ADCAIJ: Ad ances in Dis ibu ed Compu ing and A i icial In elligence Jou nal
Regula Issue, Vol. 9 N. 4 (2020), 55-70
eISSN: 2255-2863
DOI: h ps://doi.o g/10.14201/ADCAIJ2020945570
55
Jose A. Maes o-P ie o, Sa a Rod iguez, Robe o
Casado and Juan M. Co chado
Agen o ganisa ions: om independen agen s o i ual
o ganisa ions and socie ies o agen s
ADCAIJ: Ad ances in Dis ibu ed Compu ing
and A i icial In elligence Jou nal
Regula Issue, Vol. 9 N. 4 (2020), 55-70
eISSN: 2255-2863 - h ps://adcaij.usal.es
Ediciones Uni e sidad de Salamanca - cc by-nc-nd
Agen o ganisa ions: om independen
agen s o i ual o ganisa ions and
socie ies o agen s1
Jose A. Maes o-P ie oa, Sa a Rod igueza, Robe o Casadoa
and Juan M. Co chadoa
a BISITE Resea ch G oup. Depa men o Compu e Science and Au oma ic, Uni e si y o
Salamanca, 1 Escuelas S ., Salamanca, 37003
[email p o ec ed], [email p o ec ed], [email p o ec ed], [email p o ec ed]
KEYWORD ABSTRACT
Open mul i-
agen sys ems;
Vi ual
in as uc u es;
Re iew.
Real wo ld applica ions using agen -based solu ions can include many agen s
ha needs o communica e and in e ac wi h each o he in o de o mee hei
objec i es. In o ganisa ions; Agen open mul i-agen sys ems, p oblems can
include no only he o ganisa ion o a la ge numbe o agen s, bu can also be
he e ogeneous and o unp edic able p o enance o beha io . An o e iew o
he al e na i es o dealing wi h hese p oblems is p esen ed, highligh ing he
way hey y o sol e o mi iga e hem. This app oach allows he de elopmen o
complex sys ems in which he e a e agen s ha show e y di e en beha iou s
and ha a e able o adap o un o eseen changes in he en i onmen . This
makes i possible o simula e socio- echnical o na u al en i onmen s and
obse e hei possible e olu ion wi hou he e hical conside a ions in ol ed in
expe imen ing in eal en i onmen s.
1 This wo k has been de eloped as pa o “Vi ual-Ledge s-Tecnologías DLT/Blockchain y C ip o-IOT sob e o ga-
nizaciones i uales de agen es lige os y su aplicación en la e iciencia en el anspo e de úl ima milla”, ID SA267P18,
p ojec inanced by Jun a Cas illa y León, Conseje ía de Educación, and FEDER unds. I has been pa ially suppo ed by
he Eu opean Regional De elopmen Fund (ERDF) h ough he In e eg Spain-Po ugal V-A P og am (POCTEP) unde
g an 0631_DIGITEC_3_E (Sma g ow h h ough he specializa ion o he c oss-bo de business ab ic in ad anced
digi al echnologies and blockchain.).
56
Au o es
Tí ulo
ADCAIJ: Ad ances in Dis ibu ed Compu ing
and A i icial In elligence Jou nal
Regula Issue, Vol. 9 N. 4 (2020), 55-70
eISSN: 2255-2863 - h ps://adcaij.usal.es
Ediciones Uni e sidad de Salamanca - cc by-nc-nd
1. In oduc ion
Dis ibu ed A i icial In elligence (DAI) is a way o dealing wi h complex sys ems. Agen -based
sys ems a e conside ed one o he h ee ca ego ies in which DAI solu ions a e classi ied in (Do i e
al., 2018). Pa allel AI and Dis ibu ed P oblem Sol ing (DPS) a e he o he wo ca ego ies. Pa allel AI
in ol es he de elopmen o pa allel al e na i es o classical AI algo i hms as a way o inc ease e i-
ciency, whe eas DPS in ol es he di ision o a ask in o sub- asks and hei assignmen o compu ing
en i ies wi h a p ede ined communica ion scheme which may limi hem. In (Ye e al., 2017) i is said
ha when a p oblem is complex o la ge enough o he domain o he p oblem is unp edic able, he
only easonable way o add ess i is o de elop se e al speci ic unc ionali ies and componen s (agen s)
specialized in he solu ion o a pa icula aspec o he p oblem. This allows each agen o use he mos
app op ia e app oach o sol e i s pa icula p oblem(s). Whene e an in e ac ion occu s, he agen s
mus
coo dina e wi h each o he o ensu e ha he in e ac ion is p ope ly managed.
A good guide o agen -based modelling (ABM) o complex sys ems ha includes analysis, e i-
ica ion and alida ion o ABM can be ound in (Wilensky and Rand, 2015). The usage o agen -based
app oaches o p oblem
sol ing is sp eading in many ields, so ecen examples o agen -based
solu ions can be ound o indus y in (He e a e al., 2020; Ho mann, 2019), biology in (Zhang
and DeAngelis, 2020; Soheilypou and Mo ad, 2018), communica ions ne wo k modelling in (De-
akhshan and Youse i, 2019; Su agunda e al., 2019), supply chain modelling in (Dominguez, 2020;
A i ida, 2018), ene gy op imiza ion in (Al-Issaei e al., 2019), e c. As mo e and mo e o hese agen -
based sys ems a e implemen ed, hey will e en ually ha e o communica e and in e ac wi h each o he
o mee inc easingly complex objec i es ha may equi e comple ing asks ou o hei own scope.
This pape p o ides a b ie o e iew o agen -based sys ems and open mul i-agen sys ems and
and how hey can be o ganised o mee an objec i e. The es o his pape is o ganized as ollows. The
nex sec ion in oduces
he agen -based sys ems and mul i-agen sys ems and hei expec ed ea u es.
Sec ion 3 p o ides an o e iew o
he di e en ways a MAS can be o ganized, including i ual o gani-
sa ions, o e lays and simula ions. Sec ion 4
includes a b ie lis o pla o ms and amewo ks and some
e e ences o cu en examples o hei use. Finally,
Sec ion 5 concludes he pape .
2. Agen -based Sys ems
Se e al common de ini ions o agen a e included in (Do i e al., 2018), as well as he p oposal
o a new de ini ion. Focusing on he ea u es and capabili ies o he agency, in (Do i e al., 2018) is
p oposed o de ine an agen as an en i y si ua ed in an en i onmen ha pe cei es di e en pa ame e s
ha i uses o make a decision based on he en i y’s goal and akes he necessa y ac ion in he en i-
onmen in acco dance wi h he decision made. Acco ding o he au ho s, he en i y e e s o he ype
o agen (whe he so wa e, ha dwa e o hyb id) and he en i onmen e e s o he place he agen is
loca ed. The agen may be in luenced by he ea u es o he en i onmen (accessibili y, de e minism,
dynamism, con inui y,…). The pa ame e s e e o he da a pe cei ed by he agen and he ac ion e e s
o he changes ha he agen p oduces in he en i onmen . The goal o he agen is o sol e a ask may-
be wi h some cons ain s (e.g., a limi ed amoun o ime). Sol ing complex asks and ha ing a wide
applica ion equi es agen s o ha e he ollowing ea u es (Do i e al., 2018):
Sociabili y ha allows agen s o sha e hei da a and knowledge.
Au onomy which allows agen s o independen ly execu e hei p ocesses, make decisions and pe -
o m hei ac ions.
57
Au o es
Tí ulo
ADCAIJ: Ad ances in Dis ibu ed Compu ing
and A i icial In elligence Jou nal
Regula Issue, Vol. 9 N. 4 (2020), 55-70
eISSN: 2255-2863 - h ps://adcaij.usal.es
Ediciones Uni e sidad de Salamanca - cc by-nc-nd
P oac i i y which allows agen s o use hei da a and knowledge, and also ha o o he s, o o esee
u u e ac ions and mee hei objec i es.
O he au ho s, e.g. (Ye e al., 2017), include o he ea u es such as eac i i y, as agen s a e si ua ed
in an en i onmen , pe cei e da a and hence can espond o changes in he en i onmen o mee hei
objec i es.
Simple examples o agen s can be ound in he e iewed p oposals. Fo ins ance, in (Ye e al.,
2017) wo
clea examples a e in oduced: a empe a u e senso and a demon (a so wa e p og am). Bo h
ob ain in o ma ion
om hei en i onmen , and make decisions based on he da a and some ules o
ac ing on hei en i onmen s. Mo e ex ensi e and de eloped examples can be ound in (Wilensky
and Rand, 2015). Howe e , as concluded
in (Rabuzin e al., 2006), o en one agen is no enough
o sol e any p oblem and he e o e mo e agen s may
be needed. In addi ion, a g oup o agen s can
p o ide in e es ing ea u es, as hey can achie e be e esul s by pe o ming pa allel p ocessing ha
inc eases speed and pe o mance, g ea e sys em s abili y can be achie ed, and
lexibili y and eusabili y
can be imp o ed. Consequen ly, he idea o collabo a ion o sol e complex p oblems aises na u ally as
agen s a e a lexible way o o ganise so wa e modules due o hei abili y o lea n and make
decisions
au onomously. Howe e , (Gómez-C uz e al., 2017) poin s ou wo consequences o he non-linea
in e ac ion among agen s and be ween agen s and he en i onmen . Fi s , eme ging and global pa e ns
may a ise
om his in e ac ion. These pa e ns can be classi ied as s uc u al, beha io al, o unc ional.
And secondly, he e
is a coupling be ween he agen s and hei en i onmen s; he agen s may be sensi i e
o he ini ial condi ions and
hei ela ionships. These can cause ne wo k e ec s and in e dependencies
a di e en le els ha can lead o
cascading ailu es and can limi he con ollabili y and p edic abil-
i y o he en i e sys em. ABM ies o mi iga e
hese limi a ions by abs ac ing he sys em componen s,
hei ac ions, in e ac ions and he en i onmen (Wilensky
and Rand, 2015). A key ea u e o ABM is
ha he e is an explici model o hese in e ac ions (Macal, 2016),
which equi es he speci ica ion o
which agen is linked o which o he agen s and he de ini ion o in e ac ion s a egies.
Rela ed o ABM is also he concep o Agen Based Simula ion (ABS) which is he compu a ional
implemen a ion
o he model and ob aining i s dynamics o e ime (Gómez-C uz e al., 2017).
The simula ion manages o link he beha io o indi idual agen s wi h he mac o-beha io al
pa e ns ha a ise om hei in e ac ions. ABS is use ul hen (Gómez-C uz e al., 2017; Wilensky
and Rand, 2015):
• The e is mul iple au onomous and he e ogeneous agen s in he sys em.
• The agen s ope a e in a local, pa allel and dis ibu ed way, wi hou global knowledge. The e a e
non- linea , asynch onous and discon inuous in e ac ions. Small ac ions can p opaga e, igge -
ing e ec s and luc ua ions.
• The global dynamics o he sys em is sel -o ganizing and eme gen . Fea u es such as memo y,
lea ning, adap a ion o e olu ion a e p esen in he sys em.
• The sys em is s uc u ed in space- ime dimensions.
• The en i onmen is unce ain.
The e o e, ABS can be used o es concep s in an expe imen al en i onmen . As he implemen a-
ion o a sys em can ha e possible nega i e implica ions ha canno be easily o eseen, ABS allows
es ing i unde a ious condi ions o es ima e i s impac . This can help o e alua e he cha ac e is ics
o socio- echnical sys ems, such as complex human-machine in e ac ions (Gómez-C uz e al., 2017).
58
Au o es
Tí ulo
ADCAIJ: Ad ances in Dis ibu ed Compu ing
and A i icial In elligence Jou nal
Regula Issue, Vol. 9 N. 4 (2020), 55-70
eISSN: 2255-2863 - h ps://adcaij.usal.es
Ediciones Uni e sidad de Salamanca - cc by-nc-nd
2.1. Mul i-Agen Sys ems
A Mul i-Agen Sys em (MAS) is basically a se o au onomous agen s si ua ed in an en i onmen ,
ha can espond o changes in he en i onmen and y o achie e hei own o a common goal. In a
MAS, some beha iou s such as delega ion o goals o asks may occu be ween he agen s and, he e-
o e, in e ac ion and collabo a ion be ween hem mus be possible (Abbas e al., 2015), bu compe i-
ion may also occu ins ead o coope a ion o denial o in o ma ion exchange (Ye e al., 2017). Recen
e iews ela ed o MASs and he o ganisa ion o MASs can be ound, o example, in (Do i e al.,
2018) i can be ound a e iew o MASs, in (Abbas e al., 2015) is e iewed he o ganisa ion o MASs
and in (Ye e al., 2017) a e su eyed he sel -o ganisa ion mechanisms o MASs.
The implemen a ion o a MAS is a complex ask ha includes all he ea u es o a adi ional dis-
ibu ed and concu en sys em and speci ic ones such as au onomy, lexibili y o complex pa e ns o
in e ac ion be ween indi idual agen s (Ye e al., 2017). In e u n, a MAS allows o inc eased e icien-
cy by di iding up he wo k and aking ad an age o i s dis ibu ed na u e. Agen s can sol e hei ask
acco ding o hei own knowledge. This adds lexibili y o he model and can inc ease eliabili y as a
ask can be eassigned in case o ailu e o an agen (Do i e al., 2018). Howe e , some communica-
ion o e load may occu , as agen s mus communica e wi h each o he . This o e load can be educed i
agen s only communica e be ween close neighbou s. This app oach allows he sys em o expand wi h-
ou inc easing he communica ions o e load much, bu i is necessa y o c ea e in e media e agen s
o p o ide some se ices, such as main aining a lis o he se ices o e ed by each agen ,
which
imp o es he agen loca ion p ocess. A MAS has i s speci ic ea u es (Do i e al., 2018):
Leade ship whe he o no he e is an agen who de ines he goals o he o he agen s acco ding
o a gene al goal.
Decision- aking unc ion ea u es whe he i is linea wi h he alue o he inpu s o no .
He e ogenei y whe he all agen s ha e he same ea u es o no .
Ag eed pa ame e s whe he agen s should ag ee on pa ame e s called “me ics”.
Delays whe he accoun is aken o possible delays (in communica ions, in access o esou ces) o
no .
Topology whe he he loca ion o he agen s and he ela ions be ween hem a e s a ic o dynamic.
Da a ansmission equency i he agen s obse e hei en i onmen pe iodically and communi-
ca e i o he es o he agen s o i hey only obse e i a pa icula e en occu s.
Mobili y s a ic agen s a e always loca ed in he same posi ion in he en i onmen , while mobile
ones can change hei loca ion.
A key ea u e o a MAS is i s abili y o eo ganise o adap o changes in i s en i onmen (Abbas
e al., 2015). Sel -o ganisa ion is a challenge in i sel , bu is pa icula ly p oblema ic in an open MAS.
An open MAS allows new unknown agen s o join (o lea e) eely and in e ac wi h o he s a un ime.
In his ype o MAS, agen s de eloped by di e en s akeholde s, as ’black boxes’ and wi h hei own
goals, join he MAS. This ea u e is o in e es in some applica ions, o example in sma ci ies o IoT
ne wo ks, bu o ganisa ional and secu i y p oblems a e pa icula ly ele an (Se ano and Bajo, 2020;
De akhshan and Youse i, 2019; Bijani and Robe son, 2014).
59
Au o es
Tí ulo
ADCAIJ: Ad ances in Dis ibu ed Compu ing
and A i icial In elligence Jou nal
Regula Issue, Vol. 9 N. 4 (2020), 55-70
eISSN: 2255-2863 - h ps://adcaij.usal.es
Ediciones Uni e sidad de Salamanca - cc by-nc-nd
3. MAS o ganisa ion
In (Abbas e al., 2015) classi y he de elopmen o MAS in wo main app oaches:
ACMAS (Agen Cen e ed MAS) ocuses on indi idual agen s. ACMAS assumes ha he gene al
unc ionali y o he sys em will appea as a esul o he in e ac ion be ween indi idual agen s
in a bo om-up app oach. The o ganisa ion will eme ge as a pa e n o collabo a ion be ween
agen s, which may be unp edic able and unce ain: undesi able beha iou s may appea ha
can a ec he pe o mance o he sys em. The e o e, his app oach canno be alid o complex
sys ems.
OCMAS (O ganisa ion Cen e ed MAS) ocuses on he s uc u e o he sys em. The o ganisa ion
and coo dina ion o he agen s is designed sepa a ely om he local beha iou o he agen s. I
is a op-down app oach in which he o ganisa ion imposes some ules ha he agen s use o co-
o dina e hei local beha iou and in e ac ions. This p o ides some esou ces o coo dina ion
ha enable he global sys em o achie e i s gene al goal. OCMAS is maybe mos app op ia e
in a numbe o cases such as: he e a e a la ge numbe o agen s, agen s need a lo o ime o
comple e hei asks, many asks need sha ed esou ces, he e a e many collabo a i e asks,
many agen s a e specialized, ew agen s a e able o do some asks o he e a e ew esou ces.
In (P emm and Ki n, 2015), he o ganisa ion is desc ibed as a use ul me apho o desc ibing,
s udying and designing dis ibu ed so wa e sys ems; and in (Dignum, 2009) i is s a ed ha he o -
ganisa ion o agen s can be conside ed om wo di e en poin s o iew: as a p ocess and as an en i y
in i sel : in he i s one, o ganisa ion is he p ocess o o ganisa ion o indi idual agen s and implies
limi a ions in he allowed in e ac ions o he agen s. In he second, he o ganisa ion is conside ed as an
en i y wi h i s own goals and is ep esen ed as a g oup o agen s, al hough i is no exac ly he same.
Howe e , in (Abbas e al., 2015) an o ganisa ion in eg a es bo h poin s o iew and makes ex ensi e
use o he e ogenei y and open MAS. An o ganisa ion can also p o ide a way o g oup agen s, di ide
he sys em and hus achie e a g ea e deg ee o abs ac ion and modula i y o he sys em, inc easing
lexibili y and acili a ing main enance. Fu he mo e, each g oup o agen s can be a con ex in which
he agen s can eely in e ac wi h each o he .
Se e al o ganisa ional pa adigms ha e been p oposed o use in MASs, including: hie a chies,
hola chies,
eams, ede a ions, ma ke s, e c. Mo e comple e lis s and he ea u es o hese p oposals
can be ound in (Do i
e al., 2018; Abbas e al., 2015; Ho ling and Lesse , 2004). Howe e , none o
hese o ganisa ional pa adigms
always i well in any possible si ua ion. The bes possible pa adigm
depends on he goals o he agen s, he
a ailable esou ces and he en i onmen in which he agen s
will be ope a ing. Fu he mo e, a MAS can be
o ganised no only s a ically using one o mo e o he
a o emen ioned pa adigms, bu also dynamically, which makes i s design mo e complex. An AMAS
(Adap i e MAS) is designed o be able o au onomously adap i s o ganisa ion o un o eseen si ua ions,
eo ganising i sel acco ding o he changes in i s en i onmen .
3.1. Reasoning abou he o ganisa ion
The wo design possibili ies in he de elopmen o a MAS men ioned, depend on whe he o no he
agen s a e awa e o he exis ence o he o ganisa ion. I he agen s do no ha e an in e nal ep esen a ion
o he o ganisa ion
and a e he e o e no awa e ha hey belong o an o ganisa ion, o only ha e an
in e nal ep esen a ion o
he coope a ion pa e ns ob ained by pe cep ion, communica ion o by hei

60
Au o es
Tí ulo
ADCAIJ: Ad ances in Dis ibu ed Compu ing
and A i icial In elligence Jou nal
Regula Issue, Vol. 9 N. 4 (2020), 55-70
eISSN: 2255-2863 - h ps://adcaij.usal.es
Ediciones Uni e sidad de Salamanca - cc by-nc-nd
own easoning, hen hey use he ACMAS app oach. I he o ganisa ion exis s as a speci ied scheme
(pe haps e en ha d-coded) and he agen s mus comply wi h o ganisa ional cons ain s o he agen s
ha e an explici ep esen a ion o he o ganisa ion in which hey ha e been de ined and can use i in
hei easoning p ocess, hen hey a e conside ed o use he OCMAS app oach.
The e exis s some me hodologies o o ganisa ional models ha ollows hese app oaches. Fo in-
s ance, swa m-based mul i-agen sys ems such as he desc ibed in (Duan e al., 2012) ollows he
ACMAS app oach, on he o he hand, he MaSE (Mul iagen Sys ems Enginee ing) me hodology (De-
Loach, 2004) and he MOISE+ (Model o O ganisa ion o mul I-agen Sys Ems) model (Hübne e
al., 2002; Hübne e al., 2007) a e examples o he OCMAS app oach. Howe e , some o he p oposals
include a hyb id o bo h app oaches, e.g., he o ganisa ional model NOSHAPE (Abbas and Shaheen,
2017).
Th ee o ganisa ional dimensions a e iden i ied in (Hübne e al., 2007):
• unc ioning o he o ganisa ion, ela ed o he speci ica ion o he global plans, he policies o
assignmen o asks o he agen s, he coo dina ion o he execu ion o he plans o he accoun -
abili y o he plans ( ime consump ion, use o esou ces,…),
• o ganisa ional s uc u e, e.g., oles, ela ionships be ween oles, ole g oups, e c., and
• de ini ion o high le el no m ha agen s mus obey, which can be used o egula e wha an
agen can do in an o ganisa ion. Fo ins ance, he middlewa e o MOISE+ egula es wha an
agen can do by blocking ce ain ac ions (Jensen e al., 2014).
A sligh ly di e en app oach is aken in AORTA (Adding O ganiza ional Reasoning o Agen s),
ha is desc ibed in (Jensen, 2015); i is a amewo k ha allows agen s o eason abou an o ganisa ion.
AORTA allows he s udy o en i onmen s whe e agen s enac oles and sol e objec i es o he o gani-
sa ion. This amewo k has been in eg a ed in o di e en agen pla o ms such as Jason (Jensen e al.,
2014) o GAMA (La sen, 2018).
3.2. Sel -o ganisa ion
Sel -o ganisa ion can be de ined as a mechanism o p ocess ha allows a sys em o change i s o -
ganisa ion
wi hou an explici ex e nal command while i is unning (Di Ma zo Se ugendo e al., 2005).
This implies ha i
is c ea ed some pa e n by he coope a i e beha iou o indi idual agen s wi hou
ex e nal con ol o in luence,
as his is conside ed an e ec i e way o dealing wi h dynamic equi e-
men s in dis ibu ed sys ems (Ye e al.,
2017). Sel -o ganised sys ems show he ollowing ea u es:
No ex e nal explici con ol : as he sys em is au onomous, adap a ion and change mus depend
only on in e nal decisions made by he componen s o he sys em.
Decen alised con ol : sel -o ganisa ion can be achie ed as a esul o local in e nal in e ac ions
wi h sys em componen s wi hou any cen al con ol.
Dynamic and e olu iona y ope a ion : Such a sys em should be able o e ol e o adap o chang-
es in he en i onmen .
Task and esou ce alloca ion mechanisms in sel -o ganising sys ems can a oid he single poin o
ailu e o cen alised alloca ion mechanisms. Fu he mo e, hese mechanisms in sel -o ganising sys-
ems a e scalable and allow each agen o adap i s beha iou o ob ain an e icien alloca ion o asks
and esou ces in open and dynamic sys ems, wi hou global in o ma ion (Ye e al., 2017).
61
Au o es
Tí ulo
ADCAIJ: Ad ances in Dis ibu ed Compu ing
and A i icial In elligence Jou nal
Regula Issue, Vol. 9 N. 4 (2020), 55-70
eISSN: 2255-2863 - h ps://adcaij.usal.es
Ediciones Uni e sidad de Salamanca - cc by-nc-nd
3.3. Vi ual O ganisa ions o Agen s
The idea behind a Vi ual O ganisa ion (VO) is aken as he basis o o mula ing a new pa adigm
o an agen o ganisa ion.
A VO can be de ined as a se o indi iduals and ins i u ions ha need o coo dina e esou ces and
se ices ac oss ins i u ional bounda ies (A gen e e al., 2011). VOs a e he e o e composed o agen s
and MAS ha collabo a e h ough he execu ion o se ices. In (Cama inha-Ma os e al., 2010) i is
said ha VOs a e empo a y alliances o en i ies ha sha e hei skills, compe encies and esou ces o
be e espond o business oppo uni ies and whose collabo a ion is suppo ed by compu e ne wo ks.
Howe e , his is no achie ed wi hou many obs acles: he he e ogenei y o he au onomous pa ici-
pan s and he ime needed o c ea e us a e men ioned. (Cama inha-Ma os e al., 2010) p o ides se -
e al examples o he use o i ual o ganisa ions and simila app oaches in di e en en i onmen s such
as ag o-indus y, anspo sys ems, ene gy and wa e managemen , biodi e si y, sus ainable ou ism
and o he s.
This idea o using he o ganisa ional model as a way o a ange agen s and i s beha iou in a MAS
can be
ound in se e al p oposals:
MOISE is a model ha di ides he speci ica ion o an o ganisa ion o agen s in o h ee di e en
pa s: s uc u al, unc ional and deon ological speci ica ions. The s uc u al speci ica ion de-
ines he s uc u ing o he agen s using oles, ela ionships be ween oles and g oups. Roles
de ine a se o cons ain s ha an agen mus mee o become a membe o a g oup. The ela-
ionships be ween oles a e links be ween oles and also compa ibili ies om a sou ce ole o a
des ina ion ole. A g oup speci ica ion consis s o oles, subg oup de ini ions (decomposi ion),
linkage de ini ions and compa ibili ies and ca dinali ies o g oups and oles. The unc ional
speci ica ion desc ibes how an agen o ganisa ion achie es i s objec i es.An o e all objec i e
can be decomposed (using a planning algo i hm) and dis ibu ed o agen s (as missions). The
decomposi ion o he o e all objec i e in o plans and missions is simila o a ee, which in
MOISE is called a social scheme. The missions a e assigned o he agen s, who a e esponsible
o ul illing all he objec i es o he mission o which hey a e commi ed.
MaSE (O-MaSE, O ganiza ional Mul i-agen Sys em Enginee ing) is a me hodology o he de el-
opmen o agen o ganisa ions. In MaSE, agen classes a e iden i ied only by he oles o which
he agen can be assigned. O-MaSE de ines he o ganisa ional adap a ion p ocess based on he
OMACS amewo k, whe e he bes combina ion o oles and agen s o he cu en ly es ab-
lished objec i e is calcula ed. The MaSE app oach equi es agen s o use global knowledge, so
i is a kind o cen alized app oach, whe eas sel -o ganised sys ems end o be decen alized.
PANGEA (Pla o m o Au oma ic coNs uc ion o o Ganiza ions o in Elligen Agen s) is a pla -
o m o he de elopmen o mul i-agen sys ems modelled as i ual o ganisa ions (Za o e
al., 2013). I is based on he concep s o oles, o ganisa ions and ules ha a e conside ed o
be i s -o de en i ies. PANGEA is based on THOMAS (Rod iguez e al., 2011; A gen e e al.,
2011; Gi e e al., 2010; Ca ascosa e al., 2009). The THOMAS amewo k suppo s he FIPA
(Founda ion o In elligen Physical Agen s) speci ica ion (FIPA, 2002), in oduces he idea
o o e ing a chi ec u al se ices as Web Se ices and is no dependen on any in e nal agen
pla o m, so i is p epa ed o open mul i-agen sys ems. The THOMAS a chi ec u e is shown
in Figu e 1. PANGEA includes di e en models o agen a chi ec u e (BDI and CBR-BDI
a chi ec u es). I includes a communica ion p o ocol ha allows o b oadcas communica ion,
mul icas communica ion o oles o subo ganisa ions o di ec agen - o-agen communica ion.
62
Au o es
Tí ulo
ADCAIJ: Ad ances in Dis ibu ed Compu ing
and A i icial In elligence Jou nal
Regula Issue, Vol. 9 N. 4 (2020), 55-70
eISSN: 2255-2863 - h ps://adcaij.usal.es
Ediciones Uni e sidad de Salamanca - cc by-nc-nd
I also includes a module o in e ac wi h FIPA-ACL agen s, a se ice managemen and ools
o disco e se ices, se ices o dynamically eo ganise he o ganisa ion, se ices o dis ibu e
asks and balance he wo kload and a business ules engine o gua an ee compliance wi h he
ules de ined in he o ganisa ion.
Figu e 1: THOMAS a chi ec u e. A new module e med O ganisa ion Managemen Se ice (OMS)
is inco po a ed in o THOMAS and manages he VO li e cycle (in he same way as he AMS manages
he agen li e cycle) and also ce ain VO managemen ules. The OMS module manages c ea ion and
es ic s he manne in which en i ies pa icipa e in he VO. The Se ice Facili a o (SF), based on an
adap a ion o he SOA Di ec o y Facili a o , is included in he pla o m. This makes i possible o o e
he egis e ed se ices o agen s and VOs.
NOSHAPE The NOSHAPE o ganisa ional model is he newes p oposal so a and is designed o
de elop la ge agen o ganisa ions. I is desc ibed in mo e de ail in he Subsec ion 3.4.
(a) Two o ganisa ions, O0 and O1, o e -
lapped, which allows he in e ac ion
be ween agen s belonging o bo h o -
ganisa ions.
(b) A complex in e ac ion in NOSHAPE. Two uni e ses, ep esen ed as
dashed ec angles, U
0
and U
1
, a e pa ially o e lapped. Uni e se U
0
con ains wo wo lds W
0.0
and W
0.1
depic ed as g ey o als. Each wo ld
con ains one o mo e o ganisa ions. O ganisa ions in same wo ld in
he same uni e se can o e lap, such as 0
0.0.1
and 0
0.0.2
. Also can o e lap
o ganisa ions belonging o di e en wo lds in he same uni e se,
such as 0
0.1.0
and 0
0.0.2
, i he wo lds o e lap. And wo o ganisa ions in
di e en uni e ses, and he e o e in di e en wo lds, can also o e lap,
such as 0
0.0.0
and 0
1.0.0
, i he uni e ses and wo lds a e o e lapping.
Figu e 2: Two examples o o e laps in NOSHAPE.
63
Au o es
Tí ulo
ADCAIJ: Ad ances in Dis ibu ed Compu ing
and A i icial In elligence Jou nal
Regula Issue, Vol. 9 N. 4 (2020), 55-70
eISSN: 2255-2863 - h ps://adcaij.usal.es
Ediciones Uni e sidad de Salamanca - cc by-nc-nd
3.4. NOSHAPE o ganisa ional model
The o ganisa ional model NOSHAPE (Abbas and Shaheen, 2017; Abbas and Shaheen, 2015; Abbas
and Shaheen,
2014) combines he ACMAS and OCMAS app oaches o p o ide a mo e gene al model
han o he p oposals.
O ganisa ional beha iou is sepa a ed om he agen ’s indi idual beha iou , so
ha agen s can ocus on hei
asks wi hou ega d o o ganisa ional issues. Roles a e used implici ly,
wi hou ha ing p ede ined pa e ns o ac i i ies (Abbas and Shaheen, 2017). Agen s a e o ganised
mainly in o ganisa ions. An o ganisa ion ha e some s a ic oles, which help o keep he o ganisa ion
wo king, and o he dynamic oles (agen s) ha depend on he domain. A basic ope a ion on o ganisa-
ions is de ined, he o e lap, which implies he sha ing o some oles (dynamic agen s) be ween wo
(o mo e) o ganisa ions. The o e lap equi es in e ac ion and nego ia ion be ween he o ganisa ions. In
p ac ice, sha ing he dynamic oles means ha he sha ed oles o he O2 o ganisa ion a e egis e ed in
he O1 o ganisa ion’s yellow pages se ice ( he one eques ing he oles) in o de o become “local” o
he O1 o ganisa ion and be accessible o he agen s belonging o O1 (Abbas and Shaheen, 2015). G aph-
ically, he o e lap (see Figu e 2a) is ep esen ed as an in e sec ion be ween he wo o ganisa ions.
Th ee ypes o abs ac ions a e de ined in NOSHAPE as a way o manage he complexi y: o gani-
sa ion, wo ld and uni e se. The mos complex o ganisa ion possible, he ul a-la ge scale o ganisa ion,
equi es h ee s uc u es and he o e lap o wo o ganisa ions equi es ha hei espec i e uni e ses
and wo lds a e p e iously o e lapped. A do no a ion is p oposed o iden i y he di e en s uc u es.
Fo example, wo uni e ses can be iden i ied as U0 and U1, espec i ely. A wo ld in he uni e se U0 is
iden i ied as W0.0, and W1.0, espec i ely, o a wo ld in he uni e se U1. An o ganisa ion iden i ied as
O0.0.0 belongs o he wo ld W0.0 in he uni e se U0. A g aphic example is shown in Figu e 2b. O he in e -
ac ions a e allowed such as undoing an o e lap, o mo e complex ones such as agen s mo ing om one
o ganisa ion o ano he o e lapped one. Only dynamic agen s can be mo ed o ano he o ganisa ion,
hose ha main ain he s uc u e o he o ganisa ion canno be ans e ed. In
he case o he ans e o
an agen , he o ganisa ion o which i belonged o sh inks. To manage hese e en s
(o e laps, sepa a-
ion o o ganisa ions, ans e o agen s, e c.) i is p oposed o use an e en queue managed by he ole
ha main ains he s uc u e in ope a ion (Abbas and Shaheen, 2017).
3.5. Socie ies o Agen s
The de ini ion o VOs also esembles he idea o he open agen socie y desc ibed in (Pi and A -
ikis, 2015),
as a lexible ne wo k o he e ogeneous agen s each awa e o he oppo uni ies a ailable,
capable o making
au onomous decisions and coope a ing o mee ansien needs and condi ions.
The key ideas in VOs o
he e ogenei y, au onomous decision-making and coope a ion o mee an-
sien objec i es also appea in agen
socie ies. This de ini ion allows agen s o ope a e in a non-ideal
way, as compe i ion and con lic ing goals mus
be aken in o accoun when sel -in e es ed and he e o-
geneous agen s om unp edic able p o enance a e p esen .
Agen s and MAS ha e also been used o
simula e and s udy he beha io o socie ies. A simula ion can allow o ill he gap be ween heo e ical
p oposals and empi ical esea ch. Social simula ion can p o ide a way o es he hypo heses o mula -
ed and also a way o obse e social p ocesses and, he e o e, can be a ool o model socie ies (Gilbe
and Con e, 1995; Eps ein, 2006). Social in e ac ion is conside ed o be a balance be ween coope a ion and
con lic , al hough no in i s mos simple way. In a socie y he e can be many o ms o coope a ion
and con lic , al uism and sel ish and desi able and undesi able beha io s. In a socie y, an agen is no a
sel -su icien being, as he e a e usually some ela ionships wi h o he agen s. A ela ionship be ween
agen s can be o a ious ypes, om an acciden al in e e ence ha hinde s o acili a es any o he
70
Au o es
Tí ulo
ADCAIJ: Ad ances in Dis ibu ed Compu ing
and A i icial In elligence Jou nal
Regula Issue, Vol. 9 N. 4 (2020), 55-70
eISSN: 2255-2863 - h ps://adcaij.usal.es
Ediciones Uni e sidad de Salamanca - cc by-nc-nd
Walpole, J., Mac Gabhann, F., Pei ce, S. M., and Chappell, J. C., 2017. Agen -based compu a ional
model o e inal angiogenesis simula es mic o ascula ne wo k mo phology as a unc ion o pe icy e
co e age. Mic oci cula ion, 24(8):e12393. doi:h ps://doi.o g/10.1111/micc.12393.
Wilensky, U. and Rand, W., 2015. An In oduc ion o Agen -Based Modeling: Modeling Na u al, Social,
and Enginee ed Complex Sys ems wi h Ne Logo. The MIT P ess.
Ye, D., Zhang, M., and Vasilakos, A. V., 2017. A Su ey o Sel -O ganiza ion Mechanisms in Mul iagen
Sys ems. IEEE T ansac ions on Sys ems, Man, and Cybe ne ics: Sys ems, 47(3):441-461.
Za o, C., Villa ubia, G., Sánchez, A., Bajo, J., and Co chado, J., 2013. PANGEA: A New Pla o m o
De eloping Vi ual O ganiza ions o Agen s. In e na ional jou nal o a i icial in elligence, 11:93-
102.
Zhang, B. and DeAngelis, D. L., 2020. An o e iew o agen -based models in plan biology and ecology.
Annals o Bo any, 126(4):539-557. doi:10.1093/aob/mcaa043.