
processes that are organized as nested hierarchies, and if it is to
synthesize the evolutionary transitions that give rise to them. There are
numerous examples of alife systems that have successfully included
instances of the processes of horizontal self-organization dealt with in
Section 2, such as mutualism and reciprocity (e.g., [2, 17, 18, 22, 27,
33]). Not only have these alife systems evolved cooperation when
cooperative adaptations are initially incorporated in the system and set
in competition with non-cooperators, but some systems have also discovered cooperative adaptations not explicit in the initial system (e.g. Ray [27]; and Lindgren and Nordahl [22])

However, if artificial life is to synthesise the evolutionary transitions that give rise to new levels of organisation, the processes of horizontal self-organisation need to be complemented by vertical self-organisation. To date, artificial life has not been synthesised explicitly to encourage the emergence of vertical organisation, and the synthesis of entities that undergo an evolutionary transition to form higher level entities is seen as an important challenge for the future (e.g., see [28, 32]). Furthermore, when alife practitioners have recognized the importance of the synthesis of evolutionary transitions they have tended to focus on the transition from the cellular to the multicellular level where the vertical system is initially constituted by lower-level management, rather than the transition to cells and to modern human social systems where the vertical system constituted by upper-level management is more significant. To point the way to how this synthesis can be initiated, it is necessary to develop and to operationalize the concept of hierarchical relationship, identifying the critical features of the relationship so that they can be incorporated in alife systems.

I will begin by briefly examining a range of illustrative examples of the
evolution of management at various levels of organization to develop a
more concrete understanding of the evolution of essential elements of the
hierarchical relationship in living processes. I will conclude by
abstracting from the examples the relationships and processes that are
common to the various instances of the evolution of vertical
self-organization and that need to be included or synthesized in an alife
system if vertical organization is to emerge.

4.2 Evolution of Modern Human Organization

I will commence with an illustrative example of the evolution of human
organization. However, it should be noted that this example concerns the
evolution of modern, hierarchical human organizations, which have arisen
during the 12,000 years to the present, and which have largely replaced
the more egalitarian hunter-gatherer societies that preceded them (the
evolution of the hunter-gatherer phase is dealt with by, for example,
Knauft [20], Boehm [5], and Wilson and Sober [37]). It also should be
noted that, as aptly pointed out by Erdal and Whiten [12], these modern
human hierarchical organizations "are not merely reborn ape hierarchies,
but uniquely human in both their behavioural detail and their cultural
recognition" (p. 178). It is this form of organization that is responsible
for the extraordinary level of cooperative differentiation in modern human
societies, which matches that found in the other major evolutionary
transitions that are the focus of this article.

Consider a small, stand-alone horizontal organization--a tribe or an
agricultural community. An individual or alliance of individuals that has
the ability to coerce other members of the organization (due to physical
strength or superior weapons) may be able to extract a disproportionate
share of the resources and services produced in the organization (e.g.,
food and reproductive opportunities). Critically, these hierarchical
individuals do not have to participate in the competitive interactions or
mutually advantageous exchange relations of the horizontal organization to
obtain these resources. This manager may emerge from within the human
group or may itself be a free-living band that moves from group to group
plundering resources.

A distinction can be drawn between interventions made by the manager in
the horizontal organization that simply appropriate resources, and those
that actually cause the production of benefits that provide direct and
immediate benefits to the manager (e.g., the coercion of individuals to
make weapons or grow food for the manager). The latter class of
intervention is a short step away from interventions that increase the
resources available to the manager by promoting the efficiency of the
organization as a whole. For example, the manager may punish cheats who
would otherwise undermine beneficial cooperation in the horizontal
organization by not reciprocating in exchanges of goods; and it may
provide resources to individuals to promote the performance of actions
that are beneficial to the organization but that would otherwise not be
sustained because they would not produce sufficient direct benefit for the
individual (e.g., group defense).

These distinctions suggest an evolutionary sequence that begins with a
manager whose relationship with the group is largely limited to
appropriating benefits, moves through a phase in which the manager also
intervenes in the group to cause the production of benefits that directly
benefit the manager (but without improving the overall productive capacity
of the group), and then moves to a relationship where the manager manages
the group to increase its overall capacity to produce benefits, some of
which are appropriated by the manager for its maintenance (e.g.,
taxation). The end result of this evolutionary sequence is a manager that
has the capacity to improve substantially the competitive capacity of the
organization at the intergroup level, whose success and continued
existence is dependent on the organization of which it is an obligate
part, and whose interests lie to a significant extent in using this
capacity to promote the success of the organization as a whole,
particularly when the group is in competition with other groups.

To the extent that the interests of the organization and the interests of
a ruler who heads the manager coincide, the ruler's adaptation of its
management of the organization will tend to maximize the success of both
the organization and the ruler--through the adaptation of the ruler, the
organization is able to adapt continually to changing internal and
external events. In these circumstances, heritable adaptation at the level
of the group does not have to rely on a between-group selection process,
and all heritable variation is not suppressed within the manager or within
the horizontal organization.

This contrasts with forms of organization in which the human group is
primarily managed by a lower-level manager composed of genetic elements
and/or of clusters of socialized behavior patterns--in these organizations
the potential for competition within the manager is considerable, and it
must be tightly controlled. Consequently, adaptation of the manager has to
rely on differential success between groups, except in the limited
circumstances when some variation within the lower-level manager can be
successfully managed. (Rappaport [26] and Boehm [5] identify some
arrangements within small human groups that could maintain the necessary
control over competition while allowing some flexibility during the life
of a group.) This difference in adaptive capacity is likely to have been a
significant factor in the competitive superiority of human organization
managed by an upper-level manager [29].

In this illustrative example, the hierarchical relationship is founded
upon the capacity of the manager to use coercion to influence and
constrain the horizontal organization without in turn being influenced by
it. This hierarchical relationship is advantageous to the manager because
it gives the manager the capacity to appropriate resources and services
unilaterally that would otherwise be available to the horizontal
organization. Significantly, this capacity to appropriate benefits also
creates the potential for the manager to benefit from advantageous
cooperative arrangements, which it establishes by appropriate
interventions: The capacity enables the manager to harvest benefits
flowing from the cooperative arrangements. However, this potential will
not be realized until interventions are discovered that enable the manager
to promote cooperation, and unless the manager's association with the
horizontal organization is sufficiently prolonged to enable it to harvest
the benefits produced. The capacity of the manager to influence the
horizontal organization unilaterally forms a basis for the development of
these interventions that may, for example, entail selective punishment and
the differential redistribution of appropriated resources.

As is necessary if this proposed evolutionary sequence is to be considered
plausible and is to avoid the second-order problem, the individuals that
participate in the sequence are not required to act other than in their
direct individual interests at any phase of the sequence. In particular,
the sequence envisages that the manager intervenes in the horizontal
organization to promote cooperation only when the interventions produce
net benefits to the manager. The second-order problem is avoided through
the capacity of the manager to appropriate the wider benefits that flow to
the organization from the cooperation that its interventions
promote--unlike members of the horizontal organization that are limited by
the second-order problem, the manager is not dependent on having to obtain
net benefits from the interventions themselves.

Consideration of other forms of hierarchical human organization indicates
that the required capacity of the manager to influence the horizontal
organization can be initially established other than by coercion: Examples
can be readily found where it is established by informed consent arising
from common interests within the relevant horizontal organization (e.g.,
voluntary associations), or by manipulated consent of the horizontal
organization (e.g., religious cults), or by combinations of informed
consent, manipulation, and force. However, irrespective of its initial
basis, if the manager is to be capable of optimally managing the
horizontal organization in the interests of the organization as a whole,
the manager must have sufficient power, scale, and scope to act across the
organization to influence the interests of its members in the domain
covered by the organization: To the extent that it is unable to deter
cheating across the organization (e.g., by fines, imprisonment, or
execution) or to promote beneficial cooperation across the organization
(e.g., by paying employees, providing awards, or conferring status), the
interests of the organization will not be maximized. Beneficial management
will also be impaired to the extent that the access of the manager to
benefits is not dependent on the manager managing in the interests of the
organization as a whole: For example, policing will tend to fail to the
extent that police can obtain benefits (e.g., bribes) from the horizontal
organization they manage; an executive may get too close to his or her
staff (i.e., may value the esteem of staff more than some incentives
offered by the organization); and a bureaucrat may be able to avoid
accountability to the organization as a whole.

It should also be noted that in modern huma organization, lower-level
management as well as upper-level management is often operative in
creating the conditions for the emergence of cooperative organization: For
example, it has long been recognized that socialized behavior patterns
(lower-level management) that produce trust and honesty in individuals
participating in economic exchange relations can reduce the incidence of
cheating and lessen the need for its control by upper-level management
(e.g., see [35]).

4.3 The Evolution of RNA Management of Molecular Processes

As a further illustrative example, consider a horizontal organization at
the molecular level that comprises an autocatalytic set of proteins.
Single self-replicating RNA molecules or small groups of RNA molecules
that are autocatalytic in combination may be able to manage (appropriate)
some of the metabolic constituents of the protein-based autocatalytic set
to maintain themselves and reproduce. The RNA may also discover the
capacity to catalyze particular processes within the autocatalytic set
that directly assist the maintenance or reproduction of the RNA. In this
way, an RNA upper-level manager of sufficient size and stability would
stand outside the competitive interactions and exchange relations of the
horizontal organization, managing them for its own benefit. The RNA could
move from organization to organization, draining them of resources and
services.

As for the example of human organization, the capacity of the RNA
upper-level manager to manage the horizontal organization also creates the
potential for it to intervene in a way that promotes beneficial
cooperative arrangements and to harvest the benefit of any cooperative
arrangements that are promoted in this way. For example, the RNA might
catalyze a protein that provides benefits to other members of the
autocatalytic set, but which itself receives no benefits in return, and
which would be an altruist in the absence of support from the RNA; and the
RNA might catalyze a process that inhibits the reproduction within the
autocatalytic set of a cheat or freeloader that takes benefits from the
set without providing any benefits in return (the capacity for cheating
and competition to prevent optimal cooperation in autocatalytic
organization in the same way as in other horizontal organization is dealt
with in detail by Maynard Smith [23], Bresch et al. [7], and Bagley and
Farmer [4]). Once this potential is realized, the RNA manager would find
advantage not only in harvesting the horizontal organization, but also in
managing and intervening in the horizontal organization to make available
greater benefits for harvesting. The discovery of such arrangements is
more likely when the manager and horizontal organization live in close
association: The manager is more likely to be able to capture the benefits
of any cooperation promoted by it the more prolonged its association with
the horizontal organization.

Again, the end result of this evolutionary sequence is a manager that has
the capacity to enhance the competitive ability of the organization at the
intergroup level, whose success and continued existence is dependent on
the organization of which it is an obligate part, and whose evolutionary
interests lie to a significant extent in using this capacity to promote
the success of the organization as a whole. Once again we have an
evolutionary sequence in which selection operating on the manager is
progressively brought into alignment with selection operating at the level
of the organization. However, in contrast to modern human organization,
RNA does not have any internal capacity to adapt heritably, variation must
be suppressed within the organization, and heritable adaptation must rely
on between-group selection. Dyson [10] has proposed a similar evolutionary
sequence for early cells, but without the general and unifying
hierarchical perspective developed here.

5 Conclusion

If artificial life is to meet its central objective of synthesizing key
biological phenomena, it must be able to synthesize entities that are
organized as nested hierarchies and synthesize evolutionary transitions in
which differentiated higher-level entities are formed from cooperative
organizations of lower-level entities. This will necessitate the inclusion
in alife systems of the processes of vertical self-organization to
complement the processes of horizontal self-organization that have already
been synthesized in some systems.

The particular way in which vertical organization is actually constituted
in any specific case differs both within and between the various levels of
organization, depending in each instance on contingencies such as what it
is that determines the success of entities, and how entities can influence
each other's success.

However, at a higher level of abstraction it is possible to identify the
key structures and relationships that are common to the various instances
of vertical organization that have been discussed above, and which would
need to be incorporated or synthesized in an alife system for vertical
organization to emerge. These are

1. Horizontal organizations of adaptive agents (i.e., evolvable entities),
with each organization constituted by relationships that arise and persist
due to the capacity of the adaptive agents to engage in cooperative
interactions within the organization in which the success of each agent
participating in the interaction is increased (e.g., cooperation that is
mutually or reciprocally beneficial);

2. Within each horizontal organization, the possibility of adaptive agents
arising that have the capacity either to

(a) engage in interactions with other adaptive agents within the
organization in which the success of the other agents is increased, but
their own success is decreased (e.g., altruists); or

(b) engage in cooperative interactions with other adaptive agents within
the organization without making any contribution to the interactions
(e.g., cheats that do not reciprocate or do not contribute to mutualistic
cooperation);

3. Managing adaptive agents associated with each horizontal organization,
with each agent having the capacity to

(a) engage in interactions with the agents of the horizontal organization
in which the success of the managing agents is increased and the success
of the agents of the horizontal organization is decreased (e.g., the
unilateral appropriation of resources by the manager from across the
horizontal organization); and

(b) engage in interactions differentially with agents of the horizontal
organization in which the success of the agents of the horizontal
organization is either increased or decreased, and the success of the
managing agents is decreased (e.g., the capacity of the manager to
intervene across the organization to support altruists or inhibit cheats);
and

(c) exhibit adaptations that suppress differential success among the
managing adaptive agents within the organization (e.g., recursive
suppression of competition within the manager).

Acknowledgments

I gratefully acknowledge the benefit of useful comments from Jeremy Evans, David Richards and Wilson Kenell, and from a reviewer, David Sloan Wilson.

REFERENCES

1.Axelrod, R. (1986). The evolution of norms. Amer. Pol. Sci. Rev. 80: 1095-1111.

2.Axelrod, R. (1987). The evolution of strategies in the iterated Prisoner’s Dilemma. In L. D. Davis (Ed.), Genetic algorithms and simulated annealing. Research Notes in Artificial Intelligence. Los Altos, CA: Morgan Kaufmann.