CONFLICT: Agents in conflict situation

(extended abstract of the MSc dissertation)

Henrique Teles Campos

Departamento de Engenharia Informática
Instituto Superior Técnico

Advisor: Professor Ana Maria Severino de Almeida Paiva

Abstract—The objective of this document is to address and engage today’s children. Information technologies and
the problem of how to create groups of agents that engage computer and video games have show to be a great tool for
in natural conflict situations, inspired by what happens in teaching [2], educating [3] and raising awareness [4] about
real world scenarios. For such, we propose that emotional
processes are a key aspect to convey conflict situations’ aspects, subjects on students.
such as, emergence and escalation. Further, we intended to Motivated by the conflict resolution education and the
populate an educational conflict resolution game with these uprising of serious games for teaching children, the SIREN
agents. To achieve this, we investigated conflict theory in project [5] intends to create a conflict resolution game.
Social Psychology. Then we reviewed serious games, where we In this research, we are concerned about the NPCs
analysed their purposes and how they were achieved. With
that, we proposed a conflict model, where the agent observes a (agents) that will populate the serious game. In order to do
situation, then makes assumptions about it and finally behaves that, a rather general problem that we try to handle is stated
towards it. This model was implemented in a emotionally- as follows:
driven agents’ architecture. Then, it was integrated in a conflict
resolution game demonstrator, the Dream Theatre game. A How can we create autonomous agents, for a
world simulation framework supported the integration. With serious game, that are able to identify social
this, we conducted an experiment to assess our model’s conflict situations with other agents and handle
effectiveness. The experiment consisted in preliminary tests them in a believable way?
and a final between groups evaluation. In both, participants
had to watch a recorded video of a user interacting with Our research focuses on agents’ behaviours towards con-
the Dream Theatre demonstrator. The final evaluation results flict between themselves and how this aspect affects agents’
were consistent with our hypothesis, which states that agents believability. Our approach to solve this problem will be
with a decision-making process which stems from an emotion- by investigating what are the most important characteristics
oriented architecture will provide a believable display of overt of human interactions within a conflict situation. We will
manifestation of conflicts.
aim particularly at what causes conflict, its protagonists, its
effects in individuals and how it can be resolved. Within
I. I NTRODUCTION the latter referenced aspect of conflict, we are particularly
Over the past few years, there has been a great increase of concerned about which strategies individuals use to handle
investigation in the AI field of Multi-Agents systems (MAS). conflict and what factors makes them use such handling
Particularly, in the field of synthetic characters, researchers mechanisms and what makes conflicts escalate.
are aiming towards greater believability. Based on the several aspects of conflict identified in the
On this research, we investigate on how to imbue synthetic previously described investigation, we will develop a model
characters with human-like behaviours towards conflict sit- for conflict, which will be integrated in the architecture of
uations. Therefore, increasing their believability. an agents’ society, providing the agents with the ability to
Conflict is a natural part of your lives. It is present recognise conflicts and handle them autonomously. We will
in our everyday basis. In our society, conflict has been also integrate our agents in a simple serious game scenario
considered to be something to avoid due to the negative that will be populated with the agents’ society. Thus, with
feelings and destructive behaviours [1]. However, recent this research we will try to prove the following hypothesis:
research acknowledges that conflict can yield beneficial If the agents’ decision-making process stems from
aspects as well [1]. As such, conflict resolution strategies an emotion-oriented architecture, their behaviours
are essential behavioural skills that all of us need to learn will be based on emotional grounds and intensity,
since early childhood. In fact, many schools now embrace which will play an important role in overt mani-
conflict resolution learning schemes to help children to festations of conflict and users may recognise such
acquire attitudes towards others in solving their divergences behaviours as believable.
and problems in a positive and constructive manner. Additionally, note that this problem does not focus on
On the other hand, over the years, there has been several agents’ ability to resolve conflicts. This investigation does
research on developing new teaching tools that motivate not aim in modelling agents’ strategies to act with and

1
towards other agents in order to achieve a certain goal. What plied in school in order to better educate children on conflict
we intend to model is the emotional influence in conflict awareness and proper handling.
emergence and escalation, by affecting agents’ behaviours.
III. R ELATED W ORK
Further, as conflict is quite a broad subject, throughout
this document we will specify which aspects of conflict will Our related work centres in games as learning mecha-
be considered for our research. nisms, serious games. More precisely, we reviewed games
that somehow approach conflict situations. Each game was
II. BACKGROUND - T HEORY OF C ONFLICT reviewed as follows (see Table I):
There is not a reconciled definition for conflict. However, • The Prom – focus on relationship conflicts, where
there are some accepted characteristics that allow us to the user had to manage social relationships by taking
characterise the concept in terms of its structure. We may say actions that will balance the social world. Conflicts
that conflict varies along five dimensions. Participants refer emerged due to characters’ distinct personalities (in-
to individuals that may take part in the conflict situation. terests, needs, traits, social networks and social status).
Causes are linked to adverse or external conditions that Further, the user is directly involved in the conflict, as
initiate a conflict episode, as for example, relationship man- he may cause and resolve it.
agement, self-interest oriented actions or lack of information. • Global Conflict – concerns conflicts of values (the
Based on the initiating action, the actor’s responses (one’s at- Israeli-Palestinian conflict), where players had to frame
titudes, behaviours or strategies) will determine the course of the perspective of both conflicting parties. To do so,
action. Actors’ behaviours determined to bring the situation players need to maintain neutrality with both sides.
to an end are embedded in the conflict resolution dimension, Given the intractable aspect of the conflict portrayed
which following Thomas’ taxonomy [6] the approaches are by this game, the player role is merely to gather infor-
generalised as: accommodation, avoidance, competition, col- mation of both sides, as the player does not interfere
laboration and compromise. The dimensions of assertiveness with the situation.
and cooperativeness are phrased as intentional terms, in • FearNot! – handles conflicts of relations in the context
which assertiveness refers to the extent to which protagonists of bullying, where the user takes responsibility for a
try to achieve their own goal and cooperativeness refers to victim of bullying and has to help him make decisions.
the extent of protagonists trying to satisfy the concerns of Conflicts happened due to the bully’s unreasonable
others. Depending on the approaches towards the conflict, perception and conduct towards the victim. The bully’s
outcomes refer to the consequences that conflict brings actions towards the victim elicit emotions in both of
about. them. The player’s role is to advice the victim character
As this description may suggest, the conflict episode is a on how to handle such situations, thus, this advices in-
temporal sequence of events [7], which may be compared fluence how the victim will behave on the next bullying
to the typical plot of a narrative. This comparison was episode, hence the user is indirectly involved in the
made clear by Laursen and Pursell [7], who time-distributed conflict. These advices may improve or deteriorate the
the conflict episode. The conflict-plot is identified by an victim’s ability to confront the bully. Therefore, we can
initiating action (complication), a rising action (set of actions assume that whenever the victim learns to effectively
that contribute to conflict escalation), a climax and the confront the bully, the conflict de-escalates. In contrary,
outcome. deteriorating advices may escalate the conflict.
When the conflict gets worse, we say that it escalates.
Conflict
When it reaches the turning point (climax) and the mag- Game Type User Involvement Emotional influence Escalation De-escalation
nitude of the situation decreases, we say that de-escalates. The Prom
GC: Palestine
Relation
Values
Yes - Direct
No - Perspective taking
No
No
No
No
No
No
Every conflict escalates, even if it is only in a unilateral FearNot! Relation Yes - Indirect Yes Yes Yes

way. Escalation occurs when one or both parties engage in Table I

the conflict, moving it from a less severe stage to a more C OMPARISON OF CONFLICTS PORTRAYED IN GC: Palestine, The Prom
AND FearNot!.
contentious and heavy state [8]. According to Pruitt [8], an
example of such situation is persistent annoyance, which
may lead to a sequential progression of angry statements,
threats and eventually the conflict escalation to an extreme, IV. C ONFLICT M ODEL
where harassment and abuse occurs. However, what may In agents’ societies where agents have incompatible goals,
make one more prone to escalation and which triggers conflicts are bound to emerge. Yet, such conflicts are usually
promote progression to more aggressive tactics it is not undertaken by the agents’ architectures, which try to find a
clear in the literature [8]. Nevertheless, we may say that way to cooperate with the parties involved. In this work, we
escalation is driven by inner triggers [9], that is, emotions take a different approach by trying to address not so much
that weight one’s current goals and assess the affective value a cooperative-based concept but rather give a step towards
of the situation [10]. a more contingent view of conflict and its resolution.
Due to destructive outcomes in conflict amongst children Emotions are at the heart of social interaction and they
and teenagers, conflict resolution educational has been ap- play a relevant role triggering events such as conflict.

2
Therefore, the cognitive appraisal is an essential element C. Conflict Behaviour Selection Module
for understanding conflict. Given that such situations emerge As described in Section II, behaviours for handling con-
from one’s subjective evaluations of the environment. flict range within two dimensions assertiveness and cooper-
In this Section we describe a conflict handling model ativeness and for simplification reasons we only considered
for agents, where emotions affect their behaviours towards attacking and evading behaviours (from Raider’s AEIOU
the other participants in the conflict situation. Our model model of communication in conflict [14]), which are asso-
stems from FAtiMA’s emotional model for agents [11] and ciated to [high assertiveness, low cooperativeness] and [low
in Tessier’s et al. agents’ conflict handling action model [12]. assertiveness, low cooperativeness], respectively. The values
Our model is illustrated in Figure 1 and further described
of assertiveness and cooperativeness are balanced by the
below. The conflict dynamics specified by the model tries
agent’s emotional state. The reason behind that choice is
to capture the essence of Thomas’ [6] definition of the
based on the assumption that negative emotions are linked
phenomenon, in which conflict is defined as “the process
to less cooperative approaches [10], which will lead to more
which begins when one party perceives that another has
conflicts and their consequent escalation. For example, an
frustrated, or is about to frustrate, some concern of his”.
agent becomes less cooperative as he gets more frustrated
Therefore, first the agent has to perceive that some action
with the situation at hands. In this model, whether an agent
performed by any other agent frustrated (or is going to
is more prone to one of the aforementioned behaviours is
frustrate) its goals, then the event is diagnosed not only
determined by their behavioural predispositions.
according to its congruence to the agent’s goals, but also
other contextual elements (e.g. social relationships). Finally, D. Behavioural Predispositions
behaviours are selected in line with the state of the world
and the agent inward beliefs. To sum up, an agent initially Our current investigation only aims to model simple
observes, then makes assumptions and finally communicates behaviours, in order to demonstrate escalation. We are aware
with others. that several kinds of behaviours can be present in destructive
conflict situations [15]. One’s personality, for instance, can
have impact on that matter. However, we aim to explore the
evolution of a conflict by the means of overt manifestations.
Therefore, we decided to focus on a set of behaviours
that might lead to potential conflicts and their escalation
to explore what believable conflicting behaviour might be
at the eyes of the human perceiver. In this way, we set that
negative emotions will affect negatively the agent’s actions
towards conflict diminishing the possibility for cooperation
Figure 1. Conflict handling model.
[16]. Section V describes the scenario where this model was
applied and then tested. A broader and more detailed set of
The implemented model works as follows. behaviours will be developed in future work.
A. Conflict Recognition Module 1) Attacking Behaviour: Agents with the tendency to
Others’ actions or events that affect (positively or nega- Attack follow a destructive path to cope with the conflict
tively) a certain concern of the agent are perceived. These [14]. These agents are prone to have high assertiveness
are checked by the agent itself to evaluate whether they and low cooperativeness. This falls into the competitive
raise potential conflicts or contribute to the escalation of approach [6] (see conflict handling approaches in Section
the current situation. It is specified the urgency of conflict II), in which the agent only wants to benefit himself. The
that determines how intense the situation is. actions taken by agents with this tendency range from a low
level of aggressiveness to an extreme. For example, as the
B. Conflict Diagnosis Module
agents’ emotional state worsens, their actions may progress
This module diagnosis the conflict oriented event in 3 as follows: lesser insult, criticise negatively, harsh insult, and
steps: threat.
1) analyses the conflict event and generate a conflict de- 2) Evading Behaviour: On the other hand, an agent with
scription – this description consists in what is the cause an Evading tendency may try to avoid conflict situations.
(goal frustrated), participants involved and relationship Initially an agent with these characteristics may want to
between them and the importance of the conflict; cooperate [14], however the build up of negative emotions
2) generates emotional reactions from the conflict analysis as a result of a negative social exchange makes the agent
– an emotion reaction is generated in function of what to become less cooperative. Therefore, the agent will have
was appraised. a tendency to apply Avoiding approaches [6] (see conflict
3) generates emotions from the emotion reactions, where handling approaches in Section II), followed by hostile
the intensity of the emotion reflects the urgency of the evasions and withdraw of his interests. This pattern of action
conflict [13], and integrates into the emotional state characterises the Evading behaviour [14]. Furthermore, as
(this process is undertaken by FAtiMA [11]); the emotional state gets worse the actions performed by an

3
agent with this kind of behaviour progresses as follows: A. Agents’ Minds
ignore the situation, sacrifice own’s goals to avoid further Our conflict model, described previously, was imple-
involvement and, finally, leave the scene. mented in FAtiMA emotional agents’ architecture [11] (see
V. T HE D REAM T HEATRE G AME Figure 3). No significant changes were made to FAtiMA’s
The Dream Theatre1 is an educational game that aims architecture, as we took advantage of the reactive appraisal
at teaching children, aged 9 to 11, some conflict resolu- process and the deliberative planning process to implement
tion skills. The game setting is a theatre company and our conflict model in the agents.
the user/child needs to select the adequate cast for each
performance. Every week the child is challenged with a new
“production” and needs to select adequately his/her cast, and
assign adequately the roles to the actors. The user has a
set of possible virtual actors/characters to assign the roles,
and needs to do so, trying to optimise the final performance
results. The only problem is that the agents may not accept
the assignment made by the learner, and conflicts may occur
between the agents. The role of the child is to manage the
conflict, advice the agents, and try to do so in a manner that
the conflict is resolved, otherwise, if the emotional states
of the agents at the end is very negative, the performance
results will not be good.
Figure 3. Agents’ implementation in FAtiMA’s architecture (diagram from
[17]), where the circle notations represent where each conflict model’s step
was implemented.

Therefore, the model was implemented as follows.

1) Conflict recognition and diagnosis: On integrating our
conflict model in FAtiMA agents’ minds architecture, we
found plausible to simplify some of the model’s aspects. The
conflict recognition was integrated into conflict diagnosis
due to the fact that we used FAtiMA’s deliberative layer
to implement these steps from our conflict model.
To begin with, the CR is defined by a set of active-pursuit
goals in FAtiMA’s deliberative layer. In FAtiMA, active-
pursuit goals are goals that the agents actively try to achieve
[11]. The agents plan on achieving such goals whenever they
are activated, i.e., when the goals’ preconditions are satisfied.
In the conflict recognition’s active-pursuit goals definition,
Figure 2. Dream Theater’s screenshot, showing two characters having a the pre-conditions must define the situation in which a
discussion about a role. certain action is directed to a certain interest of the agent.
Agents in the cast have a set of characteristics, in particu- Furthermore, the goal’s success conditions must specify the
lar they have preferences for roles (for example some agents “internal action”2 which will cause the conflict diagnosis
may prefer to be a “Hero”, whereas others may prefer to be step.
a “Villain”). As the player grants roles to characters, conflict The conflict diagnosis step is supported by FAtiMA’s
situations emerge when characters perceive an obstruction to deliberative planner, which will plan to perform the “internal
achieve their desired roles. They also have social relations action” (FAtiMA’s deliberative planner process’ description
(friendship relations, allowing for in-group and out-group can be found in [17]). For this, it is specified a set of actions
situations), and proficiency. All the agents are modelled that determine the analysis of the event, where each action
according to an agent architecture, where emotions are presents the conflict description.
appraised by the situations created by the user, and the With the conflict description, an emotional reaction must
agents reacting emotionally to the situations. be generated, according to how the agent agent believes that
In the current version of our demonstrator, the player the event will affect his interest. By taking advantage of
does not have specific strategies on approaching the conflict FAtiMA’s reactive appraisal [11], this step of the conflict
situation, other than granting roles, dismissing from roles or diagnosis is defined by a set of emotional reactions directed
firing characters from the theatre company. to the conflict descriptions from the “internal actions”.
1 The assets of the game scenario were developed by Serious Games 2 We use the term “internal action” to specify an agent’s action which is
Interactive - www.seriousgames.dk not visible to others, only to the one who performs it.

4
 After the emotional reaction is triggered, FAtiMA’s ap- The values for the mood ranges were determined by empir-
praisal process generates the emotions [17], following ical tests with the FAtiMA architecture.
OCC’s theory of emotions [18]. This process consists in
Emotionally-driven behaviour Mood(m) Description
two major steps. The first generates potential emotions and Do not mind m≥0 Agent performs a rather “neu-
tral” behaviour towards the
the second will send these emotions to the emotional state. situation.
2) Conflict behaviour selection: In a nutshell, the conflict Ask why −3.0 < m ≤ 0 If approached by another
agent, this agent will try to ask
behaviour selection step is defined as a set of active- the reason for such behaviour.
Do not respond −5.0 < m ≤ −3.0 Agent does not fight back ag-
pursuit goals towards the conflict situation, where emotions gressive moves towards him-
influence which goals the agent will be capable of activating. self (e.g., deny a critic or a
insult towards himself).
As we stated previously, in our model’s description (Sec- Give up −8.0 < m ≤ −5.0 Agent gives up desired inter-
est, in order to avoid any more
tion IV), we aim at modelling behaviour which is affected by involvement in the conflict.
the agents’ emotional states. With such, for this version of Leave scene m ≤ −8.0 Agent final withdrawal, in
which he leaves the scene.
the agent’s minds implementation, we use agents’ moods as
the key emotional affect to influence behaviours. However, Table III
Evading AGENT ’ S BEHAVIOURS , ACCORDING TO MOOD ’ S VALUES .
this assumption can be quite arguable, as we could model
conflict behaviours related to specific emotions, such as
anger or fear [10][19]. However, as mood is a pervasive
affect that defines a general disposition state of an individual, Similarly to the Aggressive agent, this agent’s emotional
we believe that this emotional aspect is more suitable to reactions were authored for this agent to find undesirable
influence characters’ behaviours towards conflict. aggressive attacks from others towards himself.
Further, we decided to implement two behavioural pre- B. Overall System
dispositions (as presented in Section IV): Attacking and
Even though we presented the agents’ minds architecture
Evading; as we believe these are more likely to generate
and the demonstrator, we still do not know how these two
escalation as a result of what those behaviours bring to the
modules communicate with each other. Thus, in this Section
social interaction.
we will describe the overall system. As we can see in Figure
The Attacking behaviour has a set of aggressive be- 4, three main modules compose the overall system: agents’
haviours towards the situation of conflict. In Table II, we minds, world simulation and realisation engine. We already
present the Attacking agents’ emotionally-driven behaviours presented the first and the last modules in the previous
to approach a conflict situation. Each one represents a group Sections. However, we still need to describe the world
of active-pursuit goals, which can be activated if agents’ simulation, which integrates these two modules.
emotional states are in those mood ranges.

Emotionally-driven behaviour Mood(m) Description

Do not mind m≥0 Agent performs a rather “neu-
tral” behaviour towards the
situation.
Insult other −2 < m < 0 Agent uses insults towards
other agents.
Criticize negatively other −4.0 < m ≤ −2.0 Agent uses destructive critics
towards other agents.
Threaten other m ≤ −4.0 Agent threatens the other to
withdraw from the threatening Figure 4. Dream Theatre demonstrator’s overall system is composed by
agents’ concern.
three main modules: the agents’ minds, in FAtiMA’s agents architecture,
Table II the world Simulator, in ION Framework, and the demonstrator, developed
Attacking AGENTS ’ SET OF BEHAVIOURS , ACCORDING TO MOOD ’ S in Unity3D.
VALUES .
1) World Simulation: The world simulation module, as
we stated earlier, functions as the integration layer between
It is also important to explain that this agent’s emotional the demonstrator and the agents minds. On the other hand,
reactions are manipulated, so that the agent finds very unde- the world simulator, provided by the ION Framework [20],
sirable the others’ attempts to approach the conflict. Hence, also aims at providing means to simulate dynamic environ-
the more the conflict situation prolongs, more aggressive ments.
the agent will be, as others’ actions will only deteriorate his To achieve this, the ION framework [20] provides the
emotional state. means to simulate dynamic environments. To do so, it
In contrast, the Evading behaviour agent has a set of identifies four basic elements: Entities, Properties, Actions,
less aggressive behaviours towards a situation of conflict. and Events. Entities populate the simulation universe. They
As we referred, in Section IV, Evading agents may initially can have Properties and change the world through the use
have constructive approaches towards the conflict situation, of their Actions. As the ION Framework model is based on
however, if the emotional state deteriorates, this agent will the Observer pattern, Entities register to Events. In which,
begin to follow evasive approaches. Following that, Table III these Events are raised whenever any change to the world
points out Evading agents’ emotionally-driven behaviours. occurs. For example, if a Property is changed, an Action’s

5
state changes (such as, when it starts or stops), or an Entity and identifies it as an event that interferes positively to his
enters or leaves the simulation. interests, thus generating positive emotions. However, when
2) Integrating Agent’s Minds: FAtiMA agents’ minds Andy appraises this event, he will identify it as an event that
are linked to their respective graphical representations, in interferes negatively to his self-interests and thus generate a
Unity3D, through their representations in the ION simulation negative emotional state. Further, this is aggravated by the
(see Figure 5). For such, besides the elements we mentioned fact that Andy considers the role highly important to him and
previously in the ION simulation, agents’ minds need an Bob is not his friend. So, with the emotions fired, Andy will
additional element in the simulation: the Remote Mind. get upset enough to approach Bob aggressively, by verbally
insulting him. This process then escalated, and Bob starts
feeling upset, but, given that he was given his preferred role,
he limits himself to only question the reason of the insult
trying to resolve this situation. However, Andy appraises
the situation again negatively, disapproving Bob’s approach,
thus getting even more upset. These emotions triggered will
lead Andy to plan even more aggressive behaviours towards
Bob. As the situation gets even more intense, Bob eventually
also reaches a high anger situation, which will lead Bob to
plan a way internally how to cope with such pressure. In the
end, if not dealt with, Bob ends up giving up the Hero role,
thus making the whole cast selection by the child a failure.

VI. E VALUATION
This research aims at developing a prototype conflict
Figure 5. Communication details between: Unity3D character’s script model to be integrated into an adaptive serious game for
components and ION simulation elements; and between those elements and
a FAtiMA agent. The solid arrows define communication through functions, teaching resolution skills to children. The model is based
while the dashed arrows signify communication through socket messages. on an emotional system (FAtiMA [11]) and we claim that
this emotional process is essential to convey concepts as
A script in Unity3D launches the FAtiMA agent’s mind conflict emergence and escalation. Therefore, we conducted
process and connects it with the correspondent representa- an between groups experiment to assess our hypothesis. For
tion, in the ION simulation. For that, the FAtiMaMind script that, we intended to compare our conflict model, wish we
creates the Remote Mind element in the ION simulation, will emotionally-driven or full-model (FM) condition, with a
which registers to all the world events, such as, when control version where the emotional process was deactivated
an Entity enters or leaves the simulation, an Action being on agents, we will name it non-emotional or simplified
performed, or a Property value changed. With that, the model (SM) condition.
Remote Mind reports all these Events to the FAtiMA agent
mind (see Figure 5). A. Measures
In order to communicate with the FAtiMA agent’s mind, With this final evaluation we wanted to measure: (1)
the Remote Mind establishes a socket communication with whether the participants had the perception of a conflict
it. Therefore, all the registered events that occur in the situation and the actors’ behaviours towards it; (2) if the
simulation are transformed into FAtiMA events and then sent users perceived the influence of emotions on the charac-
to the agent’s mind. ters’ behaviours towards the conflict and its influence on
On the other hand, the actions that the agent will perform the conflict escalation; and (3) participants’ perceptions of
(by action tendencies or deliberated actions) are sent to the the conflict end-state, if characters’ emotionally oriented
RemoteMind (see Figure 5). These actions are parsed and behaviours were reasonable and if there is a chance of
generated into Action start requests to the simulation. renewable conflict.

C. Illustrative Example B. Evaluation Procedure and Method

In order to illustrate how Dream Theater works, let’s A total of 80 participants (19 females, 61 males aged 14-
analyse a small situation. 48) took part in the study, which was available through an
In the first week of playing, the user is faced with a online questionnaire. The scenario for evaluation was the
play that only needs two characters. He has Andy and Dream Theater and the illustrative example on Section V-C
Bob to chose from, but both share the desire for the same was the base of our study.
role, the “Hero” role. However, the player can only choose The participants had to watch a video of a user interacting
one of them to have it. Bob has higher proficiency than with (Dream Theater). Similar to the example presented in
Andy, thus he is the most suitable choice for this role. Section V-C, the user gives the role to Bob, who has a
Yet, when Bob receives the role, he appraises this event higher proficiency level and seems to be the natural choice

6
for the given situation, instead of Andy. From this action, a behaviour and presents the differences between the two
sequence of interactions occurs between Andy and Bob. conditions (FM and SM).
After watching the video, participants had to answer a set
Descriptive Statistics
of questions, regarding what they saw on the video. The set Question
1st Set Andy’s
FM (N = 40) SM (N = 40) Mann-Whitney
of questions were adapted from a self-serving questionnaire Behaviour Measure Mdn[Quartiles] Mdn[Quartiles]
U = 298.500
on conflict behaviour and escalation [21]. Hence, our ques- (a) hostile
Q5 1[1, 2] 3[2, 3] p < 0.001
or friendly r = −0.565
tionnaire is divided in three main set of questions. In these U = 455.500
(b) competitive
sets, a 5-point likert scale was provided to allow participants Q6
or collaborative
1[1, 2] 2[2, 3] p < 0.001
r = −0.390
classify each sentence. Note that participants were blind to (c) frustrating
U = 647.000
the hypotheses. Q7 2[1, 3] 2[2, 3] ns
or stimulating
1) 1st Set – Characters’ behaviours measurement: First, (d) evil-minded
U = 365.500
Q8 1[1, 2.5] 3[2, 3] p < 0.001
participants were asked to rate the characters’ behaviours (or good-hearted −0.488
through 5-point likert scales. More precisely, they had to Table IV
indicate to what extent each character’s behaviours was [21]: M ANN -W HITNEY STATISTICS FOR PARTICIPANTS ’ PERCEPTION OF
(a) hostile or friendly (1 - very hostile to 5 - very friendly); A NDY ’ S BEHAVIOUR WITHIN THE TWO CONDITIONS (FM AND SM).
(b) competitive or collaborative (1 - very competitive to 5 -
very collaborative), (c) frustrating or stimulating (1 - very
frustrating to 5 - very stimulating), and (d) evil-minded or Taking a look at the results, from Q5, users’ consid-
good-hearted (1 - very evil-minded to 5 - very good-hearted). ered Andy’s behaviour in the FM condition significantly
2) 2nd Set – Escalation measurement: After that, a (p < 0.001) more hostile (M dn = 1.5), in comparison with
second set of questions asked participants to rate the process the result from the SM condition (M dn = 3). The same
of escalation of the interaction. The questions regarded if happened in Q7 and Q8, where his behaviour was found
participants acknowledged escalation in terms of [21]: (a) significantly p < 0.001 more competitive (M dn = 1) and
to what extent did parties obstruct each other, (b) to what evil-minded (M dn = 1) in the FM condition, comparing
extent did frustration increase, (c) to what extent did the to the SM condition (M dn = 2 and M dn = 3). However,
atmosphere worsen; and de-escalation: (d) to what extent Q7 there were no significant differences in the participants’
did parties work out an ideal solution, (e) to what extent perception of Andy’s frustrating behaviour between the FM
did parties come close, and (f) to what extent were ideas condition and SM conditions.
explored. Each question was rated from 1 - not at all to 5 - Despite that, the obtained results are consistent with its
very much. internal drives to follow a destructive path in a conflict inter-
3) 3rd Set – End-state measurement: Finally, the third action, modelled in the FM condition. In the SM condition,
set of questions probed participants about the end-state of Andy’s neutral attitude contributed for results distributed in
the situation. Due to the presented scenario, we maintained the middle of the scale.
only one of the questions presented in the self-serving On the other hand, Table V specifies the results gath-
questionnaire [21]: if there is a chance of a renewed conflict ered for participants’ perception of Bob’s behaviour and it
between the characters. Further, for exploratory reason we presents the difference between the FM and SM conditions.
also assessed participants’ perception of the reasonableness
Descriptive Statistics
of characters’ behaviours and if participants believe that Question
2nd Set - Bob’s
FM (N = 40) SM (N = 40) Mann-Whitney
Behaviour Measure Mdn[Quartiles] Mdn[Quartiles]
such behaviours need improvements. These measures were U = 375.500
(a) hostile
rated by participants through 5-point likert scales, from −2 Q9 4[3, 4] 3[2, 3] p < 0.001
or friendly r = −0.470
- totally disagree to 2 - totally agree. U = 287, 000
(b) competitive
Q10 4[3, 4] 2[1, 2] p < 0.001
C. Results or collaborative r = −0.566
U = 793.500
(c) frustrating
The data was analysed using the Mann-Whitney test to Q11
or stimulating
3[2, 3] 3[2, 3] ns

compare the difference between the two conditions (agents (d) evil-minded
U = 428.000
Q12 4[3, 4.5] 3[2, 3] p < 0.001
with the model’s FM version and agents with the model’s (or good-hearted r = −0.421
SM version). Each measure reported the following results.
Table V
Does the user have a perception of a conflict interac- M ANN -W HITNEY STATISTICS FOR PARTICIPANTS ’ PERCEPTION OF
tion? B OB ’ S BEHAVIOUR WITHIN THE TWO CONDITIONS (FM AND SM).
As we explained previously, this Set of questions rates
the agents’ behaviours towards the situation. To begin with,
questions Q5 to Q8 correspond to the (a) to (d) variables Bob’s attitude was considered to be more constructive.
of the conflict behaviour measures (p. 7) directed to Andy’s In Q9, Q10 and Q12 of the FM condition, selected values
behaviour. After that, questions Q5 to Q8 relates to the same go towards the other extreme of the scale (M dn = 4,
variables addressed to Bob’s behaviour. M dn = 4, M dn = 4), compared to Andy’s score. Further,
First, we address Andy’s behaviour. Table IV summarises Bob was considered significantly (p < 0.001) more friendly
the results obtained for the participants rating of Andy’s than the SM condition (M dn = 3, M dn = 2, M dn = 3).

7
Despite the significant (p < 0.001) differences between were due to participants’ interpretation of Andy’s gleeful
the two conditions, the non-emotional condition (SM) has smile, as this emotional expression may have conveyed to
provoked more responses towards the destructive side of participants that he achieved his goals.
the spectrum. We believe that such responses were due On the other hand, in Q16, the significant (p < 0.001)
to participants’ subjective view on Bob’s neutral posture difference between Bob’s increase of frustration in the FM
towards the conflict situation. On the SM condition, in (M dn = 4) and the SM (M dn = 1) conditions was as
contrast to FM, Bob did not give in to Andy’s actions. predicted. Since Bob opted out of performing the role in the
Therefore, participants might have perceived this as being FM condition, due to reaching the limits of his emotional
competitive, as we can see in Q10 (M dn = 2) of the control, participants could easily perceive an increase in
SM condition, where participants have reported a more his level of frustration. In contrast, Bob’s neutral posture
competitive behaviour in Bob compared to the FM condition throughout the SM condition demonstrated no signs of
(M dn = 4). frustration increase.
Nevertheless, the results are consistent with Bob’s initially Finally, as we forecasted, in Q17, participants’ ratings of
cooperative behaviour modelled in the FM condition. the atmosphere worsening were significantly (p < 0.001)
Does the user recognise that the conflict escalates? different between the FM (M dn = 5) and the SM (M dn =
This Set of questions assesses the escalation (questions 4) conditions. This result is consistent with our process of
Q13 to Q17) and the de-escalation (questions Q18 to Q21) conflict escalation which emerges from the emotional states
processes of the conflict situation. First, questions Q13 and deterioration.
Q14 are adapted to the context of the scenario to determine Further, considered Q18 to Q21, which assessed conflict
the (a) variable from the conflict escalation measurement de-escalation, no significant differences where found be-
(p. VI-B2), where Q13 regards to Andy frustrating Bob and tween test conditions, given that no de-escalation processes
Q14 regards Bob frustrating Andy. Similarly, Q15 and Q16 were portrayed in both conditions. In both conditions, Q18
are also adapted in order to assess the (b) variable, regarding to Q21 had the same ratings (M dn = 1).
Andy and Bob’s increase of frustration, respectively. Finally, Does the user recognise a chance for future conflict
questions Q17 to Q21 relate to the variables (c) to (f). and characters’ behaviour reasonableness?
Starting with the escalation process measurement, Table The final Set of questions (Q22 to Q26) presented in our
VI contains the results and it compares condition FM with questionnaire aimed at measuring participants’ perception
condition SM. of the conflict end-state. The first two questions from this
Set (Q22 and Q23) assessed participants’ belief of the
Descriptive Statistics
Question
3rd Set
FM (N = 40) SM (N = 40) Mann-Whitney reasonableness of characters’ behaviours. After that (Q24
Escalation Measure Mdn[Quartiles] Mdn[Quartiles]
U = 109.500 and Q25), participants stated if they believe that characters’
(a) To what extent did
Q13 Andy obstruct Bob 5[4, 5] 1[1, 2]
p < 0.001
r = −0.766
should improve their behaviours. Finally, it was asked for
(a) To what extent did
U = 284.000 participants to rate their belief on the chance of renewed
p < 0.001
Bob obstruct Andy
Q14 2[1, 2] 4[2.5, 5] r = −0.572 conflict between characters (Q26).
U = 623.000
(b) To what extent did Andy
become more frustrated
ns For these measures, Table VII summarises the results of
Q15 3[2, 4.5] 4[3, 5]
U = 232.000
participants’ ratings.
(b) To what extent did Bob
p < 0.001
Q16 become more frustrated 4[3, 5] 1[1, 2.5] r = −0.627 Descriptive Statistics
U = 368.500 Question 4th Set - End-state Measure FM (N = 40) SM (N = 40) Mann-Whitney
(c) To what extent did
p < 0.001 Mdn[Quartiles] Mdn[Quartiles]
Q17 the atmosphere worsen 5[4, 5] 4[3, 4.5] r = −0.503 U = 297.000
Andy’s behaviour
Q22 −2[−2, −2] 0[−1, 1] p < 0.001
was reasonable
Table VI r = −0.585
U = 662.500
M ANN -W HITNEY STATISTICS FOR PARTICIPANTS ’ PERCEPTION OF Bob’s behaviour
ns
Q23 was reasonable 0[0, 1] 0[−1, 1]
CONFLICT ESCALATION WITHIN THE TWO CONDITIONS (FM AND SM).
U = 389.000
Andy needs to improve his
Q24 2[2, 2] 1[0.5, 2] p < 0.001
behaviour towards this situations
r = −0.505
U = 737.000
Bob needs to improve his
Q25 1[−0.5, 2] 1[0, 2] ns
As expected, the results obtained from this set of questions behaviour towards this situations
U = 724.500
were quite expressive. First, in question Q13, participants Q26
There is a chance for Andy
and Bob to discuss again 2[1, 2] 1.5[1, 2] ns
significantly (p < 0.001) reported that Andy frustrated Bob
more in the FM condition (M dn = 5), compared to the SM Table VII
M ANN -W HITNEY STATISTICS FOR PARTICIPANTS ’ FINAL REMARKS ON
condition (M dn = 1). In contrast, in question Q14, Bob was THE CONFLICT SITUATION WITHIN THE TWO CONDITIONS (FM AND
significantly (p < 0.001) rated as less frustrating in the FM SM).
condition (M dn = 2), then in the SM condition (M dn = 4).
Next, in Q15, no significant difference was detected
between conditions regarding Andy’s increase of frustration. To begin with, in Q22, the results on participants rating
This was quite expected given that in both conditions, Andy of the reasonableness of Andy’s behaviour was significantly
could not achieve the role. However, in the FM condition, it (p < 0.001) different between FM (M dn = −2) and
was not expected so many ratings in the middle of the scale SM (M dn = 0) conditions. The result of this measure
(M dn[Quartiles] = 3[2, 4.5]). We believe that these ratings in the FM condition (M dn = −2) demonstrated that

8
our emotionally-driven conflict model was able to portray been used for raising awareness [4], teaching about subjects
behaviours destructive enough for participants to rate them [2], how to deal with bullying [3], and others.
as not being reasonable, compared to “neutral” behaviour Motivated by this, the SIREN project intends to create a
represented in the SM condition (M dn = 0). series of conflict resolution games to teach children on how
On the other hand, Bob’s ratings of reasonable behaviour to deal with conflict situations [25]. Our contribution for this
were not as conclusive as Andy’s ratings. No significant project concerns the creation of a conflict resolution game
difference was found in Q23 between the FM and SM that is populated with autonomous virtual characters.
conditions (M dn = 0 and M dn = 0). We believe that such In order to achieve our hypothesis, we investigated theory
results may be due to the duality of Bob’s behaviours in of conflict in social psychology. From this we demonstrated
the FM condition. Given that in the FM condition, initially, special regard to Thomas’ [6] definition of conflict as “the
Bob’s behaviours towards Andy’s increasingly aggressive process which begins when one party perceives that another
attacks might be rather constructive, as the conflict escalates has frustrated, or is about to frustrate, some concern of his”
to an extreme, Bob starts to lose some emotional control and and Castelfranchi’s [26] definition of “full social conflict”,
copes with this by following a more evasive behaviour (see which exists when “there is the subjective awareness of the
p. 3). Therefore, we assume that this duality lead participants competitive situation”.
to rate Andy’s reasonable behaviour in the middle of the Further, we reviewed educational games for raising aware-
scale, in the FM condition. Further, in the SM condition ness on general populate [4], teaching children about a
the ratings in the middle of the scale are due to Bob’s subject [2] and games that involve conflict in the matter
“neutrality” throughout the situation, in the SM conditions. of: social interactions [27], intractable conflicts [28], and
Moreover, in Q24, participants pointed that Andy’s be- bullying [3]. However, from the reviewed games, only
haviour in the FM condition needed improvement. A signif- FearNot! [3] assessed emotion in the conflict situations,
icant (p < 0.001) difference was found between this result in which is supported by FAtiMA architecture of emotional
the FM condition (M dn = 2) and SM condition (M dn = 1). agents [11].
This result is coherent with the result of Andy’s reasonable With the conflict theory review, we implemented an
behaviour rating. agents’ model that comprises emotional affect in the conflict
Regarding Bob, in Q25, the results were not conclusive handling process.
concerning participants’ view on behavioural improvement. To demonstrate our model, we implemented a conflict res-
There was no significant difference between FM (M dn = olution game’s demonstrator, Dream Theatre. This demon-
1) and SM (M dn = 1) conditions. Similar to our beliefs strator was populated with synthetic characters modelled
for Bob’s rating of reasonable behaviour (Q23), we posit by our conflict model in an emotional agents’ architecture,
that the result in the FM condition is due to the duality FAtiMA [11].
in Bob’s behaviours. Therefore, the result for Q24 in the In order to prove our hypothesis, we performed a between
FM condition had a wide discrepancy (M dn[Quartiles] = groups experiment with 80 participants. Each participant was
1[−0.5, 2]). randomly assigned to one of two test conditions. One condi-
Finally, in Q26, participants’ ratings on the possibility of tion supported our conflict model, where agents’ behaviours
a renewed conflict situation did not provide a significant are emotionally-driven, we called it emotionally driven or
difference between the FM (M dn = 2) and the SM (M dn = full-model (FM) condition. Another condition, which served
1.5) conditions. This result was quite unexpected. Taking a as a control condition to test our hypothesis, imbued agents
look at each condition, even though the result for the FM with a variation of our model where the emotional process
condition was as expected, the SM condition also had the was deactivated, we called it non-emotionally driven or
same score. Hence, we have to discard the argument that simplified-model (SM) condition.
participants rated that there was a high probable chance of With the experiment’s results, we compared both con-
renewed conflict due to the conflict escalation portrayed in ditions to assess if our model’s condition (FM) issued
the FM condition. better perceptions of the conflict situation, compared to the
VII. C ONCLUSIONS simplified-model (SM) condition. From the result’s analysis,
we confirmed that in the FM condition, participants had
As we argued in the beginning of this document, conflict a better perception of the characters’ behaviours towards
is natural to human life, as it is present in our everyday conflict, and their roles in the situation. Further, in the FM
basis [22]. However, conflict is mostly seen as a negative condition, the characters’ escalating behaviours influenced
act that needs to be eradicated. Contrarily, researchers have by their emotional state conveyed the clear perception of
proven that embracing conflict in a constructive way serves conflict escalation. These results where consist with your
as a tool for constructive personal growth [22], especially hypothesis.
in young ages [23]. For that, conflict resolution education
has been applied in schools in order to teach students about R EFERENCES
conflict and how to handle it constructively [24]. [1] C. Morgado and I. Oliveira, “Mediação em contexto escolar:
Concerning education, new teaching tools have been re- transformar o conflito em oportunidade,” Exedra, vol. Volume
searched, such as, serious games. This kind of games have 1, pp. 43–56, June 2009.

9
 [2] J. P. Rowe, B. W. Mott, S. W. McQuiggan, J. L. Robinson, [16] C. Bell and F. Song, “Emotions in the conflict process: an
S. Lee, and J. C. Lester, “Crystal island: A narrative-centered application to the cognitive appraisal model of emotions
learning environment for eighth grade microbiology,” 2009. to conflict management,” International Journal of Conflict
Management, vol. 16, no. 1, pp. 30 – 54, 2005.
[3] R. Aylett, M. Vala, P. Sequeira, and A. Paiva, “Fearnot! –
an emergent narrative approach to virtual dramas for anti- [17] J. Dias, I. Leite, M. Kriegel, M. Y. Lim, C. Martinho, S. Mas-
bullying education,” in Virtual Storytelling. Using Virtual carenhas, A. Paiva, A. Pereira, R. Prada, and A. Wichert,
Reality Technologies for Storytelling, ser. Lecture Notes “Architecture development with action-selection mechanism,”
in Computer Science, M. Cavazza and S. Donikian, Eds. 2010, unpublished Deliverable 5.3.
Springer Berlin / Heidelberg, 2007, vol. 4871, pp. 202–205.
[18] A. Ortony, G. L. Clore, and A. Collins, The Cognitive
[4] G. Rebolledo-Mendez, K. Avramides, S. de Freitas, and Structure of Emotions. Cambridge University Press, July
K. Memarzia, “Societal impact of a serious game on raising 1988.
public awareness: the case of floodsim,” in Sandbox ’09
Proceedings of the 2009 ACM SIGGRAPH Symposium on [19] A. L. Miller and S. L. Olson, “Emotional expressiveness
Video Games. ACM New York, NY, USA, 2009, pp. 15–22. during peer conflicts: A predictor of social maladjustment
[5] G. Yannakakis, J. Togelius, R. Khaled, A. Jhala, K. Kar- among high-risk preschoolers,” Journal of Abnormal Child
pouzis, A. Paiva, and A. Vasalou, “Siren: Towards adaptive Psychology, vol. 28, pp. 339–352, 2000.
serious games for teaching conflict resolution,” in 4th Eu-
ropean Conference on Games Based Learning (ECGBL10), [20] M. Vala, G. Raimundo, P. Sequeira, P. Cuba, R. Prada,
Copenhagen, Denmark, 21-22 October 2010, 2010. C. Martinho, and A. Paiva, “Ion framework – a simulation
environment for worlds with virtual agents,” in 9th Interna-
[6] K. W. Thomas, “Conflict and conflict management: Re- tional Conference on Intelligent Virtual Agents - IVA2009,
flections and update,” Journal of Organizational Behavior, ser. Lecture Notes in Computer Science, vol. 5773 / 2009.
vol. 13, pp. 265–274, 1992. Amsterdam, The Netherlands: Springer Berlin - Heidelberg,
September 2009, pp. 418 – 424.
[7] B. Laursen and C. A. Hafen, “Future directions in the study
of close relationships: Conflict is bad (except when it’s not),” [21] C. K. W. De Dreu, A. Nauta, and E. Van de Vliert, “Self-
Social Development, vol. 19, no. 4, pp. 858–872, November serving evaluations of conflict behavior and escalation of the
2010. dispute,” Journal of Applied Social Psychology, vol. 25, pp.
2049—-2066, 1995.
[8] D. G. Pruitt, The Handbook of Conflict Resolution: Theory
and Practice, 2nd ed. Jossey-Bass, 2006, ch. Some Research [22] M. J. M. C. Luı́s, “A assembleia de turma como dispositivo
Frontiers in the Study of Conflict Resolution, pp. 849–880. de mediação de conflitos,” Master’s thesis, Universidade de
[9] L. Kriesberg, Conflicts: From Escalation to Resolution, Lisboa, Faculdade de Psicologia e de Ciências da Educação,
3rd ed. Rowman & Littlefield Publishers Inc, 1998, ch. 2009.
Analyzing Social Conflicts, pp. 1–26.
[23] W. J. Kreidler, Teaching Conflict resolution through children’s
[10] E. G. Lindner, The Handbook of Conflict Resolution: Theory literature (Grades K-2). Scholastic Inc., 1994.
and Practice, 2nd ed. Jossey-Bass, 2006, ch. Emotion and
Conflict, pp. 268–311. [24] D. W. Johnson and R. T. Johnson, “Conflict resolution
and peer mediation programs in elementary and secondary
[11] J. Dias and A. Paiva, “Feeling and reasoning: A computational schools: A review of the research,” Review of Educational
model for emotional characters,” in Progress in Artificial In- Research, vol. 66, no. 4, pp. 459–506, Winter 1996.
telligence, ser. Lecture Notes in Computer Science, C. Bento,
A. Cardoso, and G. Dias, Eds. Springer Berlin / Heidelberg, [25] Y.-G. Cheong, R. Khaled, C. Grappiolo, J. Campos, C. Mart-
2005, vol. 3808, pp. 127–140. inho, G. P. D. Ingram, A. Paiva, and G. Yannakakis, “A com-
putational approach towards conflict resolution for serious
[12] C. Tessier, H.-J. Müller, H. Fiorino, and L. Chaudron, “Agents games,” in Proceedings of the Sixth International Conference
conflicts: New issues,” in Conflicting Agents, ser. Multiagent on the Foundations of Digital Games, ser. FDG ’11. ACM,
Systems, Artificial Societies, and Simulated Organizations, 2011.
C. Tessier, L. Chaudron, H.-J. Müller, and G. Weiss, Eds.
Springer US, 2002, vol. 1, pp. 1–30. [26] C. Castelfranchi, Conflict Ontology. Springer-Verlag New
[13] M. Maiese, “Emotions,” in Beyond Intractability, G. Burgess York, Inc., 2000, pp. 21–40.
and H. Burgess, Eds. University of Colorado,
Boulder: Conflict Information Consortium, July 2005, [27] J. McCoy, M. Treanor, B. Samuel, B. Tearse, M. Mateas,
http://www.beyondintractability.org/bi-essay/emotion Link and N. Wardrip-Fruin, “Comme il faut 2: A fully realized
active in May 2012. model for socially-oriented gameplay,” in Proceedings of the
Intelligent Narrative Technologies III Workshop. ACM New
[14] E. Raider, S. Coleman, and J. Gerson, “Teaching conflict York, NY, USA, 2010.
resolution skills in a workshop,” in Handbook of Conflict
Resolution, 2nd ed. Jossey-Bass, 2006, ch. 31. [28] T. Buch and S. Egenfeldt-Nielsen, “The learning effect of
global conflicts: Palestine,” Media@Terra conference, 2006.
[15] H. Xie, D. J. Swift, B. D. Cairns, and R. B. Cairns, “Ag-
gressive behaviors in social interaction and developmental
adaptation: A narrative analysis of interpersonal conflicts
during early adolescence,” Social Development, vol. 11, no. 2,
pp. 205–224, 2002.

10