An Architecture for Autonomous Normative BDI Agents based on

Personality Traits to Solve Normative Conflicts

Paulo Henrique Cardoso Alves, Marx Leles Viana and Carlos José Pereira de Lucena

Laboratory of Software Engineering (LES), Department of Computer Science,

Pontifical Catholic University of Rio de Janeiro – PUC – Rio, Rio de Janeiro, RJ, Brazil

Keywords: Solving Normative Conflicts, Normative Agents, Multiagent Systems, Personality Traits.

Abstract: Norms are promising mechanisms of social control to ensure a desirable social order in open multiagent

systems. Normative multiagent systems offer the ability to integrate social and individual factors to provide

increased levels of fidelity with respect to modelling social phenomena such as cooperation; coordination;

decision-making process, and organization in artificial agent systems. However, norms eventually can be

conflicting — for example, when there is a norm that prohibits an agent to perform a particular action and

another norm that obligates the same agent to perform the same action, the agent is not able to fulfill both

norms at the same time. The agent’s decision about which norms to fulfill can be defined based on rewards,

punishments and agent’s goals. Sometimes, the analysis between these attributes will not be enough to allow

the agent to make the best decision. This paper introduces an architecture that considers the agent’s personality

traits in order to improve the normative conflict solving process. In addition, the agent can execute different

behaviors with equal environment variables, just by changing its own internal characteristics. The

applicability and validation of our approach are demonstrated by an experiment that reinforces the importance

of the society’s norms.

1 INTRODUCTION

Multiagent Systems (MASs) are societies in which

these heterogeneous and individually designed

entities (agents) work to accomplish common or

independent goals (Viana et al., 2016). In order to

deal with autonomy and diversity of interests among

the different members, such systems provide a set of

norms, which are mechanisms used to restrict the

behavior of agents by defining what actions to which

the agents are: (i) obligated (agents must accomplish

a specific outcome); (ii) permitted (agents can act in

a particular way) or (iii) prohibited (agents must not

act in a specific way) all to encourage the fulfillment

of the norm through rewards definition and

discouragement of norm violation by pointing out the

punishments (Figueiredo et al, 2010).

Norms must be complied with by a set of agents

and include normative goals that must be satisfied by

the addressees. In addition, norms are not always

applicable, and their activation depends on the

environment in which agents are situated. In some

cases, norms suggest the existence of a set of

sanctions to be imposed when agents fulfill, or

violate, the normative goal.

The decision-making process about which norms

will be fulfilled or violated might be defined based on

the agent’s goals, rewards and punishment analysis

(Viana et al., 2016). Since an agent’s priority is the

satisfaction of its own goals, before complying with

the norms the agent must evaluate their positive and

negative effects on its goals (Lo

pez and Ma

rquez,

2004) without hurting the agent’s autonomy. Both

rewards and punishments are the means by which the

agents know what might happen independently of the

agent’s decision to comply, or not, with the norms.

However, norms sometimes may conflict or be

inconsistent with one another (Mccrae and John,

1992). For instance, different norms can, at the same

time, prohibit and obligate a state that the agent wants

to fulfill and the simple balance between goals,

rewards and punishments might not be enough to

permit the agent to make the best decision.

The abstract normative agent architecture

developed by (Lo

pez and Ma

rquez, 2004), has four

main steps: (i) agent perception, i.e., when the agent’s

beliefs and a set of norms are updated; (ii) norm

Alves, P., Viana, M. and Lucena, C.

An Architecture for Autonomous Normative BDI Agents based on Personality Traits to Solve Normative Conﬂicts.

DOI: 10.5220/0006599300800090

In Proceedings of the 10th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2018) - Volume 1, pages 80-90

ISBN: 978-989-758-275-2

adoption, i.e., when agents verify which norms are

addressed to them; (iii) norm deliberation, i.e., when

agents verify which norms they intend to fulfill, or

violate, and (iv) norm compliance, i.e., when agents

verify which norms they will comply with. Within the

norm deliberation step, conflicting norms are verified

and a set of these norms is added to the norm

compliance set.

We changed the internal process of the norm

deliberation step to deal with conflicting norms by

adding the agent’s personality traits. These

characteristics will help the software agents make

some different decisions involving personality traits

based on the OCEAN model (Mccrae and John, 1992),

setting a weight for each one of these characteristics. We

will present an experiment comparing different

approaches to deal with normative conflicts based on

social profiles and personality traits. This will illustrate

the new deliberation process proposed in this paper.

Within this context, we present an approach that

builds BDI agents with personality traits (Barbosa et al.,

2015) to improve the decision-making process for the

solution of normative conflicts. This approach aims at

offering new resources for the agent to deal with

conflicting norms supported by personality traits. As

such, more human characteristics can be considered

in order to improve the deliberation process. By using

these new functions, it is possible to build agents that:

(i) use personality traits to improve the solution

between normative conflicts, and (ii) evaluate the

effects on its desires with respect to the fulfilment, or

violation, of a norm and thus use all of these functions

to conduct experiments to learn how different

strategies could change an agent’s behavior.

The paper is structured as follows: Section 2

focuses on the background, while Section 3 discusses

related work. Section 4 presents the BDI-agent

approach to personality traits to solve normative

conflicts. Section 5 presents the experiment that

evaluates our approach. Finally, Section 6 shows our

conclusion and future work.

2 BACKGROUND

This section describes the main concepts related to

agents and multiagent systems. First, we will discuss

norms and BDI (Belief-Desire-Intention)

architecture. We will also discuss the relation

between normative conflicts.

2.1 Norms

Norms are designed to regulate the behavior of the

agent, and therefore, a norm definition should include

the address of the agent being regulated (Bordini et

al., 2007). However, norms are different from laws,

and they cannot force agents to comply with them.

Agents are autonomous entities, so norms can only

suggest and present the expected behavior to which

the agent will decide to comply with, or not.

In this work, we used the norm representation

described in (Viana et al., 2015) . Norms properties

are briefly described in Table 1. For example, the

property Addressee is used to specify the agents or

roles responsible for fulfilling the norm.

Table 1: Norm Description.

Property

Description

Addressee

It is the agent or role responsible

for fulfilling the norm

Activation

It is the condition for the norm to

become active

Expiration

It is the condition for the norm to

become inactive

Rewards

It represents the set of rewards to

be given to the agent to fulfill a

norm

Punishments

It is the set of punishments to be

given to the agent for violating a

norm

Deontic

Concept

It indicates if the norm states an

obligation, a permission or a

prohibition

State

It describes the set of states being

regulated

In order to better understand the application of

norms to regulate agents with a different social

profile, we made a comparison between the social

contribution and the individual satisfaction of the

agent for fulfilling, or violating, the norms for each

approach. Furthermore, to better understand the

definition of norms and their representation, imagine

a user scenario where the employee agent has to

decide the transportation type to go home. The

agent’s goal is to increase physical conditioning and

has the following options to go home: (i) by bicycle,

which is a way to satisfy the agent’s goal, and (ii) by

bus, if it is raining, in which case, the agent cannot

accomplish its goal at this time.

In addition, each employee agent should decide

according to specific norms. Eventually, a norm is

sent to each employee agent with the following

statement: “go home by bus, it is raining”. This norm

has the following attributes: (i) addressees are

employee agents; (ii) the required deontic concept is

prohibition, because it prohibits the agent to go home

by bicycle, and (iii) when an agent agrees to a norm,

An Architecture for Autonomous Normative BDI Agents based on Personality Traits to Solve Normative Conﬂicts

it will receive a reward. In this case, the reward may

be not getting the flu. If the employee agent violates

the norm, the agent will receive the punishment

associated with the norm. For example, when it is

raining and the employee agent really wants to work

out, it then will violate the norm by going home by

bicycle, which will result in the decrease of the

agent’s health, because the agent will probably come

down with the flu. In this case, a punishment

associated with the norm will be applied to the agent,

i.e., the agent cannot work the next day because it is

sick. Note that the norm is activated when it is raining.

In turn, the norm expires when the weather is sunny.

2.2 Conflicting Norms

Norms eventually may conflict, i.e., an action may be

simultaneously prohibited and permitted, or it may be

inconsistent, i.e., when an action is simultaneously

prohibited and obligated (Vasconcelos et al., 2007).

These conflicts and inconsistencies may be caused by a

norm that prohibits an agent to perform a particular

action while another norm obligates the same agent to

perform the same action at the same time. The agent can

realize any action in the environment until an active

norm restricts its goals. For example, Figure 1 presents

a scenario of conflicting norms — when a norm

defines that the buyer agent cannot bring back the

product bought and at the same time another norm

defines that the buyer agent can return the product

bought before opening it.

Figure 1: Conflict - Prohibition and Permission.

Figure 2 presents another scenario of conflicting

norms — the seller agent can only re-price the

products before the store opens and another norm

permits the seller agent to re-price them when the

store is open and there is a sale.

Figure 2: Conflict - Permission and Obligation.

In short, conflicts may occur in different cases and

situations, and dealing with them is extremely

important to make the best decision.

2.3 BDI Architecture

The BDI (Belief-Desire-Intention) model was

proposed by (Bratman, 1987) as a philosophical

theory of practical reasoning, representing, the

information, the motivational and deliberative states

of the agent, respectively. There are two main steps:

(i) applying a filter to make a set of goals to which the

agent must commit to serve as the basis of its beliefs,

and (ii) finding a way to understand how the desires

produced can be fulfilled based on the agent’s

available resources (Wooldridge and Ciancarini,

1999).

Figure 3: Generic BDI architecture (Wooldridge et al.,

1999).

Figure 3 shows the BDI model, which is

composed of three mental states: (i) beliefs, which

represent the environment factors that are updated

after each perceived action — they represent the

agent’s world knowledge; (ii) desires, which have

information about the goals to be fulfilled — they

represent the agent’s motivational state, and (iii)

intentions, which represent the action plan chosen.

The BDI architecture starts with a Belief Revision

Function that makes a new belief set based on the agent’s

perception. Next, the Option Generation Function

sets the agent’s available options and desires, based

on its own environment beliefs and intentions. The

next function is a Filter that sets the agent’s intentions

based on its own beliefs, desires and intentions.

Finally, the Action Selection Function sets the actions

to be executed based on the current intentions.

Most BDI systems are inspired by the Rao and

Georgeff (Rao and Georgeff, 1995) model. The

authors presented an abstract BDI interpreter. This

interpreter works with beliefs, goals and agent plans.

As such, the goals are a set of concrete desires that

may be evaluated all together, avoiding a complex

goal deliberation step. The interpreter’s main

functionality is the means to the end process,

achieved by plan selection and plan execution given

a goal or event.

ICAART 2018 - 10th International Conference on Agents and Artiﬁcial Intelligence

2.4 Personality Traits

The big-5 model (Mccrae and John, 2011), also

known as OCEAN model, provides a mechanism to

define personality traits based on such concepts and

defines five factors: (i) Openness, describing a

dimension of personality that portrays the

imaginative, creative aspect of the human character,

(ii) Conscientiousness, determining how much an

individual is organized and careful, (iii) Extroversion,

related to how outgoing and sociable a person is, (iv)

Agreeableness, which is about friendliness,

generosity and the tendency to get along with other

people, and (v) Neuroticism, referring to emotional

instability and the tendency to experience negative

emotions.

Each factor is composed of many traits, which

basically are used to describe people (Mccrae and

John, 2011) (Goldberg, 1990). The factors presented

will be used to help the agent’s decision-making

process and plan selection, according to the agent’s

individual goals and intended norms.

Based on the OCEAN model, the personality traits

may be built through the distribution of weights

between the factors: (i) Openness to experience; (ii)

Conscientiousness; (iii) Extroversion; (iv)

Agreeableness, and (v) Neuroticism. In Figure 4,

agent 1 may be creative and adventurous, while agent

2 may be careful.

Figure 4: OCEAN model application example.

3 RELATED WORK

This section describes some related work: (i) the

solution for normative conflicts (López, 2003),

(Criado et al., 2010), (Neto et al., 2011); (ii)

architecture designs considering the agent’s

emotional state (Pereira et al., 2005), and (iii) the

agent’s personality (Barbosa et al., 2015), (Jones et

al., 2009).

Pereira et al. (Pereira et al., 2005) proposed an

architecture based on the BDI (Belief-Desire-

Intention) model to support artificial emotions,

including internal representations of the agent’s

capabilities and resources. This research introduces

subjects, such as artificial emotions, agent means and

BDI architecture. Furthermore, a common-sense

definition of new mental states, such as emotions, was

developed, and influenced the BDI architecture

through the common-sense understanding of the way

they positively affect human reasoning. The authors

defined a new concept: Fear, an informational data

structure that reports situations which an agent should

avoid. This work presents the Personality Traits in

BDI-Agent architecture as an extended version of the

classic BDI. However, the authors do not compare the

results with other approaches that may, or may not,

apply emotions and neither provide support to solve

normative conflicts (Pereira et al., 2005).

The authors in (Barbosa et al., 2015) built a

decision process to work as part of the story-telling

systems wherein narrative plots emerge from the

acting characters’ behaviors and personality traits.

The process evaluates goals and plans to examine the

plan commitment issue. The drives, attitudes and

emotions play a major role in the process. However,

the personality traits were not applied to MASs,

which creates an opportunity to improve the agent’s

decision-making process to deal with normative

conflicts.

Jones et al. (Jones et al., 2009) developed a BDI

extension to consider physiology, emotions, and

personality. It is used to model crisis situations such

as terrorist attacks, for instance. The emotions were

used in pairs such as fear/ hope, anger/gratitude and

shame/pride. The physiology may be affected by the

simulation environment and may change the agent’s

health. The following characteristics were

considered: stress, hunger/thirst, temperature, fatigue,

injuries and contamination. The personality is a set of

characteristics that determines that agents are

psychologically, mentally and ethically different

from each other. However, this approach was not

applied in Normative Multiagent Systems to evaluate

different behaviors that may emerge with personality

traits applications.

Some approaches (López, 2003), (Criado et al.,

2010), (Neto et al., 2011) have been proposed in the

literature to develop agents that evaluate the effects

of solving normative conflicts. For instance, the

n-BDI architecture defined by Criado et al. (Criado et

An Architecture for Autonomous Normative BDI Agents based on Personality Traits to Solve Normative Conﬂicts

al., 2010) presents a model for building environments

governed by norms. Basically, the architecture selects

objectives to be performed based on the priority

associated with each objective. An objective’s

priority is determined by the priority of the norms

governing a specific objective. However, it is not

clear in this approach how the properties of a norm

can be evaluated. In addition, the approach neither

supports a strategy nor considers the agent’s

personality traits to deal with conflicts between

norms.

Lopez et al. (López, 2003) defined a set of

strategies that can be adopted by agents to deal with

norms, as follows: Pressured, Rebellious and Social.

For example, the Pressured strategy occurs when

agents fulfill the norms to achieve their individual

goals considering only the punishments that will harm

them. Another is the Rebellious strategy, in which

agents consider only their individual goals and violate

all the environment’s norms. Finally, the Social

strategy happens when agents first of all comply with

norms and after verify if it is possible to fulfill some

individual goals. Although this work provides some

mechanisms for the agents to collect norms, the

authors do not provide a framework that can be

extended to create simulations of normative

multiagent systems by including new strategies. In

addition, this work can neither extend mechanisms to

collect information during the simulations nor extend

mechanisms to generate norms and agent goals.

Furthermore, the agent cannot detect and overcome

normative conflicts.

Finally, Santos Neto et al. (Neto et al., 2011)

propose the NBDI architecture, based on the Criado

et al. (Criado et al., 2010) research, to develop

goal-oriented normative agents whose priority is the

accomplishment of their own desires while evaluating

the pros and cons associated with the fulfillment or

violation of the norms. To make this possible, the BDI

architecture was extended by including norms-related

functions to check incoming perceptions, and select

norms based on the agent’s desires and intentions. A

detection conflict and a solving conflict algorithm

were developed based on norms contributions; in the

case of conflicts between norms, the one with the

highest contribution to the achievement of the agent’s

desires and intentions can be selected. If the norm

contributions have equal values, then the first norm

will be selected. Therefore, as it is possible to

observe, sometimes the norm contribution is not

enough for the agent to make a better decision. We

identified this gap and improved the decision-making

process, adding the personality traits concept.

As none of this related work deals with norm

conflicts using personality traits, this was the gap that

we based on to propose our work. We aim at

providing a better way to balance goals, rewards,

punishment and personality traits to solve normative

conflicts and improve the deliberation process. To

evaluate the norm contribution, we first use rewards

and punishment values. With these values, we then

continue to evaluate the norm contribution, now

adding personality traits.

4 PERSONALITY TRAITS IN

BDI-AGENTS

This section describes the main concepts required to

understand the approach based on BDI agents with

personality traits. This architecture improves the

solution of normative conflicts and, after helping the

deliberation process, it deals with non-conflicting

norms and agent goals. In addition, we provide a

software framework overview and discuss its

different components.

4.1 The Architecture

The Personality Traits in BDI agents approach that

can solve the normative conflicts were inspired on the

concepts presented in the background and the related

work sections.

Figure 5: The architecture.

We added both BDI features and personality traits

in the conflicts resolution and normative deliberation

process. The architecture foundation was based on the

abstract normative agent architecture developed in

(Lo

pez and Ma

rquez, 2004). Figure 5 presents our

BDI agent with personality traits architecture to solve

normative conflicts.

The most significant change was adding to the

deliberation process a reasoning step that involves the

BDI architecture and the personality traits approach.

Both strategies work in a complementary way to

ICAART 2018 - 10th International Conference on Agents and Artiﬁcial Intelligence

change the agent’s behavior, considering factors that

were not used in the norms deliberation process in

previous work. All of these changes refer only to the

internal agent process. The decision-making process

proposed has four steps, which are described below.

The first step involves the agent’s perception in

the Belief Revision Function, where the agent

perceives the active norms in the environment

addressed to it by means of its sensors. Then, the

agent inserts into the Norms set the norms that it

wants to fulfill by using the Norms Adoption

function. After that, the agent updates its beliefs,

taking into account these new norms.

The second step is the Desire Normative

Generator, which is composed of three processes: (i)

Norm Status Evaluation function, where the agent

verifies which norms are activated or deactivated; (ii)

Norms Conflict Detection function, where the agent

verifies what the normative conflicts are, and (iii)

Solution Normative Conflicts function, where the

agent evaluates the norms contribution and solves the

normative conflicts, also considering its personality

traits based on the OCEAN model. Table 2 shows

some examples of personality traits composition that

we consider: drives, attitudes and emotions, as in

(Mccrae and John, 1992). Our personality traits

model has only two properties: (i) a name and (ii) a

value indicating its weight.

Table 2: Personality Traits Examples.

Drives

Attitudes

Emotions

Sense of duty

Careful

Anger

Material gain

Adaptable

Fear

Spiritual

endeavor

Self-controlled

Surprise

The norms analyses are based on the normative

contribution, which is composed by the evaluation

between rewards, punishments, goals and personality

traits. Figure 6 shows the normative contribution

equation.

Figure 6: Normative Contribution Equation.

The normative contribution concept was extended

from (Neto, 2011). We added the goals and

personality traits weights. The m bound refers to the

summation of the activated norms addressed to the

agent; the p bound refers to the summation of the

agent’s goals, and q refers to the summation of the

agent’s personality traits. Table 3 describes the goal

properties.

Table 3: Goal Properties.

Property

Description

Name

It is the name of the goal

Value

It is the value that represents the

importance of this goal

Norm Required

It is the set of the norms required

to permit that the goal be

accomplished

Personality

Trait

It is the set of the personality traits

required to permit that the goal be

accomplished

Belief Required

It is the set of the beliefs required

to permit that the goal be

accomplished

The personality traits are used only in two

situations: (i) at Solution Normative Conflicts

through the equation shown in Figure 6 and (ii) at the

plan selection step. Table 4 describes the plan

properties. A set of non-conflicting norms is exported

to the next step. The goals that are not restricted by

the norms are the agent’s Desires.

Table 4: Plan Properties.

Property

Description

Name

It is the name of the goal

Value

It is the value that represents the

importance of this plan

Personality

Trait

It is the set of the personality traits

that contribute to this plan

execution

Goal Required

It is the set of the goals required to

permit that the plan be activated

The third step is the Normative Filter, which is

composed of two processes taking into account the

agent’s personality traits: (i) Norms Evaluation

function, where the agent evaluates the Desires set

and it decides which norms will be fulfilled, and (ii)

Plan Selection function, where the agent will choose

its best plans in the Intentions set.

Finally, the fourth step is the Action Selection

function, which is composed of the Normative

executor and selector. This function receives the

Norms set, which are the norms that the agent intends

to fulfill. Last but not least, all of these steps help to

improve the normative conflict solving process,

considering personality traits inserts into the BDI

reasoning process.

An Architecture for Autonomous Normative BDI Agents based on Personality Traits to Solve Normative Conﬂicts

4.2 The Framework

Inspired by the JSAN architecture (Viana et al.,

2015), which uses different normative strategies to

deal with norms and takes into account the different

agent’s social levels, as in (López, 2003), we built a

new approach by introducing personality traits

aiming to improve the solution of the normative

conflict. Our framework provides the

decision-making process described in Section 4.1.

Figure 7 shows the framework architecture.

The Normative BDI Agent class is composed of

goals, role, norms, beliefs, desires, intentions, and

personality traits. By using these attributes, the agent

starts the decision-making process to solve normative

conflicts. In the normative conflict solving process,

the agent will choose the norms that it will add to the

Intentions set and finally will decide which norms

will be fulfilled according to the agent’s social

profile, as in (Bordini et al., 2007) and (López, 2003).

The solving process of normative conflicts starts

with the calculation of each norm’s normative

contribution, wherein the agent evaluates its rewards

and punishments and compares each normative

contribution with other norms addressed to it.

Furthermore, we added a new step to improve this

process, also taking into consideration the agent’s

goals and its personality traits. This new step consists

of the choice of the normative goals that can be

fulfilled according to the agent’s goals and its

personality traits.

Figure 7: The Framework architecture.

The agent will verify which goal can be fulfilled

based on its personality traits, so the agent uses its set

of goals and analyzes each conflicting norm, adding

to the normative contribution an integer value to

represent the compatibility between the agent’s goals

and the normative goals. The compatibility is defined

by the evaluation of which of the agent’s goals can be

achieved if a norm is fulfilled. As a result, some

conflicting norms may have changed their normative

contribution based on the use of the agent’s

personality traits. For instance, imagine one norm that

obligates an agent to cross a damaged bridge. If the

agent is careful (careful meaning the agent's

personality trait) its normative contribution will be

decreased because the agent does not have the intent

to cross a damaged bridge — it is dangerous.

5 EXPERIMENT

Our initial experimentation includes different kinds

of agents to deal with norms, such as described in

(Lo

pez and Ma

rquez, 2004) and (Neto et al., 2011).

The (Lo

pez and Ma

rquez, 2004) approaches deal with

norms considering the following strategies: (i) Social,

i.e., the agent fulfills all of the active norms addressed

to it and then it verifies which goals can be fulfilled;

if there are conflicts, it randomly selects one norm

from each conflicting norms set to be complied with,

(ii) Rebellious, i.e., the agent violates all norms and

fulfills all goals, and in this case it does not matter if

there are conflicting norms; the agent will never

fulfill any norms, and (iii) Pressured, i.e., the agent

only fulfills the norms whose normative punishment

value is bigger than the value of the importance of the

goals; thus the agent feels pressured to comply with

the norm to avoid punishments. In (Neto et al., 2011)

the authors present the NBDI approach, which

considers the normative contribution generated by

evaluation between: (i) the norms’ rewards and

punishments, and (ii) the importance of the goals.

We chose these examples to compare with our

approach because they represent the most common

strategies followed by agents when they face a norm

compliance decision. Our approach is based on (Neto

et al., 2011) and was improved by adding personality

traits. A normative conflict is identified when

different norms are active and have opposite deontic

concepts. The norm contribution is then evaluated for

each one of the conflicting norms and there are a few

steps to follow: (i) for each goal, its importance is

increased by a weight given to a personality trait, (ii)

for each goal allowed by a norm (the norm does not

restrict this goal), the norm contribution is increased,

adding the importance of the goal, and (iii) for each

norm that is active at the same time and has opposite

deontic concepts, the norm with the better norm

contribution value is selected.

For the non-conflicting norms (i) a set of norms

indexed by the goals that the norm restricts is created,

ICAART 2018 - 10th International Conference on Agents and Artiﬁcial Intelligence

(ii) for each non-conflicting norm, the norm contribution

is increased adding the norm contribution value to

each norm in this set that restricts the same goal, (iii)

the norms contribution and goals increased by

personality traits are evaluated, and (iv) the better

value is selected and this norm or goal is selected to

be fulfilled. Our interest here is the observation of

how both the social contribution and the agent’s

individual satisfaction change, according to the norm

compliance strategy it chooses, the increase in the

number of conflicts between the norms it has to play

with and its personal goals. The social contribution of an

agent is defined by the number of times the agent has

fulfilled the norms addressed to it. The agent’s

individual satisfaction is the number of goals

achieved in relation to the number of goals generated.

We reproduced the experiment created in (Lo

pez

and Ma

rquez, 2004) using all of these different

approaches and comparing them with our approach.

First, a base of goals to represent all the goals that an

agent might have is randomly created. Second, a

motivation value is associated to each goal in this set to

represent their importance. In addition, each goal might

have a personality trait associated, meaning that if

there is an agent that has this personality trait, this

goal will be increased by the personality trait value.

Both punishments and rewards in each norm are also

randomly generated, as well as the deontic concept

and activation time. Thus, the norms are evaluated by

agents following different strategies so that similar

inputs produce different outcomes.

We observed both the social contribution and the

agent’s individual satisfaction taking into account the

different percentages of normative conflicts over a

period of time. First, no conflicts were considered,

meaning that all norms and goals could be fulfilled.

Then the experiment was repeated, with the number of

conflicts increased in a proportion of 25% until all norms

conflicted among themselves. Each experiment

consisted of 100 runs, and in each run, 10 goals and

10 norms were used.

Table 5 and Table 6 show the properties of

random norms and random goals used in this

experiment, respectively.

Table 5: Random Norm Properties.

Property

Description

Addressee

Agent “X”

Activation

All norms are activated

Expiration

There is no expiration

Rewards

Random value in the range [0,5]

Punishments

Random value in the range [0,5] +

Set of goals restricted by this norm

Deontic

Concept

Random value in the range [-1,1],

where -1 represents a prohibition, 0

represents a permission and 1

represents an obligation

Table 6: Random Goal Properties.

Property

Description

Name

Goal + random value in the range

[0,9]

Value

Random value in the range [0,5]

Norm Required

Random set of norms

Personality

Trait

Two personality traits with a

random value [5,10]

Belief Required

No belief was required

First, the Pressured strategy shows that the agent

achieves more individual goals rather than

contributes to the society. Figure 8 shows the agent’s

behavior in different conflicting norm situations.

Figure 8: Pressured Strategy.

The Social strategy shows that initially, with

no-conflicting norms, the agent fulfills all norms

because, first, the agent complies with all the adopted

active norms and then decides which goals will be

achieved. Figure 9 shows that, as a result, the number

of goals achieved increases gradually.

Figure 9: Social Strategy.

Agents using the Rebellious strategy violate all

norms and, as no goal are restricted, all of them will

200

400

600

800

1000

0 2 5 5 0 7 5 1 0 0

GOALSANDNORMS

CONFLICTSPERCENTAGE

PRES S URED

IndividualSatisfaction SocialCon tributio n

200

400

600

800

1000

0 2 5 5 0 7 5 1 0 0

GOALSANDNORMS

CONFLICTSPERCENTAGE

SOCIAL

IndividualSatisfaction SocialContribution

An Architecture for Autonomous Normative BDI Agents based on Personality Traits to Solve Normative Conﬂicts

be achieved. Figure 10 shows this behavior. It is

important to notice that the rewards and the

punishment values are not taken into account. In this

situation, the agent always will receive a punishment

for violating norms that restrict its goals.

Figure 10: Rebellious Strategy.

The agent using the NBDI strategy considers the

value of the social contribution to fulfill, or violate,

each norm before deciding to comply with it, or not.

Figure 11 shows that more goals are achieved when

the normative conflicts increase.

Figure 11: NBDI Strategy.

The Personality traits strategy considers the norm

contribution developed in NBDI adding the

personality traits value. The experiment results are

similar to the NBDI strategy, although the agent

meets more individual goals. Figure 12 shows the

agent’s behavior regarding norms compliance and

goals achievement.

Figure 12: Personality Traits Strategy.

As can be observed, the personality traits strategy

encourages the agent to fulfill its goals and, if there is

a personality trait with a null value, the performance

will be the same as presented by NBDI. The greater

the weight of the personality traits, the higher the

number of individual goals.

Figure 13 shows the comparison between all of

the five strategies. As a result, the Personality Traits

strategy achieved more goals than the Social,

Pressured and NBDI strategies. It shows that the

Personality Traits strategy helps the agent to fulfill

more individual goals and increases the individual

satisfaction.

Figure 13: Individual satisfaction overview.

Figure 14 shows all of the five different strategies,

comparing the social contribution between them. As

can be observed, strategies that achieve more goals

comply with fewer norms; therefore, the Personality

Traits strategy fulfills fewer norms than other

strategies, except the Rebellious strategy, which

always violates all the norms. Thus, the developed

approach is a middle ground between Rebellious

strategy and NBDI strategy.

200

400

600

800

1000

0 2 5 5 0 7 5 1 0 0

GOALSANDNORM S

CONFLICTSPERCENTAGE

REBELLIOUS

IndividualSatisfaction SocialContribution

200

400

600

800

1000

0 2 5 50 75 10 0

GOALSANDNORMS

CON FLICTSPERCENTAGE

NBDI

IndividualSatisfaction SocialContribution

200

400

600

800

1000

0 25 5 0 75 1 0 0

GOALSANDNORMS

CONFLICTSPERCENTAGE

PERS O N ALITYTRAITS

IndividualSatisfaction SocialContribution

500

550

600

650

700

750

800

850

900

950

1000

0 2 5 5 0 7 5 1 0 0

GOALS

CONFLICTSPERCENTAGE

IN DIVIDUALSATIS FACTIO N

NBDI Pressured PersonalityTraits Rebellious Social

ICAART 2018 - 10th International Conference on Agents and Artiﬁcial Intelligence

Figure 14: Social Contribution overview.

6 CONCLUSION AND FUTURE

WORK

This paper proposes an approach to deal with

normative conflicts by adding personality traits

characteristics to the BDI architecture to improve the

decision-making process that will decide which

norms the agent shall fulfill. The main contributions

of this research are: (i) include personality traits in the

BDI architecture to change the solving process of

normative conflicts; (ii) implement different agent

behaviors according to different personality traits,

and (iii) make it possible to build software agents with

different behaviors. The BDI-agent with personality

traits was able to reason about the norms it would like

to fulfill, and to select the plans that met the agent’s

intention of fulfilling, or violating, such norms.

Moreover, the experiment developed showed that the

Personality Traits strategy results were similar to the

NBDI strategy, although the agent with personality

traits chooses to achieve more goals than with the

other strategies.

As future work, we are deciding on an

experimental study in order to apply fuzzy logic to

deal with changes found in the real world, such as the

chance to become sick if you stay in the rain.

Furthermore, the punishment for becoming ill is also

variable. An agent's punishment may range from

sneezing to pneumonia. The severity of the illness

could be a factor for the agent's current health state

and how fast the recovery takes place may also be part

of the agent's personality profile. So, when the agent

must decide whether to ride the bike in the rain, it

must calculate the reward (fitness gained) against the

possibility of becoming sick (may or may not get

sick) and the consequences (punishment) that could

range from very mild (sneezing) to very serious

(pneumonia). We also plan to implement this

approach in other more complex scenarios that take

personality traits into account. For example: (i) in risk

areas, where firefighters are responsible for planning

people’s evacuation, and (ii) in crime prevention,

where the police are responsible for arresting

criminals and keeping civilians safe. Last but not

least, we will apply these different strategies to

environments that have more agents, in order to

analyze their behavior and evaluate the norms

addressed to the agent, and the agent’s internal goals.

REFERENCES

Barbosa, S. D. J., Silva, F. A. G. Da, Furtado, A. L.,

Casanova M. A., 2015. Plot Generation with Character-

Based Decisions, In Comput. Entertain. 12, 3, Article

Bordini, R. H.; Hübner, J. F.; Wooldridge, M., 2007.

Programming Multi-Agent Systems in AgentSpeak

using Jason. Wiley-Blackwell. 1

edition.

Bratman, M., 1987. Intention, plans, and practical reason,

Harvard University Press.

Criado, N., Argente, E., Noriega, P., and Botti, V., 2010.

Towards a Normative BDI Architecture for Norm

Compliance, In COIN@ MALLOW2010, p. 65-81.

Jones, H., Saunier, J., Lourdeaux, D., 2009. Personality,

Emotions and Physiology in a BDI Agent Architecture:

The PEP -> BDI Model, In IEEE, p. 263-266.

Lopez, F. L., 2003. Social Power and Norms, In Diss.

University of Southampton.

pez, L. F., Márquez, A. A., 2004. An Architeture for

Autonomous Normative Agents, IEEE, Puebla,

México.

Mccrae, R., John, O., 1992. An introduction to the five-

factor model and its applications, In Journal of

Personality, vol. 60, pp. 175–215.

Neto, B. F. S., Silva, V. T., Lucena, C. J. P., 2011.

Developing Goal-Oriented Normative Agents: The

NBDI Architecture, In International Conference on

Agents and Artificial Intelligence.

Pereira, D., Oliveira, E., Moreira, N., Sarmento, L., 2005.

Towards an Architecture for Emotional BDI Agents, In

Portuguese conference on artificial intelligence.

Rao, A. S., Georgeff, M. P., 1995. BDI agents: From theory

to practice, ICMAS. Vol. 95.

Vasconcelos, W.W., Kollingbaum, M.J., Norman, T. J.,

2007. Resolving conflict and inconsistency in norm-

regulated virtual organizations, In Proceedings of the

6th international joint conference on Autonomous

agents and multiagent systems (AAMAS '07). ACM,

New York, NY, USA, Article 91 , 8 pages.

Viana, M. L., Alencar, P., Everton Guimarães, Cunha, F. J.

P., 2015. JSAN: A Framework to Implement Normative

Agents, In The 27th International Conference on

Software Engineering & Knowledge Engineering, p.

660-665.

100

200

300

400

500

600

700

800

900

1000

0 2 5 5 0 7 5 1 0 0

NORMS

CONFLICTSPERCENTAGE

SOCIALCONTRIBUTION

NBDI Pressured PersonalityTraits Rebellious Social

An Architecture for Autonomous Normative BDI Agents based on Personality Traits to Solve Normative Conﬂicts

Viana, M. L., Alencar, P., Lucena, C. J. P., 2016. A

Modeling Language for Adaptive Normative Agents. In

EUMAS, 2016, Valencia. European Conference on

Multi-Agent Systems.

Wooldridge, M. Ciancarini, P., 2001. Agent-Oriented

Software Engineering: The State of the Art, In LNCS

1957.

Wooldridge, M., Jennings, N. R., Kinny, D., 1999. A

methodology for agent-oriented analysis and design, In

Proceedings of the Third International Conference on

Autonomous Agents Agents 99.

ICAART 2018 - 10th International Conference on Agents and Artiﬁcial Intelligence