ATTENTION, MOTIVATION AND EMOTION IN COGNITIVE

OFTWARE AGENTS

Daniela C. Terra

Depto Ciˆencias Exatas, Instituto Federal de Educac¸˜ao, Ciˆencia e Tecnologia de Minas Gerais

Faz. Varginha, LMG 827 - Km 05, Bambu´ı, Brazil

nrique E. Borges, Paulo E. M. Almeida

Laborat´orio de Sistemas Inteligentes, Centro Federal de Educac¸˜ao Tecnol´ogica de Minas Gerais

Avenida Amazonas 7675, Belo Horizonte, Brazil

ywords:

Emotional-based agents, Bio-inspired systems, Cognitive agents, Situated cognition, Software agents

architecture.

Abstract:

The observations that the emotional phenomenon is an essencial component to the living beings cognition

has inﬂuenced the conception of artiﬁcial intelligent mechanisms. This inﬂuence has lead to the discussion

whether it is possible to elaborate an inteligent system without including into it the emotions’s role. The

proposed model implements an affective mechanism inside an architecture to build cognitive software agents,

called ARTIFICE. Its conception was inspired in a biological bottom-up approach for classifying affections

considering ideas and neuroscientiﬁc concepts about emotions and their inﬂuence on learning. Also it is consi-

dered that the attention, motivation and emotion are interdependent aspects that need to be considered together

by an affective mechanism. The main objective of this work is to reproduce the adaptive functions of survival

value in software. Moreover this study also aims to presents how the autonomy in artiﬁcial organisms can be

acquired inserting appropriate synthetic emotions. The experiments suggest that the model is appropriate to

allow agents to adapt themselves in generic environments, according to their incorporated affective structures.

Their learning was accomplished solely from live interaction experiences with environment and other existing

entities. No previous information about the artiﬁcial world were built-in into these agents.

1 INTRODUCTION

There were many frustrated attempts by engineers to

apply artiﬁcial intelligence (AI) following that tra-

ditional and non biological approachs (Terra et al.,

2004). According to (Varela et al., 2003), the ﬁeld of

cognitive sciences has ﬁnally yielded to openings in

alternative approaches to cognition. AI mechanisms

have been built without including environmental rep-

resentations (Clancey, 1997). Some agree that emo-

tions perform an essential role to the living systems,

so emotions need to be abstracted to constitute a bio-

inspired model (Edelman, 1987).

The affective mechanism proposed here is part of

an architecture of building cognitive situated software

agents (CSSA), named ARTIFICE (Santos, 2003).

CSSA are agents biologically inspired, built without

representations of the external environment, with

sensory-motor capacities to interact and adapt to the

unpredictability involved in their ontogeny. The AR-

TIFICE model is based in a view of ”intelligence”

under the perspective of situated cognition (Maturana

and Varela, 2002; Varela et al., 2003; Edelman, 1987;

Bateson, 1972).

The affective mechanism model was conceived to

integrate the nervous system (NS) of a CSSA. It seeks

to reproduce in the agents some basic adaptive capa-

cities assigned to emotions. The model considers that

emotion, motivation and attention are all intrinsic as-

pects of the affective process. It is due to this in-

terdependent perspectives the nomitation ”emotional-

motivational-attentional” (AME) mechanism.

Some concepts, ideas and theories that based the

AME mechanism conception are cited in the follow-

ing section. The AME model is explained in sec-

tion 3, section 4 describes the environment and the

CSSA instantiated for the computational experiments

and the conclusions follows.

426

C. Terra D., E. Borges H. and E. M. Almeida P..

ATTENTION, MOTIVATION AND EMOTION IN COGNITIVE SOFTWARE AGENTS.

DOI: 10.5220/0003179304260429

In Proceedings of the 3rd International Conference on Agents and Artiﬁcial Intelligence (ICAART-2011), pages 426-429

ISBN: 978-989-8425-41-6

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

2 BIOLOGICAL INSPIRATION

This study tried to recontextualize some concepts and

ideas from a biological affects typology (Buck, 1999)

and from other theories and researches in neuro-

science (Edelman, 1987; LeDoux, 2003; Varela et al.,

2003) for a software model.

2.1 The Motivational-emotional System

The (Buck, 1999) typology of biological affects de-

ﬁnes motivation as a potential to the behavior due the

dynamic induced by the cerebral neurochemical sys-

tems. Emotion is seen as the expression of this moti-

vational potential when the individual faces a challen-

ging stimulus. Motivation and emotion are like two

sides of the coin, aspects of a motivational-emotional

phenomenon. This is a bottom-up approach where

emotions are seen as based in biological systems,

structured by evolution. (Buck, 1999) named primary

motivational-emotional system (primes) the affective

systems wich signal to the organism the perception of

its internal environment for auto-regulation.

The primes can be hierarchically classiﬁed accor-

ding to an increasing degree of interaction with the

general proposed systems. This hierarchy classiﬁes:

a)reﬂexes which involve completely inﬂexible reac-

tions (e.g., the patellar reﬂex); b)instincts like ﬁxed

patterns of innate behavior (e.g., the migratory beha-

vior of birds); c)drives of primary needs that re-

quire speciﬁc consumatory learning: hunger, thirst,

pain, etc.; d)primary affects that motivate the auto-

regulation of fear, rage, etc. The higher the hierarchy

of primes more inﬂuenced by learning.

2.2 Learning, Forgetting and Attention

Another essential concept of the (Buck, 1999) typo-

logy is the expectancy system. The operation of this

non-speciﬁc affective system is connected to the ope-

ration of some primes such as drivers, among others.

The expectancy system associates to reward (activa-

tion) and punishment (inhibition) behaviors. It is a

ﬂexible mechanism of instrumental learning through

which the behavior is directed towards environmental

stimulus and against existing threats.

It may be said that attention is an implicit concept

of motivational-emotional phenomenon. For (Dama-

sio, 2003), attention is a mechanism that allows the

maintenanceof a mental image in the consciencewith

relative exclusion of others. The inﬂuence of emo-

tional states in attention is recognized. For (Edel-

man, 1987), attention seems to comprehend multiple

mechanisms operating in many levels which involve

large brain portions, including emotional areas.

The cognitive process also includes a phe-

nomenon that can seem as an embarrassment to lear-

ning: the forgetfulness. Using the (Edelman, 1987)

TNGS terms, forgetting implies the decay population

process subjacent to maintenance of neural group se-

lection due the absence of neural activity.

2.3 From Biology to Modeling of

Intelligent Systems

These ideas and conceptssubstantiate the modelingof

AME mechanism. The primes as specialized and in-

nate systems suggest the creation of specialized affec-

tive circuits. So, artiﬁcial primes need to be integrated

to the NS of the CSSA. These primes are specialized

according to the complexity of their operation. The

reﬂections/instincts, drives and primary affects must

be incorporated to support the automatic reactions,

the survival basic needs and other important affec-

tive needs, respectively. A CSSA will only have the

primes needed to a given application. In other words,

a phylogeny or a lineage of agents will have similar

primes.

primepriority(IP)=10

reward/inhibition(RI)mechanism:

rangeofmotivationalpotencial(MP):

-indexofforgetfullness(IF)=0.005

-timetoforgetfullness(DTF)=10cycles

lowHunger normalHunger highHunger

-10 -2

S1S2S3

-5

2.5

currentreadoutlevel(RL)

sensory-motor(SM)

correlation

modulated

correlation

Figure 1: The primes structure.

The permanent operation of artiﬁcial primes inﬂu-

ences all the behaviors of the CSSA. The perception-

action (PA) dynamic of agents must express a motiva-

tion of primes’ auto-regulation. This auto-regulation

tends to establish a primes’ redout level (RL) found in

the limits of what is considered a great equilibrium.

Attention should focus in the direction of satisfaction

of the unregulated affective necessities at a certain in-

stant.

The agents’ learning occurs according to what is

suggested by the expectative system. There must be a

mechanism responsible to reinforce and inhibit beha-

vior conducive or not conducive to regulation of em-

bedded primes. According to the suggestion made by

(Buck, 1999), the developmental primes dimension

ATTENTION, MOTIVATION AND EMOTION IN COGNITIVE SOFTWARE AGENTS

427

needs a kind of attenuation to the acquired conditio-

nings, over time. For such, a selection of the most

executable behaviors must at same time reproduce

the acquired learning extinction effect associated with

learning of behavioral repulsion or reward.

3 THE AME MECHANISM

The AME model consists of the artiﬁcial primes,

of an aggregating component of these primes called

AME container, of a mechanism of behavioral re-

ward/inhibition (RI) and of the attentional mechanism

(Terra et al., 2006). The primes are divided into three

categories: innate contextual reaction for the reﬂexes

and instincts; primary needs for the drives and com-

pelling state for the primary affection.

All primes are aggregated by the AME container

its operation deﬁnes a global emotional state (AME

state) that will inﬂuence the CSSA PA at every in-

stant. Based on the AME state, the RI and atten-

tional mechanisms operate to select an environmen-

tal stimuli and a viable behavior, considering the in-

nate tendencies and the experiences acquired by the

agent throughout its ontogeny. The internal dynamic

of the primes occur via alterations of its activation

level (RL) within the pre-established maximum and

minimum limits. The speciﬁcation of the activation

states for the CSSA hunger drive used in the exper-

iments performed is illustrated in Figure 1. The dy-

namic of the AME container consists of considering

all active primes and deﬁning the AME state.

Based on the dynamic of each NS structure, the

PA cycle should select a sensory-motor (SM) corre-

lation for the agent representing its emergent global

state. The SM correlations are the elements to be

considered by the RI mechanism for application of

the coefﬁcients of learning and forgetting. Learning

occurs when a performed SM correlation corresponds

to a behavior that promotes the re-establishment of

the equilibrium or increases the disequilibrium of any

primes. When this occurs, a learning coefﬁcient pro-

portional to the degree of affective regulation/non-

regulation is calculated. This coefﬁcient will be ap-

plied to update the motivational potential (MP) of the

respectiveprimes in the SM correlationsinvolved (see

Figure 1). Thus, the MP refers to the inﬂuence that a

speciﬁc primes possesses for the establishment of a

certain SM correlation in the CSSA.

The selected action reﬂects the inﬂuence of not

only one but potentially all the affective necessities of

the agent. This is one of the aspects of the proposed

attentional mechanism. The other is that the AME

state also modulates the selection of the environmen-

updateAttention

(attentionalAtributes)

HungerDrive

VisionSensor

Eye

(peripheralcomponent)

ATTENTIONAPPROACHSOURCE

“ON”“fast”“hungerDrive”

updatePredisposition

(approach,status,

intentions)

Attentional Atributeto“High”state:

Figure 2: Attencinal mechanism: peripheral atributes.

tal stimuli percieved by the peripheric sensory-motor

components of the CSSA. An example of an atten-

tional attribute sent to a visual sensor by the agent’s

hunger drive is illustrated in Figure 2.

The operation of the AME mechanism assures a

CSSA auto-regulation (an homeostasis maintanence)

based on its AME state. It should be emphasized that

this emotional permanent inﬂuence is essential for the

incorporation of the experiences in accordance to the

situated cognition approach.

4 COMPUTATIONAL

EXPERIMENTS

The experiments for the validation of the AME me-

chanism were made in a two dimension artiﬁcial life

environment. In it is possible to insert different types

of nutrients (green, blue and red apples) and a non-

edible object (brick). The walls that limit the ”ar-

tiﬁcial world” were also inserted during the initial-

ization. All these are software components (non-

cognitive beings) that the CSSA will interact through

the exchange of stimuli by the sensory-motor compo-

nents of its peripheral system (Figure 3).

The CSSA instanciated included the following NS

structures: a) Vision Sensor; b) Mouth Touch Sensor;

c) ME Primary Temporal Need Hunger; d) Hit And

Turn Reﬂex; e) Mouth Moviment Effector; f) Trans-

lation Effector; g) Rotation Effector. The are two

primes: (i) a ”Hit and Turn” reﬂex to make a right

turn after CSSA hit a wall and (ii)a drive for ”hunger”

as a temporal primary need whose RL evolves from a

unit at each time.

It should be considered that among the possible

SM correlations some should be deﬁned as innate ten-

dencies. It is so to motivate the CSSA behavior of

approximation and ingestion of the entities detected

in its visual ﬁeld. These are seen by the agent as

a nutrient if a positive nutritional coefﬁcient (NC) is

perceived by it. For example, imagine a CSSA with

hunger drive in a high state. After the ingestion of a

blue apple (NC = 4 kcal) the hunger drive signals sat-

isfaction and generates a positive (proportional)affec-

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

428

blueApple(4kcal)

redApple(1kcal)

greenApple(8kcal)

brick(non-edible)

visualfield

mouth

CSSA’sbody

track

points

walls

Figure 3: Non-cognitive components of the CSSA’ artiﬁcial

world.

tive regulation index that should be applied to rein-

force the MP of all SM correlations associated with

that behavior. The MP scale ranges from -5 to 5, the

same is true for all CSSAs primes (Figure 1).

With respect to the learning extinction, as the SM

correlations aren’t being selected their MP is adjusted

gradually in direction of a neutral MP (0). This neu-

tral MP is the one at the beginning of CSSA ontogeny

for all primes except innate contextual reaction.

5 FINAL CONSIDERATIONS

The proposed model presents some advantages in re-

lation to other affective mechanism proposals (see

(Terra, 2007) for details). It is possible to make im-

portant conﬁgurations to the AME mechanism spec-

ifying: (i) the internal speciﬁc dynamic of each

primes; (ii) the attentional attributes to the peripher-

ical system; (iii) a strategy to generate the AME state.

Second, behavior can be inﬂuenced by more than one

emotion. The resolution of the affective need will be

limited only by the sensori-motor skills.

Third, it is considered that the outcomes that were

obtained are satisfactory. The model is coherent to the

non-representationalist view and with the cognitive-

emotional process as stated. The AME mechanism is

appropriate to tally with autonomy (auto-regulation)

of a CSSA. One aspect to be explored is the insertion

of high order emotions in the AME model. Finally,

it is necessary to clarify what are the applications of

the model proposed. If the behavior of the software

system must be determined by the received external

stimulus, the proposed model is not applied. Intel-

ligent systems that have these characteristics are not

applications of cognitive and situated agents.

ACKNOWLEDGEMENTS

The authors would like to thankFAPEMIG and CNPq

agencies for their ﬁnancial support.

REFERENCES

Bateson, G. (1972). Step to an Ecology of Mind. Chandler

Publ. Group, S. Francisco.

Buck, R. (1999). The biological affects: a typology. Psy-

chological Review, 106, No. 2:301–336.

Clancey, W. J. (1997). Situated Cognition: on human

knowlegde and computer representation. Cambridge

University Press, New York.

Damasio, A. R. (2003). O Erro de Descartes: emoc¸˜ao,

raz˜ao e o c´erebro humano. Compahia das Letras, S˜ao

Paulo.

Edelman, G. M. (1987). Neural Darwinism: the theory of

neuronal group selection. Basic Books, New York.

LeDoux, J. (2003). Sinaptic Self: how our brain became

who we are. Penguin Group, New York.

Maturana, H. R. and Varela, F. J. (2002). A `arvore do

Conhecimento: as bases biol´ogicas da compreens˜ao

humana. Palas Athena, S˜ao Paulo, 2 edition.

Santos, B. A. (2003). Aspectos conceituais e arquiteturais

para a criac¸˜ao de linhagens de agentes de software

cognitivos e situados. Mestrado em manufatura in-

tegrada por computador, Centro Federal de Educac¸˜ao

Tecnol´ogica de Minas Gerais, Belo Horizonte.

Terra, D. C. (2007). Desenvolvimento de um mecanismo

atencional-motivacional-emocional para organismos

artiﬁciais: uma abordagem a partir da cognic¸˜ao situ-

ada. Mestrado em modelagem matem´atica e computa-

cional, Centro Federal de Educac¸˜ao Tecnol´ogica de

Minas Gerais, Belo Horizonte.

Terra, D. C., Borges, H. E., and Almeida, P. E. M. (2006).

Uma implementac¸˜ao para as emoc¸˜oes prim´arias em

agentes de software cognitivos e situados. In IX

Encontro de Modelagem Computacional, Belo Hori-

zonte.

Terra, D. C., Grandi, A., and Borges, H. E. (2004). A abor-

dagem enaction para a cognic¸˜ao e suas implicac¸˜oes na

modelagem de sistemas inteligentes. In 8th Brazil-

ian Symposium on Neural Networks, S˜ao Lu´ıs do

Maranh˜ao. CEFALA.

Varela, F. J., Thompson, E., and Rosch, E. (2003). A Mente

Incorporada: ciˆencias cognitivas e experiˆencia hu-

mana. Artmed Editora, Porto Alegre.

ATTENTION, MOTIVATION AND EMOTION IN COGNITIVE SOFTWARE AGENTS

429