My agents love to conform: Norms and emotion - polsoz.fu

Document technical information

Format pdf
Size 270.0 kB
First found May 22, 2018

Document content analysis

Category Also themed
not defined
no text concepts found


Ernst Fehr
Ernst Fehr

wikipedia, lookup




Comput Math Organiz Theor (2006) 12:81–100
DOI 10.1007/s10588-006-9538-6
My agents love to conform: Norms and emotion
in the micro-macro link
Christian von Scheve · Daniel Moldt · Julia Fix ·
Rolf von Luede
Springer Science + Business Media, LLC 2006
Abstract This contribution investigates the function of emotion in relation to norms,
both in natural and artificial societies. We illustrate that unintentional behavior can be
normative and socially functional at the same time, thereby highlighting the role of
emotion. Conceiving of norms as mental objects we then examine the role of emotion in
maintaining and enforcing such propositional attitudes. The findings are subsequently
related to social structural dynamics and questions concerning micro-macro linkage,
in natural societies as well as in artificial systems. Finally, we outline the possibilities
of an application to the socionic multi-agent architecture SONAR.
Keywords Emotions . Social norms . Micro-macro link . Multi-agent systems .
Socionics . Petri Net modeling
For some time now apprehensions from the general public as well as from the scientific community have been issued concerning the controllability of artificial intelligence systems, in particular distributed systems consisting of a multitude of intelligent autonomous agents. Agent systems are feared to run out of control in such
a way that autonomously generated goals a system pursues might contradict some
of the crucial, although implicit high-level goals of the designer or the user of a
system, in consequence leading to inefficient, undesirable or even hazardous system
C. von Scheve () . R. von. Luede
University of Hamburg, School of Business, Economics and Social Sciences, Institute of Sociology,
Allende-Platz 1, D-20146 Hamburg, Germany
e-mail: [email protected]
D. Moldt . J. Fix
Faculty of Informatics, Theoretical Foundations of Computer Science, Vogt-Koelln-Str. 30, D-22527
Hamburg, Germany
C. von Scheve, D, Moldt et al.
behavior. To make matters worse, the dilemma arising out of these conflicting goals
affects some of the core strengths of artificial agent systems: their autonomy, flexibility, and discretion. Thus, approaches at resolving the issue of conflicting goals
have to ensure the autonomy of the system in question and the integrity of human
high-level goals at the same time. One feasible solution to this problem is the implementation of a system of shared norms which is not coerced by the designer, but instead
emerges from mutual interactions of agents (and users) (cf. Shoham and Tennenholtz,
1997). To support such an approach, it might be beneficial—if not mandatory—to
acquire profound knowledge of and to adapt to the computational context the mechanisms of norm emergence, maintenance, and compliance in human social systems. In
this article, we argue that emotion constitutes such a mechanism and should therefore
be taken into account in designing agents and multi-agent systems (MAS) in particular.
The concept of agents is inspired by and to a large extent relies on classical
paradigms of cognitive psychology and hence on corresponding conceptualizations of
intelligent behavior, which is still fundamentally based on cognitions and representations. Belief-Desire-Intention (BDI) architectures can well be considered an epitome
of this perspective on intelligent information processing (Wooldridge, 2000). Notwithstanding this, the interrelation of emotion and cognition and the role of emotion in
overall intelligent behavior have been long debated in the behavioral sciences, having
likewise promoted the idea that artificial intelligence (AI) systems might be improved
by taking into account mechanisms which are functionally equivalent to emotions
in their biological counterparts (Simon, 1967; Sloman and Croucher, 1981). At least
since Marvin Minsky’s programmatic and frequently evoked statement that “the question is not whether intelligent machines can have emotions, but whether machines can
be intelligent without any emotions” (Minsky, 1986: 163), efforts have been increased
within the AI community to develop emotional agents, i.e. entities endowed with
mechanisms functionally equivalent to emotion in human and non-human animals (cf.
Canamero, 1998; Hatano, Okada, and Tanabe, 2000; Trappl, Petta, and Payr, 2003).
Yet, the functions of emotion in large-scale distributed systems have not been investigated extensively, despite some initial explorative efforts in multi-agent systems
research (Aube and Senteni, 1996; Elliot, 1993; Fix, 2004; Gmytrasiewicz and Lisetti,
2000; von Scheve and Moldt, 2004). This is probably due to fact that research on emotions in natural large-scale social systems, i.e. human societies, groups, organizations,
and communities is still largely underdeveloped, although first investigations indicate
that emotion indeed plays a key role in interlinking individual action and global system behavior (Barbalet, 1998; von Scheve and von Luede, 2005). A crucial part of
these initial efforts is the analysis of the interplay of social norms and emotion. Proper
theorizing and empirical evidence both suggest that emotion is essential in sustaining
social norms and in enforcing sanctions in cases of non-compliance (Elster, 1996;
Fehr and Fischbacher, 2004). On the other hand, social norms are frequent causes of
emotions and to a certain extent determine coping and regulation processes (Heise and
Calhan, 1995; Hochschild, 1979).
Whereas the application of norms as a determinant of social action and a principle
of coordination is becoming a major area of inquiry in multi-agent systems research
and may draw on an extensive literature in the social sciences (Castelfranchi et al.,
1999; Verhagen, 2000; Tuomela, 1995), it is not surprising that the role of emotion
in normative multi-agent systems is largely not accounted for. However, pioneering
My agents love to conform
research in the computational study of social norms and emotion has been conducted by
Alexander Staller and Paolo Petta (2001) and by Ana Bazzan and colleagues (2002).
In this article, we aim at extending this line of research by following a socionic
approach, considering agent systems as a model for social science research—in fact
a modeling tool—and a software engineering paradigm alike (Jennings, 2000; von
Luede, Moldt, and Valk, 2003; Macy and Willer, 2002; Malsch, 2001; Sawyer, 2004).
In view of the fact that distributed systems are of increasing importance in many areas of
application, for example in electronic marketplaces, automated negotiations, planning
and scheduling systems, business process and workflow management, coordination of
large-scale open systems, and social simulations it seems thoroughly reasonable to us
to further promote the investigation of the functions of emotion in large-scale social
systems, natural as well as artificial.
We proceed as follows: First, we briefly illustrate some of the social functions of
emotion in a broader perspective regarding an organism’s internal functioning and
social interactions. We then argue for a model of social control that is fundamentally based on the relation between social norms and emotion, thereby referring to
two specific functions of emotion and the nature of social norms. In this model, the
emotion-based commitment to social norms in particular ensures actors’ compliance
with norms. Finally, we give a sketch of how these findings might be modeled in the
SONAR/MULAN multi-agent architectures.
Social functions of emotion
To achieve a better understanding of the relation between social norms and emotion,
we will outline some of the social functions emotion serves in individual behavior,
social interaction, and social aggregation, which may in turn be distinguished from
intraindividual (Levenson, 1999), phylogenetic (Cosmides and Tooby, 2000), and ontogenetic functions (Abe and Izard, 1999) of emotion. In fact, emotions are supposed
to be functional by definition, being regarded as “functional, organized responses to
environmental demands that prepare and motivate the person to cope with the adaptational implications of those demands” (Smith and Pope, 1992: 36). According to this
definition, a central function of emotion is the adaptational and beneficial regulation of
behavior in relation to environmental conditions, which clearly can be both, physical
and social in nature (Keltner and Gross, 1999: 468). Conceptually as well as empirically, social functions of emotion have been located on different levels, for example
biological, psychological, and social (Averill, 1992); individual, dyadic, group, and
cultural (Keltner and Haidt, 1999); organism, personality, social structure, and culture
(Gerhards, 1988) as well as micro, meso, and macro (von Scheve and Moldt, 2004).
On the level of an individual agent, emotion performs above all two functions:
On the one hand, emotion informs an agent about events in the social environment
that require immediate, reactive, and adaptive behavior, also known as the “affect
as information” paradigm (Clore, Schwarz, and Conway, 1994; Schwarz and Clore,
1988). For example, annoyance informs about the felt fairness of an action, love informs about degrees of affection and commitment, shame and embarrassment inform
about the conformity of an action (Keltner and Haidt, 1999; see below). On the other
hand, emotions prepare the organism to react adequately upon situational demands, for
C. von Scheve, D, Moldt et al.
example by triggering physiological arousal and by structuring the cognitive system
into adequate operational modes (Cacioppo et al., 2000; Clore, 1994; Oatley and Jenkins, 1996: 252). The verbal and nonverbal expression of emotion in social interaction
is a further function of emotion particularly tied to social norms (Keltner and Haidt,
1999): First of all, emotion expressions allow the mutual attribution of most interactional contingencies, including emotional state, interpretations of the situation, and
intentions. Secondly, emotion expressions may (unconsciously) evoke complementary
or reciprocal reactions in context-bound observing actors and therewith contribute to
improved mutual interpretations of a situation, which in turn is a prerequisite for
cooperation and the coordination of social action (Frank, 2001). Thirdly, emotion expressions may promote or constrain specific courses of an interaction and act as either
motivating or sanctioning signals. Large-scale perspectives on the social functions of
emotion highlight their role in identifying social groups and group members (Heise
and O’Brien, 1993), ascribing status and power resources (Kemper, 1978), the emergence and maintenance of solidarity and cohesion (Lawler, Thye, and Yoon 2000),
and in the internalization and sustenance of social norms, power structures, moral
and cultural ideas, ideologies, and the like (Barbalet, 1998; Elster, 1999; Hochschild,
As outlined above, social functions of emotion are identified on rather different
levels, all of which supposedly “refer to the history of some object (e.g. behaviour or
trait), as well as the regular consequences that benefit the system in which the object
or trait is contained”, to stick to Keltner and Haidt’s (1999: 507, italics added) view
on functional explanations of emotion. The use of different levels of abstraction in
analyzing the social functions of some trait or behavior is also common in general
social theory and in multi-agent systems design (cf. below) and has proven to be
particularly helpful in micro-macro analyses of social structural dynamics (Lawler,
Ridgeway, and Markowsky, 1993; Sawyer, 2003; Wiley, 1988). In discussing Elster’s
notion of functional explanation, Castelfranchi (2001) convincingly argues that in
order to elucidate what is functional for a social system the relationship of functional
behavior and cognitive agents’ mental representations has to be thoroughly understood.
Cognitive (BDI) architectures are probably a suitable tool for the further analysis of
this relationship, even though such an analysis requires the profound consideration of
emotion, as Castelfranchi (2001: 11) casually indicates and we will elaborate in more
detail now.
In an artificial intelligence context, the major problem arising out of this perspective on the social functions of emotion seems to be a physiological and biological
rather than a cognitive one and is indeed a problem almost every disembodied agent
system has to face. Although the cognitive aspects of emotion (and vice versa) are far
from being thoroughly understood—let alone computationally modeled—some of the
above mentioned theories, as well as our own interdisciplinary approach (von Scheve
and von Luede, 2005), highlight the role of physiological processes in analyzing the
functions of emotion. These approaches stand in sharp contrast to other social constructionist perspectives on emotion that largely deny any importance of physiology.
Simon Clarke (2003) has aptly outlined this dilemma in the case of envy, thereby
emphasizing the importance of psychoanalytic theory and the limitations of a purely
social constructionist, i.e. cognitivist, account particularly in explaining the role of
emotion in large-scale behavioral dynamics.
My agents love to conform
Given these issues, there are, nevertheless, some reasons we are optimistic that such
an endeavor might indeed be successful: First is the fact that models of emotion that
encompass the physiological level are just beginning to emerge (e.g., Avila-Garcia
and Canamero, 2005; Mendao, 2004; de Almeida, da Silva, and Bazzan, 2004) and
might well contribute to more sophisticated multi-level theories also incorporating
the physiological level. Second, the role of cognition in emotion has been exhaustively accounted for, but we still see the potential of further refinement of existing
models. Third, we are foremost concerned with modeling interactions and communication between the different layers, and might therefore duly leave some of the aspects
temporarily black-boxed.
Normative behavior and emotion
In many, if not most, normative multi-agent systems norms are conceptualized with
reference to deontic logic as explicitly represented obligations, duties, and conventions determining “that something ought or ought not to be the case” under specific
circumstances (Opp, 2002: 132). In addition, norms are characterized by various features, e.g. their acceptance, maintenance, social distribution, formalization, rationale,
function or purpose (Verhagen, 2000). Whatever aspect is highlighted in the various
approaches to normative multi-agent systems, many entities in these systems are assumed to be “norm autonomous,” i.e. may autonomously and intentionally decide to
follow or violate a norm, supposed that they have full knowledge of actually prevailing norms in a social system and a specific social situation and perfect information on
their internal milieu and external environment. Furthermore it is often assumed that
agents are able to deliberately and collectively issue norms (Castelfranchi et al., 1999;
Dignum, 1999).
This perspective on deliberative normative agents raises a couple of issues, in
particular when dealing with problems of micro-macro linkage. We will show that
actors are not as “autonomous” as they might seem when it comes to norms, and that
other forms of normative behavior are equally important when dealing with micromacro issues. First and foremost is the problem of deliberate choice: If an agent may
always autonomously decide either to follow or to violate a norm, then wherein lies the
purportedly outstanding and compulsive power of norms and on what basis are such
decisions derived? In highly functional and institutionalized domains such as markets,
law, or politics, decision-makers usually act rationally in order to maximize individual
or collective outcomes. Given a rational normative agent, its decision to adhere to a
norm will therefore be guided by the subjectively expected outcome of the different
alternatives. However, there are good examples where abiding to a norm is highly
irrational in terms of costs and benefits, e.g. in vendettas, altruistic punishment, tax
compliance, or voting in large-scale elections. Nevertheless, most actors still do follow
these norms (Elster, 2004). Another example is the fact that people follow norms even
in solitude and in absence of probable sanctions (including loss of reputation), for
example in adhering to table manners or by contributing to the provision of public
goods (e.g. protection of the environment).
In addition, the problem of deliberate choice is particularly evident when it comes
to what is commonly understood as social norms, i.e. norms that are not only socially
C. von Scheve, D, Moldt et al.
shared but in addition involve third-party punishment as a defining feature and a
normative action in itself. Evidence suggests that third-party punishers, which are
by definition materially unaffected by norm violations, carry out sanctions even at
their own costs with no immediate benefits (material, reputation, deterrence) (Fehr
and Fischbacher, 2004a). Why do they choose to do so? It has also been shown that
third-party punishment, i.e. the existence of a social norm prescribing the punishment
of violators, promotes evolutionary more stable exchange strategies in iterated dyadic
interactions than, for example, second-party punishment (Bendor and Swistak, 2001).
This brings us to another issue, namely the emergence of norms. If norms are to
be intentionally issued for instrumental, i.e. socially functional reasons, for example
to promote the fitness of some large-scale social system, then it seems highly improbable that there exists an individual or corporate norm-legislating entity capable of
determining what is in fact most functional for such a large-scale system, thereby simultaneously accounting for different units of measuring utility, differing time-scales,
imperfect information, etc. The intentional issuing of norms seems to be practicable
in small to medium-scale social systems, e.g. organizations and communities, but
highly impracticable as the size and complexity of a system increases (cf. Bendor and
Swistak, 2001: 1495). Many norms are not issued intentionally but rather emerge from
the unintended consequences of actions, for example table manners, dress codes, and
fairness rules (cf. Opp, 2002). Given these issues, some researches in the social and
behavioral sciences still embrace the idea of explaining the social functions of norms
solely by referring to their rational qualities (Opp, 2002; Horne and Cutlip, 2002;
Bendor and Swistak, 2001), while others—such as us—do not (Elster, 1996, 2004;
Frank, 1993, 2001).
Robert Axelrod in a seminal article on the evolution of norms uses a behavioral
definition of norms: “A norm exists in a given setting to the extent that individuals
usually act in a certain way and are often punished when seen acting not in this way”
(Axelrod, 1986: 1097). Although Axelrod applies this definition to a game-theoretic
context, we will use it to suggest a behavior-based approach to normative agents and to
explain the functions of norms and emotions in a micro-macro context. This does not
mean, however, that we categorically disregard the possibility of explicitly represented
norms. All we aim at is the distinction between an explicit representation of a norm
on the one hand, and the existence of unintentional normative behavior that is not
based on some propositional mental content. A norm may well emerge from initially
uncoordinated and then for some reason (or social function, if one wants) successively
increasing cooperative behavior, and later on (if at all) become explicitly represented,
for example as a formal law. Or, on the other hand, a norm is intentionally formulated,
issued and adopted, and thus promotes normative behavior.
If we assume that social norms constrain actions by denoting some options for
action as more adequate than others, then it is also plausible to assume that social
structuration emerges in such a way that certain actions under certain situational conditions are not being implemented at all, while other options are constantly preferred,
leading to the emergence of robust “structuring practices” (Knorr-Cetina, 1981). In
this respect, Cristiano Castelfranchi (2001) gives a good account of the interrelation of
norms and social functions (and, casually, emotion). Although his definition of norms
is not in line with our concept, he develops an interesting stimulus-reinforcement
learning model to account for structural effects of individual behavior. In addressing
My agents love to conform
the “foundational theoretical problem of the social sciences—the possibility of unconscious, unplanned emergent forms of cooperation, organization and intelligence
among intentional, planning agents” (Castelfranchi, 2001: 6), Castelfranchi suggests
that the reinforcement of goals and beliefs as the main determinants of social action
in effect leads to the reinforcement of corresponding goal-directed behavior. In doing
so, he postulates two central mechanisms, namely the strengthening of associations
between beliefs/goals and situational contexts on the one hand, and the confirmation
of beliefs supporting a desired behavioral outcome on the other hand.
Although Castelfranchi criticizes Bourdieu’s “habitus” for being far too deterministic and lacking an explanation of the social functions of intentional action, his model
of reinforcement learning at least to us strongly resembles the very characteristics
of habitual behavior (Castelfranchi, 2001: 25f). As far as we can assess, Bourdieu
did not favor an opposition between intentional behavior as non-functional and the
habitus as being socially functional. On the contrary, empirically—and this relates to
a general problem with Castelfranchi’s account and a common misunderstanding of
Bourdieu’s habitus concept (see Lizardo (2004) for an excellent clarification on this
matter)—there is no such thing as “pure” conscious intentional behavior devoid of
any social influence or habitualization (cf. von Scheve and von Luede, 2005). Notably,
Castelfranchi in his layered architectural approach argues for exactly this interaction
of associative, habitualized low-level mechanisms and high-level cognitive functions,
in that he proposes an additional layer of automatic, associative mechanisms resting
upon the layer of explicit mental representations, propositional attitudes, and higher
cognitive functioning. Although this kind of behavior is considered social functional
(i.e. structurally reinforcing), Castelfranchi vividly denies its normative status. “Normative behaviour has to be intentional and conscious: it has to be based on knowledge
of the norm (prescription), but this does not necessarily imply consciousness and intentionality relative to all the functions of the norm” (Castelfranchi, 2001: 31; italics
omitted). In contrast to this view, we argue that the potential of social norms in explaining social structural dynamics is vainly looked for solely in the area of conscious
and intentional action. If one conceives of norms as mental objects, i.e. “hybrid configurations of beliefs and goals”, as Conte and Castelfranchi do (1995: 192), then there
is no obvious reason to assume that behaviorally enforced goals and beliefs do not in
turn promote habitual and at the same time norm-abiding behavior.
What, then, is the crucial role of emotion in these processes? It has been repeatedly shown that the mechanisms of association and confirmation, herein proposed to
enforce goals and beliefs, are inextricably tied to cognitive-emotional interactions,
for example in memory formation, stimulus-reinforcement learning, mood congruency, state-dependent recall of information, and social judgments (Forgas, 1995, 2000;
Clore, Schwarz, and Conway, 1994; Rolls, 2004; Bless, 2000). From a design perspective, in layered architectures emotion ultimately facilitates the processes constituting
the supplementary automatic, associative layer on top of the procedural reasoning and
reactive layers, for example in Sloman’s (2001) “CogAff” system, Staller and Petta’s
(2001) TABASCO architecture, in Castelfranchi’s hybrid model (Castelfranchi, 2000,
2001) and in our own work (von Scheve and Moldt, 2004). This perspective is the
outcome of diverse empirical and theoretical investigations into the inextricability of
cognition and emotion, which we cannot account for in its entirety here (cf. Power and
Dalgleish, 1997; Frijda, 1994; LeDoux, 1994). Thus, even if we conceive of normative
C. von Scheve, D, Moldt et al.
behavior as, and only as, conscious intentional behavior, there is no reasonable way
of bypassing emotion as long as one is committed to modeling socially intelligent
behavior in human and non-human animals. Jon Elster has formulated this position
to the point in paraphrasing Max Weber, holding that “a social norm is not like a
taxi from which one can disembark at will” (Elster, 1989: 106). Also, regarding the
issues outlined by Brooks (1991), a behavioral approach to norms might at least be an
alternative worth exploring in multi-agent systems research.
Emotions enforce and maintain social norms
Besides the general influence of emotion on unintentional normative and socially functional behavior just portrayed, there is another noteworthy social function of emotion
strongly related to social norms that is particularly evident in social interactions. This
relation not only points to the fact that normative behavior is guided by emotions,
but also to the fact that the maintenance and enforcement of norms is closely tied
to emotion as well. The further analysis of this perspective strongly suggests an approach also taken by Conte and Castelfranchi (1995), namely to view a norm—as
such—as a mental object, i.e. a hybrid configuration of beliefs and goals (which in no
way, as illustrated above, underestimates the power of unintentional normative behavior). Endorsing this approach, the social, temporal, and spatial distribution of norms
makes them instances of a macro system; at the same time, however, their definition
as configurations of propositional attitudes (i.e. beliefs and goals) renders them an
instance of the micro level and accordingly a primary subject matter of different kinds
of reasoning processes (Fodor, 1981; Dretske, 1988). Conte and Castelfranchi (1995)
argue in analogy, defining norms as directives or instructions which are represented
as beliefs, substantially determining behavior by generating new goals, i.e. normative
goals. However, the decisive questions in this respect, “how and why does a normative
belief come to interfere with x’s decisions? What is it that makes her [an actor] responsive to norms concerning her? What is it that makes a normative belief turn into a
normative goal?” (ibid. 192) remain largely unanswered, though a satisfactory answer
is essential if one wants to find a solution for the “foundational theoretical problem”
mentioned by Castelfranchi.
In what follows, we argue that emotion might be a loophole here and that Jon
Elster’s (1999) conceptualization thereof is suited to serve as an addition to existing
positions. Elster delivers a definition of certain qualities of social norms rather than
of the concept of a social norm itself. Accordingly, social norms can be described as
follows (Elster, 1999: 145f): First, social norms are non-outcome-oriented phenomena.
They can have unconditional imperative character but also conditional if they refer to
past actions. Second, social norms are shared with other members of a social system in
which the process of sharing is also socially shared. Third, and this is a consequence
of the second characteristic, norms are subject to direct and third-party punishment,
whereas punishing is prescribed by a norm.
Some evidence
When analyzing the question why agents adhere to norms and social systems exhibit
large-scale normative behavior, it is critical to examine the kinds of sanctions in case of
My agents love to conform
non-compliance with norms. Particularly in economic theory sanctions resulting from
non-compliance are described as a withdrawal of material resources (Becker, 1976).
Withdrawing or withholding material resources (e.g. loss of money or other resources,
refusal to cooperate, breach of a contract), however, is by no means the definitive or
most efficient way to successful sanctioning. Even more decisive in this respect is the
fact that norm-violators usually interpret material sanctions also as a vehicle for the
expression of strong negative emotions, e.g. contempt, disdain, detestation, or disgust.
In consequence, violators feel shame or guilt. The experience of shame in this respect
will be even worse because—and in contrast to guilt—the perspective of the sanctioner
is incorporated and accounted for more vividly. Furthermore, it has been argued that
shame indicates severe threat to an actor’s social bonds (Scheff, 2003).
In this context, Jon Elster has shown that the material aspect of sanctions lies solely
within the question of how much it costs the punisher to impose the sanction, and not
within the question of how severe the sanctions are for the offender (Elster, 1999: 146).
To clarify: The higher the costs a punisher accepts to implement an intended sanction,
the more vividly aware is the offender of the negative emotions accompanying the
implementation of these sanctions, and the more strongly the offender will feel the
resulting shame. The sanctioning costs a punisher is willing to accept signals to the
offender the severity of the norm digression. It is no exception to the rule that punishers
accept enormous costs by far outreaching the damage an offender has caused. But these
costs are by no means futile, quite the contrary they exemplify the negative emotional
meaning that accompanies the sanction and emphasize the offender’s obligation to
feel guilt or shame.
These mechanisms have also been validated empirically and experimentally. For
example in one-shot ultimatum games, participants carry out altruistic punishment at
their own costs with no material rewards or possible deterrence, thereby enforcing the
norm of fairness. Ernst Fehr and his group (Fehr and Gaechter, 2002) have shown that
emotion is a proximate cause of altruistic punishment. Their experiment illustrates
that free riding in public goods games, i.e. the violation of the norm of reciprocal
cooperation, reliably causes strong negative emotions, that most free riders do indeed
expect these emotions, and that they in turn trigger punishment. Fehr and Gaechter
conclude that cooperation thrives if altruistic punishment is possible, and collapses
if altruistic punishment is ruled out. In another experiment Sanfey and colleagues
(2003) have shown that unfair offers in an ultimatum game, i.e. the violation of fairness norms, reliably modulate anterior cingulate activity, an area of the human brain
that is “responsible” for negative affects such as pain and distress, thirst and hunger.
Another astonishing study has shown that punishment of norm violators activates
human’s neural reward-circuitry (mainly the dorsal striatum), indicating that actors
derive satisfaction from punishing defectors (de Quervain et al., 2004). A well known
side effect in this case usually is the relief experienced from one’s anger ceasing as a
consequence of delivered punishment.
Further evidence comes from studies analyzing the violator’s perspective. Naomi
Eisenberger and colleagues found that social rejection, i.e. social exclusion or even
ostracism, possibly as a consequence of the violation of a social norm, modulates
human’s anterior cingulate cortex, an area usually “responsible” for processing physical pain. They conclude that social pain, as experienced in shame or guilt states, and
physical pain share some of the same neural circuits and corresponding computational
C. von Scheve, D, Moldt et al.
mechanisms (Eisenberger, Lieberman, and Williams, 2003; Eisenberger and Lieberman, 2004). Still others found that the perception of a norm transgression activates
specific brain regions usually involved in representing aversive emotional reactions in
others (Berthoz et al., 2002).
Action incentives
The evidence just outlined to us suggests that norms may well be represented as propositional attitudes, but in order to function as such, i.e. to generate normative behavior
and to be enforced and maintained, emotion is needed. What makes actors responsive
to social norms is neither external punishment nor rational interest alone. It is (also)
the anticipation of negative emotions that arise in case of a norm violation and the
satisfaction derived from punishing violators. Now, what consequences does “emotional deterrence” have for one’s options to act? Striving for emotional gratification,
i.e. the motivation to seek out interactions resulting in positive emotions and to avoid
those resulting in negative emotions is considered a basic motivator of human behavior (Turner, 1999; Collins, 1984). Thus, emotions can well relate to future actions by
substantially affecting the generation of plans and many other cognitive competencies
(Loewenstein and Lerner, 2003). Anthony Giddens, for example, considers concerns
about the loss of ontological security as a central aspect in his theoretical framework:
fear of loosing ontological security and facticity is a primary motivator of social action
(Giddens, 1991). Other authors, for example Randall Collins (1984) or Michael Hammond (1991), who regard emotional gratification as a motivator of action directly
scalable to large-scale contexts, assume that actors have an inborn need for positive
emotional exchange processes, which may solidify to “interaction ritual chains” and
thus contribute to the emergence of robust social structures (Collins, 1981; Collins,
In view of the interaction of emotion, norms, and normative behavior, we can further
assume that shame and contempt in particular serve as a vehicle for maintaining
norms by generating normative goals on the one hand, and goals of avoidance of
adverse consequences on the other hand. The goal of compliance with social norms
therefore is not necessarily generated as a consequence of the anticipation of a loss
of material resources through sanctions, but instead as a result of the fear of emotiondriven sanctions (by way of negative emotions such as contempt, disdain, detestation
or disgust in the punisher), that would result in negative emotions, e.g., shame, guilt,
or embarrassment, in the violator.
An approach to modeling
Improving AI systems by taking into account emotion or functionally equivalent mechanisms is not a new idea, its origins rooted in the contributions of researchers like Herbert Simon (1967), Aaron Sloman and Monica Croucher (1981) or Marvin Minsky
(1986). In the late 1980s first reviews of AI-models of emotion were assembled (Dyer,
1987; Pfeifer, 1988; Ruebenstrunk, 1998) and until now, research on emotion within
computer science has mainly revealed three basic motivations: performance, humancomputer interaction, and the simulation of naturally occurring phenomena (Picard,
1997; Wehrle, 1998; Scheutz, 2002). As indicated above, research on emotional agents
My agents love to conform
has largely been concerned either with isolated entities or dyadic interaction settings
(agent-agent and agent-user) and almost all of the models in question are based on
psychological and neuroscientific theories of emotion, as we have delineated for the
field of human-computer interaction previously (Moldt and von Scheve, 2002, 2002a).
However, as we have briefly shown in the preceding sections, emotion is highly socially functional not only in dyadic settings but also in large-scale social systems, one
of the facilitators of functionality being the interplay with social norms. Therefore it
seems evident that if norms are employed as a coordination principle in multi-agent
systems, the use of emotion suggests itself.
Our approach to modeling the interdependencies and reciprocities of the social
functions of emotion first and foremost aims at locating emotion in a micro-macro
framework, as initially outlined above. This is further facilitated by taking a layered perspective on sociality that can be represented by the multi-agent architecture MULAN, providing a conceptually highly flexible framework. The MULAN
architecture is a technical implementation of the agent concept that may simultaneously account for almost any aspect of an application model. Separation of the
two perspectives is achieved by the SONAR agent architecture in making use of
an array of sociological concepts (cf. von Luede, Moldt, and Valk, 2003; Koehler
et al., 2005). In particular, for each and every social entity constituting the different
societal layers—i.e. actors, social processes, and social structures—a single SONAR
agent is deployed and made available on the MULAN platform. The social units’
inherent logics are in turn represented by multi-agent systems which are directly subordinated to the respective units. This framework has proven to be quite useful in
modeling micro-macro aspects of behavior in organizational and institutional social
contexts (cf. von Luede, Moldt, and Valk, 2003).
Representing emotion and norms in SONAR
Michael Koehler and colleagues (2003) provide a modeling framework conceiving
of social entities, i.e. actors, processes, and structures, as first-order objects that can
be modeled side by side simultaneously. The framework allows the thorough representation of direct interdependencies occurring on identical layers of observation
(abstraction), whereas the internal properties of a particular social entity may still be
entirely encapsulated from direct access by other social entities of the same layer. The
social entities’ internal logics are largely autonomous and may thus differ from one
another substantially. For example, any actor might maintain an arbitrarily complex
representation of its (social) environment, which is in turn represented as a set of
specifically networked social entities, i.e. SONAR-based multi-agent systems. The
system can be quite simple in case of some primitive agents, it can become, however, highly complex when agents are characterized by increasing degrees of social
Imagine, for example, the mind of some muddle-headed professor and her representations of the external and internal world in its entire complexity—including all
mental contents and inconsistencies. Being an exemplification, few would claim to be
able to model a real human being’s mental contents. Rather, the designer might well
simplify or extend the model according to the requirements of the task at hand. The
C. von Scheve, D, Moldt et al.
same would also be true for the social processes that are usually established between
social agents as well as for considerably stable (but not unalterable) social structures.
Therefore, social entities in this framework in themselves contain the necessary references to other entities, or, to be more explicit, are to some extent made up of these
references. Evidently, these references do not have to be modeled independently. Part
of this architectural layout indeed reflects what human actors do in everyday social
interactions: continuously modeling (representing) other actors and mentally relating
to them, thereby constituting what is dubbed in the social sciences as “intersubjectivity”. And, even more important, the structural layer, i.e. social structural effects, to
some extent determine the references and representations on the lower layers while in
turn being constituted by them through, e.g., shared knowledge, shared beliefs, and
collective acceptance (Tuomela, 2002; Searle, 1995).
In designing a specific system, the designer is required to deliberately select specific
aspects of a system which are supposed to be crucial for the task at hand. SONAR
models can then exactly specify the model’s most important elements. Whereas the
MULAN environment provides the technical framework for implementing key concepts like autonomy, mobility, cooperation, and adaptation, the SONAR architecture
is used to model the internal representations of an entity by means of a multi-agent
system. Figure 1 illustrates the interplay of two autonomous social entities, in this
case an actor and the social process it is involved in. Social structural embedding is
achieved by the social processes that are involved in the evolution, reinforcement, and
reproduction of new structures. Conceptually, we have to deal with the same kind of
connection between both parts of the model by means of synchronous channels (see
Figure 1 depicts a powerful variant of Petri nets—a reference net (Valk, 1998).
Rectangles (transitions) represent actions, whereas circles (places) denote available
or unavailable resources or conditions that may be fulfilled. An arc determines the
specific context of the transition. Thus, arcs that are directed from transitions to places
can be interpreted as preconditions for action, whereas arcs that are directed from
places to transitions represent actions’ outcomes. A firing transition (or an action that
is implemented) will remove resources or conditions (for short: tokens) from places
and insert them into some other place. A notable property of a reference net is the
possibility that the tokens located on a place in a net (the system net) may again be a
reference net (an object net) (or any other Java object) (Valk, 1998). Object net and
system net are synchronized through synchronous channels, whereby one of the nets
is expected to reference the other one. For example, as illustrated in figure one, actions
of an actor are synchronized with some social process by two transitions constituting
the synchronous channel: “observe” and “act” in the actor net, in this case modeled as
an object net, and “access observability” and “expand action repertoire” in the social
process net, in this case the actor net’s system net. Needless to say, tokens a1 and a2
could also be modeled as additional actor (object) nets.
We should point out that an increase in resolution is facilitated by “zooming” into
the references to the relevant actor nets. As we have said before, any social entity is in
itself essentially autonomous. In principle, the mutual relationships between them are
assumed to be symmetrical, even though modeling them might imply some (rational)
hierarchical structure. The social interactional concepts we have used, e.g. “observation” and “action”, are accounted for solely in the process entity. Any adjustment
My agents love to conform
Fig. 1 A reference net representing actor, social process, and social structure
succeeds locally, without direct interaction with the social environment. Further details had to be omitted here, but can be found in either von Luede, Moldt, and Valk
(2003) or in Koehler et al. (2005).
Although we have already used a range of concepts from the social sciences within
the SONAR models, for example acknowledgement, observation, action, etc., the role
of emotion has hardly been taken into account. Hence, we propose some simple but
fundamental enhancements to SONAR that still allow for a conventional approach to
modeling, therefore treating all aspects concerning emotion separately for the time
being. This requires some explicit decisions in view of what belongs to an emotion
proper and what does not. Given this perspective, the problem of adequately connecting
both models becomes quite obvious. We aim at solving this issue in a simple and
homogenous manner with the applied modeling techniques: existing MULAN models
C. von Scheve, D, Moldt et al.
of organizational micro-macro behavior are to be completed with additional models
of the functions of emotion in large-scale systems.
Corresponding findings in emotion theory rather explicitly dealing with the nature
of emotion as such might be represented in a segregated emotion model. On the other
hand, in case the emotion theories under scrutiny state explicit references to social
processes and structures (e.g., sociological models), they may well be linked to the
conventional, rational elements of the model by synchronous channels, thus creating
a kind of homologous “parallel world”. Modeling each and every social entity as an
individual net allows depicting their mutual relations by means of references in the
sense of a “pointer”. Synchronous channels in this case may represent any possible
interconnection of system behavior in distinct parts of the model.
We are well aware of the fact that the explicit separation of emotion from other
parts of the model is prone to critique in that it seemingly promotes a dichotomy
between “rational” and “emotional” perspectives that most researchers would nowadays consider obsolete. However, we favor an initial explicit technical separation
and simplification in order to achieve just that: to be able to illustrate precisely and
concisely in semi-formal semantics the mutual and supposedly reciprocal and recursive interconnections between the two domains. The necessity for such models
in social science research has been put forward by Jonathan Turner: “Such models
[...] can provide a picture of process—that is, of how variables influence each other
across time. Moreover, they can also give us a view of complex causal processes.
Too often in sociology, we employ simple causal models [...] that document one-way
causal chains among empirical indicators of independent, intervening, and dependent
variables. But actual social processes are much more complex, involving feedback
loops, reciprocal causal effects, lag effects, threshold effects, and the like” (Turner,
1988: 17).
Obviously, this is no means to overcome general system complexity, although
complexity can be allocated to different layers, their integration demanding still further
efforts because the linkages between the different layers are to be located and modeled
explicitly. (In this respect we might, for example, apply solutions of similar problems
found in the combination of different views in models created with the unified modeling
language.) Although at some point an integration of the two perspectives is necessary
and desired, we currently favor the advantages of an explicit separation, in particular
the intuitive simplicity of the detached points of view.
Smaller and less complex models can usually be implemented much faster with
programming languages that do not distinguish between different points of view.
However, the separation of different points of view in order to handle a system’s
complexity is a common method in computer science. Apart from issues of technical
complexity, emotion (and other) theories in the social and behavioral sciences are
also highly complex systems that might demand even more flexibility as is generally
required in construction-oriented computer science models. Therefore, a separation
into different layers seems to be useful not only for social theory but also in view
of a sharper examination of different analytical layers. The possibility of modeling
discrete emotions and their specific components as explicit states and processes which
are integrated into the existing (rational) models still remains largely unelaborated,
albeit it is one goal of our future work.
My agents love to conform
Finally, we will very briefly evaluate the modeling approach presented above as well
as the possibilities it holds in view of modeling the interrelation of norms and emotion. Generally speaking, accounting for emotion in our modeling and simulation
framework of behavior in organizational and institutional social contexts incorporates
further aspects of reality that might contribute to better solutions, particularly when
it comes to modeling informal social interactions. In addition, modeling the social
functions of emotion is supposed to improve multi-agent systems and normative systems in particular, for example in view of alternative coordination solutions. However,
strong evidence has yet to be presented that AI-models of emotion do indeed facilitate
better solutions.
In addition, the approach allows the investigation of the relationship between cognition and emotion by means of explicitly separated components. This is particularly
important in view of the interaction of norms as explicitly represented mental objects
and emotions as processes with non-propositional output. The analysis of different
components of the emotion process belonging to different representational formats is
also of interest for original emotion research. Currently, there is much debate on the
question of how information that is represented in different representational formats
and memory structures (i.e., semantic vs. non-propositional) interacts in the generation
of emotion (cf. Reisenzein, 2001).
Thus, the interrelation between social norms and emotion—and between two different representational systems—can be examined on different levels of social abstraction. On the structural layer, norms may either be related to concepts largely disregarding emotion, or with those considering emotion on all layers of a model (actor, process,
and structure). This openness and flexibility permits an analysis of different theories
and also computational models of emotion simultaneously and to different extents.
One of the core advantages of the framework is that existing emotion-based agent
architectures which are primarily concerned with representing actor- and eventually
process-layers (e.g., Staller and Petta, 1998, Sloman, 2001; Cami, Lisetti, and Sierhuis, 2004)) might well be integrated (and extended), probably on the protocol-layer,
as emotion generating entities.
Abe JA, Izard CE (1999) The developmental functions of emotions: An analysis in terms of differential
emotions theory. Cog Emotion 13(5):523–549
de Almeida LB, da Silva BC, Bazzan AL (2004) Towards a physiological model of emotions: First steps.
In: Proceedings of the 2004 AAAI Spring Symposium. AAAI Press, Menlo Park, CA
Averill JR (1992) The structural bases of emotional behavior. In: Clark MS (ed) Emotion. Sage, Newbury
Park, CA, pp 1–24
Avila-Garcia O, Canamero L (2005) Hormonal modulation of perception in motivation-based action selection architectures. In: Proceedings of Agents that Want and Like: Motivational and Emotional Roots
of Cognition and Action. Hatfield, UK
Aube M, Senteni A (1996) Emotions as commitments operators: A foundation for control structure in multiagents systems. In: Proceedings of the 7th European Workshop on Modelling Autonomous Agents in
a Multi-Agent World (MAAMAW’96). LNAI Vol. 1038. Springer, Berlin, pp 13–25
C. von Scheve, D, Moldt et al.
Axelrod R (1986) An evolutionary approach to norms. Am Political Sci Rev 80(4):1095–1111
Bazzan AL, Adamatti DF, Bordini R (2002) Extending the computational study of social norms with a systematic model of emotions. In: Proceedings of the 16th Brazilian Symposium on Artificial Intelligence
(SBIA’02). LNCS Vol. 2507. Springer, Berlin, pp 108–117
Bendor J, Swistak P (2001) The evolution of norms. Am J Sociol 106(6):1493–1545
Becker G (1976) The economic approach to human behavior. University of Chicago Press, Chicago, IL
Berthoz S, Armony JL, Blair RJ, Dolan RJ (2002) An fMRI study of intentional and unintentional (embarrassing) violations of social norms. Brain 125(8):1696–1708
Bless H (2000) The interplay of affect and cognition: The mediating role of general knowledge structures. In: Forgas JP (ed) Feeling and Thinking. Cambridge University Press, New York, NY, pp 201–
Brooks RA (1991) Intelligence without representation. Artif Intell 47(1–3), 139–159
Cacioppo JT, Berntson GG, Larsen JT, Poehlmann KM, Ito TA (2000) The psychophysiology of emotion.
In: Lewis R, Haviland-Jones JM (eds) Handbook of emotions, 2nd Ed. Guilford, New York, NY pp
Cami A, Lisetti C, Sierhuis M (2004) Towards the simulation of a multi-level model of human emotions.
In: Proceedings of the 2004 AAAI Spring Symposium. AAAI Press, Menlo Park, CA
Canamero D (ed) (1998) Emotional and intelligent: The tangled knot of cognition. In: Proceedings of the
1998 AAAI Fall Symposium. Menlo Park, CA: AAAI Press
Carley K (1986) An approach for relating social structure to cognitive structure. J Math Sociol 12(2):137–
Castelfranchi C (2000) Affective appraisal vs. cognitive evaluation in social emotions and interactions. In:
Paiva A (ed) Affect in interactions. Springer, Heidelberg
Castelfranchi C (2001) The theory of social functions: Challenges for computational social science and
multi-agent learning. Cogn Sys Res 2(1):5–38
Castelfranchi C, Dignum F, Jonker CM, Treur J (1999) Deliberative normative agents: Principles and architecture. In: Proceedings of the 6th International Workshop on Intelligent Agents VI, Agent Theories,
Architectures, and Languages (ATAL). LNCS Vol. 1757. Springer, London, pp 364–378
Clarke S (2003) Psychoanalytic sociology and the interpretation of emotion. J Theory Soc Behav 33(2):143–
Clore GL, Schwarz N, Conway M (1994) Affective causes and consequences of social information processing. In: Wyer RS, Srull TK (eds) Handbook of social cognition. 2nd Ed. Erlbaum, Hillsdale, NJ,
pp 323–417
Clore GL (1994) Why emotions are felt. In: Ekman P, Davidson RJ (eds) The nature of emotion. Oxford
University Press, New York, NY pp 103–111
Coleman JS (1990) Foundations of social theory. Harvard University Press, Cambridge, MA
Collins R (1981) On the microfoundations of macrosociology. Amer J Sociol 86(5):984–1014
Collins R (1984) The role of emotion in social structure. In: Scherer KR, Ekman P (eds) Approaches to
Emotion. Erlbaum, Hillsdale, NJ, pp 385–396
Collins R (2004) Interaction ritual chains. Princeton University Press, Princeton, NJ
Conte R, Castelfranchi C (1995) Norms as mental objects. In: Proceedings of the 5th European Workshop
on Modelling Autonomous Agents in a Multi-Agent World (MAAMAW ’93). LNAI Vol. 957. Springer,
Berlin, pp 186–196
Cosmides L, Tooby J (2000) Evolutionary psychology and the emotions. In: Lewis R, Haviland-Jones JM
(eds) Handbook of emotions, 2nd Ed. Guilford, New York, NY pp 91–115
Dignum F (1999) Autonomous agents with norms. Artif Intell Law 7(1):69–79
Dretske F (1988) Explaining behavior. MIT Press, Cambridge, MA
Dyer MG (1987) Emotions and their computations: Three computer models. Cogn Emotion 1(3):323–347
Eisenberger NI, Lieberman MD, Williams KD (2003) Does rejection hurt? An fMRI study of social exclusion. Sci 302:290–292
Eisenberger NI, Lieberman MD (2004) Why rejection hurts: A common neural alarm system for physical
and social pain. Trends Cogn Sci 8(7):294–300
Elliot CD (1993) Using the affective reasoner to support social simulations. In: Proceedings of the 13th
International Joint Conference on Artificial Intelligence (IJCAI’93). Morgan Kaufmann, Los Altos,
CA, pp 194–200
Elliot CD (1992) The affective reasoner. A process model of emotions in a multi-agent system. PhD Thesis,
Intitute for the Learning Sciences, Northwestern University
Elster J (1989) Social norms and economic theory. J Econ Perspect 3(4):99–117
My agents love to conform
Elster J (1996) Rationality and the emotions. The Econ J 106(438):1386–1397
Elster J (1999) Alchemies of the mind. Cambridge University Press, Cambridge
Elster J (2004) Emotions and rationality. In: Manstead AS, Frijda NH, Fischer A (eds) Feelings and emotions.
Oxford University Press, New York, NY, pp 30–48
Engel P (2002) Intentionality, normativity and community. Facta Philosophica 4(1):25–49
Fehr E, Fischbacher U (2004) Social norms and human cooperation. Trends Cogn Sci 8(4):185–
Fehr E, Fischbacher U (2004a) Third-party punishment and social norms. Evol Hum Behav 25(1):63–
Fehr E, Gaechter S (2002) Altruistic punishment in humans. Nature 415: 137–140
Fix J (2004) Emotionale agenten. Diploma Thesis. University of Hamburg, Computer Science
Fodor J (1981) Propositional attitudes. In: Fodor J (ed) RePresentations. MIT Press, Cambridge, MA, pp
Forgas JP (1995) Mood and judgment: The affect infusion model (AIM). Psychological Bulletin 117(1):39–
Forgas JP (2000) Feeling is believing? The role of processing strategies in mediating affective influences on
beliefs. In: Frijda NH, Manstead A, Bem S (eds) Emotions and beliefs. University Press Cambridge,
Cambridge, pp 108–143
Frank RH (1993) The strategic role of emotions: Reconciling over- and undersocialized accounts of behaviour. Rationality and Society 5(2):160–184
Frank RH (2001) Cooperation through emotional commitment. In: Nesse RM (ed) Evolution and the
capacity for commitment. Russel Sage, New York, pp 57–76
Frijda NH (1986) The emotions. Cambridge University Press, Cambridge
Frijda NH (1994) Emotions require cognitions, even if simple ones. In: Ekman P, Davidson RJ (eds) The
nature of emotion. Oxford University Press, New York, NY, pp 197–202
Gerhards J (1988) Soziologie der emotionen. Juventa, Weinheim
Giddens A (1991) Modernity and self-identity. Polity Press, Cambridge
Gmytrasiewicz PJ, Lisetti CL (2000) Using decision theory to formalize emotions for multi-agent systems.
In: Proceedings of the 2nd ICMAS’00 Workshop on Game Theoretic and Decision Theoretic Agents.
Boston, MA
Hatano G, Okada N, Tanabe H (eds) (2000) Affective minds. In: Proceedings of the 13th Toyota Conference.
Elsevier, Amsterdam
Heckathorn DD (1989) Cognitive science, sociology, and the theoretic analysis of complex systems. J Math
Sociol 14(2/3):97–110
Heise DR, Calhan C (1995) Emotion norms in interpersonal events. Soc Psychol Q 58(4):223–240
Heise D, O’Brien J (1993) Emotion expression in groups. In: Lewis M, Haviland JM (eds) Handbook of
emotions. Guilford, New York, NY, pp 489–497
Hochschild AR (1979) Emotion work, feeling rules, and social structure. Ame J Sociol 85(3):551–
Horne C (2001) Sociological perspectives on the emergence of norms. In: Hechter M, Opp K-D (eds) Social
norms. Russell Sage, New York, NY, pp 3–34
Horne C, Cutlip A (2002) Sanctioning costs and norm enforcement. Rationality and Society 14(3):285–
Jennings NR (2000) On agent-based software engineering. Artif Intell 117(2):277–296
Keltner D, Gross JJ (1999) Functional accounts of emotion. Cogn Emotion 13(5):467–480
Keltner D, Haidt J (1999) Social functions of emotion at four levels of analysis. Cogn Emotion 13(5):505–
Kemper TD (1978) A social interactional theory of emotions. Wiley, New York, NY
Knorr-Cetina KD (1981) The micro-sociological challenge of macro-sociology. In: Knorr-Cetina KD,
Cicourel AV (eds) Advances in social theory and methodology. Routledge & Kegan Paul, Boston, MA,
pp 1–47
Koehler M, Moldt D, Roelke H (2001) Modeling the structure and behaviour of petri net agents. In: Proceedings of the 22nd International Conference on Application and Theory of Petri Nets (ICATPN’01).
LNCS Vol. 2075. Springer, Berlin, pp 224–241
Koehler M, Moldt D, Roelke H (2003) A discussion of social norms with respect to the micro-macro
link. In: Proceedings of the 2nd International Workshop on Regulated Agent-Based Social Systems
(RASTA’03). Edinburgh
C. von Scheve, D, Moldt et al.
Koehler M, Moldt D, Roelke H, Valk R (2005) Linking micro and macro descriptions of scalable social
systems using reference nets. In: Fischer K, Florian M (eds) Socionics: Its contribution to the scalability
of complex social systems. LNCS. Springer, Heidelberg [in press]
Lawler EJ, Ridgeway C, Markowsky B (1993) Structural social psychology and the micro-macro problem.
Sociol Theory 11(3):268–290
Lawler EJ, Thye SR, Yoon J (2000) Emotion and group cohesion in productive exchange. Ame J Sociol
LeDoux JE (1996) Cognitive emotional interactions in the brain. In: Ekman P, Davidson RJ (eds) The nature
of emotion. Oxford University Press, New York, NY, pp 216–223
Lizardo O (2004) The cognitive origins of bourdieu’s habitus. Journal for the Theory Soc Behav 34(4):375–
Loewenstein G, Lerner JS (2003) The role of affect in decision making. In: Davidson RJ, Scherer KR,
Goldsmith HH (eds) Handbook of affective sciences. Oxford University Press, New York, NY, pp
von Luede R, Moldt D, Valk R (eds) (2003) Sozionik: Modellierung soziologischer Theorie. [Socionics:
Modeling Sociological Theory]. Muenster: Lit
Macy MW, Willer R (2002) From factors to actors: Computational sociology and agent-based modeling.
Annu Rev Sociol 28:143–166
Malsch T (2001) Naming the unnamable: Socionics or the sociological turn of/to distributed artificial
intelligence. Autonomous Agents and Multi-Agent Systems 4(3):155–186
Mendao M (2004) Hormonally moderated neural control. In: Proceedings of the 2004 AAAI Spring Symposium. AAAI Press, Menlo Park, CA
Moldt D, von Scheve C (2002) Emotions in hybrid social aggregates. In: Herczeg M, Prinz W, Oberquelle
H (eds) Mensch & Computer 2002. Teubner, Stuttgart, pp 343–352
Moldt D, von Scheve C (2002a) Attribution and adaptation: The case of social norms and emotion in humanagent interaction. In: Proceedings of The Philosophy and Design of Socially Adept Technologies.
Workshop held in conjunction with CHI’02. Minneapolis, MN
Minsky M (1986) The society of mind. Simon & Schuster
Oatley K, Jenkins JM (1996) Understanding emotions. Basil Blackwell, Oxford
Opp K-D (2002) When do norms emerge by human design and when by the unintended consequences of human action? The example of the no-smoking norm. Rationality and Society 14(2):131–
Panzarasa P, Jennings NR (2002) Social influence, negotiation and cognition. Simulation Modelling Practice
and Theory 10(5–7):417–453
Panzarasa P, Norman TJ, Jennings NR (2001) Social mental shaping: Modelling the impact of sociality on
autonomous agents’ mental states. Comput Intell 17(4):738–782
Pfeifer R (1998) Artificial intelligence models of emotion. In: Hamilton V, Bower GH, Frijda NH (eds)
Cognitive perspectives on emotion and motivation. Kluwer, Dordrecht, pp 287–320
Picard RW (1997) Affective computing. MIT Press, Cambridge, MA
Power M, Dalgleish T (1997) Cognition and emotion: From Ordeo to disorder. Psychology Press, Hove
de Quervain DJ-F, Fischbacher U, Treyer V, Schellhammer M, Schnyder U, Buck A, Fehr E (2004) The
neural basis of altruistic punishment. Sci 305:1254–1258
Reisenzein R (2001) Appraisal processes conceptualized from a schema theoretic perspective. In: Scherer
KR, Schorr A, Johnstone T (eds) Appraisal processes in emotion. Oxford University Press, New York,
NY, pp 187–204
Rolls ET (2004) The functions of the orbitofrontal cortex. Brain and Cognition 55(1):11–29
Ruebenstrunk G (1998) Emotional computers. Computer models of emotions and their meaning for
emotion-psychological research. Unpublsihed Manuscript, available from
Saam NJ, Harrer A (1999) Simulating norms, social inequality, and functional change in artificial
societies. Journal of Artificial Societies and Social Simulation, 2(1).
Sanfey AG, Rilling JK, Aronson JA, Nystrom LE, Cohen JD (2003) The neural basis of economic decisionmaking in the ultimatum game. Sci 300:1755–1758
Sawyer RK (2003) Artificial societies. Multiagent systems and the micro-macro link in sociological theory.
Sociol Meth & Res 31(3):325–363
Sawyer RK (2004) Social explanation and computational simulation. Philosophical Explorations 7(3):219–
My agents love to conform
Scheff TJ (2003) Shame in self and society. Symbolic Interaction 26(2):239–262
Scheutz M (2002) Agents with or without Emotions? In: Proceedings of the 15th International Florida
Artificial Intelligence Symposium (FLAIRS’02). AAAI Press, Menlo Park, CA, pp 89–94
von Scheve C, Moldt D (2004) Emotion: Theoretical investigations and implications for artificial social
aggregates. In: Lindeman G, Moldt D, Paolucci P (eds) Regulated agent-based social systems. LNAI
Vol. 2934. Springer, Berlin, pp 189–209
von Scheve C, von Luede R (2005) Emotion and social structures. Towards an interdisciplinary approach.
J Theory Soc Behav 35(3):303–328
Schwarz N, Clore GL (1988) How do i feel about it? The informative function of affective states. In: Fiedler
K, Forgas JP (eds) Affect, cognition, and social behavior. Hogrefe, Toronto, pp 44–62
Searle JR (1995) The construction of social reality. Free Press, New York, NY
Shoham Y, Tennenholtz M (1997) On the emergence of social conventions: Modeling, analysis, and simulations. Artifl Intell 94(1–2):139–166
Simon HA (1967) Motivational and emotional controls of cognition. Psychological Review 74:29–
Sloman A (2001) Varieties of affect and the cogaff architecture schema. In: Proceedings of the AISB’01
Symposium on Emotion, Cognition, and Affective Computing. SSAISB, York, pp 39–48
Sloman A, Croucher M (1981) Why robots will have emotions. In: Proceedings of the 7th International
Joint Conference on Artificial Intelligence (IJCAI’81). Vancouver, pp 197–202
Smith CA, Pope LK (1992) Appraisal and emotion. The interactional contributions of dispositional and
situational factors. In: Clark MS (ed) Emotion and social behavior. Sage, Newbury Park, CA, pp 32–
Staller A, Petta P (1998) Towards a tractable appraisal-based architecture for situated cognizers. In: Proceedings of Grounding Emotions in Adaptive Systems. Workshop Notes of the 5th International Conference
of the Society for Adaptive Behaviour (SAB’98). Zurich, pp 56–61
Staller A, Petta P (2001) Introducing emotions into the computational study of social norms: A first evaluation. Journal of Artificial Societies and Social Simulation 4(1).
Trappl R, Petta P, Payr S (eds) (2003) Emotions in humans and artefacts. MIT Press, Cambridge, MA
Tuomela R (1995) Norms and agreement. European Journal of Law, Philosophy and Computer Science
Tuomela R (2003) Collective acceptance, social institutions, and social reality. Am J Econ Sociol 62(1):123–
Turner JH (1988) A theory of social interaction. Stanford University Press, Stanford, CA
Turner JH (1999) Toward a general sociological theory of emotions. Journal for the Theory of Social
Behaviour 29(2):133–161
Valk R (1998) Petri nets as token objects. An introduction to elementary object nets. In: Proceedings of
Application and Theory of Petri Nets. Lisbon, Portugal. LNCS Vol. 1420. Springer, Berlin, pp 1–
Verhagen H (2000) Norm autonomous agents. PhD Thesis, Royal Institute of Technology and Stockholm
Wehrle T (1998) Motivations behind modeling emotional agents: Whose emotions does your robot have?
In: Proceedings of Grounding Emotions in Adaptive Systems. Workshop Notes of the 5th International
Conference of the Society for Adaptive Behaviour (SAB’98). Zurich
Wiley N (1988) The micro-macro problem in social theory. Sociol Theory 6(2):254–261
Wooldridge M (2000) Reasoning about rational agents. MIT Press, Cambridge, MA
Christian von Scheve graduated in Sociology with minors in Psychology, Economics, and
Political Science at the University of Hamburg, where he also worked as a research assistant
at the Institute of Sociology. Currently, he is a 3rd year PhD student at the University of
Hamburg. He was a Fellow of the Research Group “Emotions as Bio-Cultural Processes”
at the Center for Interdisciplinary Research (ZiF) at Bielefeld University. In his doctoral
thesis he develops an interdisciplinary approach to emotion and social structural dynamics,
integrating emotion theories from the neurosciences, psychology, and the social sciences.
He has published on the role of emotion in large-scale social systems, human-computer
interaction, and multi-agent systems. He is co-editor of a forthcoming volume on emotion
C. von Scheve, D, Moldt et al.
Daniel Moldt received his BSc in Computer Science/Software Engineering from the University of Birmingham (England) in 1984, graduated in Informatics at the University of
Hamburg, with a minor in Economics in 1990. He received his PhD in Informatics from
the University of Hamburg in 1996, where he has been a researcher and lecturer at the
Department of Informatics since 1990. Daniel Moldt is also the head of the Laboratory for
Agent-Oriented Systems (LAOS) of the theoretical foundations group at the Department
of Informatics. His research interests focus on theoretical foundations, software engineering and distributed systems with an emphasis on agent technology, Petri nets, specification
languages, intra- and inter-organizational application development, Socionics and emotion
in informatics.
Julia Fix is currently a PhD student at the Theoretical Foundations of Computer Science
Group, Department for Informatics at the University of Hamburg. She studied Informatics
and Psychology at the University of Hamburg, with an emphasis on theoretical foundations
of multi-agent systems and wrote her diploma theses about emotional agent systems. Her
current research interests focus on conceptual challenges and theoretical foundations of
modelling emotions in multi-agent systems, emotion-based norm enforcement and maintenance, and Socionics. A further research focus are Petri nets, in particular the use of
Petri-net modelling formalisms for representing different aspects of emotion in agent systems.
Rolf von Lüde is a professor of Sociology at the University of Hamburg with a focus
in teaching and research in Sociology of Organizations, Work and Industry since 1996.
He graduated in Economics, Sociology, and Psychology, and received his doctorate in
Economics and the venia legendi in Sociology from the University of Dortmund. His
current research focuses on labor conditions, the organization of production, social change
and the educational system, the organizational structures of university, Socionics as a new
approach to distributed artificial intelligence in cooperation with computer scientists, new
public management, and emotions and social structures. Rolf von Lüde is currently Head
of Department of Social Sciences and Vice Dean of the School of Business, Economics
and Social Sciences, University of Hamburg.

Similar documents


Report this document