Agents and Multi-agent Systems as Actor-networks

Yury Iskanderov

and Mikhail Pautov

The St. Petersburg Institute for Informatics and Automation of RAS,

39, 14-th Line, St. Petersburg, Russia

Foscote Group, 23A Spetson St., 102A, Mesa Geitonia, 4000 Limassol, Cyprus

Keywords: Actor-network Theory, Agents, Multi-agent Systems, Translation.

Abstract: Agents and multi-agent systems are looked at through the lens of actor-network theory which plays a promi-

nent role in the avant-garde of postmodern studies on socio-technological systems. The authors use the ele-

ments of applied semiotics and logics of action to formalize the basic actor-network concepts; they discuss

prospective intercourses between the actor-network paradigm and the agent based approaches, potential syn-

thesis of the two methods and the new semantics of certain MAS concepts suggested by actor-network theory.

1 INTRODUCTION

The core principle of the actor-network theory (ANT)

is based on the idea that the actions of any human or

nonhuman agent (or “actor” in terms of ANT) are me-

diated by the actions of a set of other heterogeneous

actors. This set is informally defined as “actor-net-

work”. The origins of any actor-network are consid-

ered a “rhizome” in Deleuze-Guattarian understand-

ing of this term, i.e. a self-organizing multiplicity

with totally decentralized structure and no hierar-

chical relations between its heterogeneous elements

having only an initial affinity with each other. It can

hardly be identified as a system since it lacks order

and never inherits any order from its predecessors or

constituents, however the initial affinity between its

heterogeneous members helps establish functional

links between them (Deleuze et al., 1993). Heteroge-

neity of actors traditionally understood in the actor-

network theory as their belonging to one of the two

opposite worlds (human/social or nonhuman/natu-

ral/technological) and ability to form steady networks

of human-nonhuman (socio-technological) hybrids

(or quasi-objects) needs to be redefined with the re-

cent advent and coming out of the blue of the artificial

intelligence. The pendulum of the actor-network the-

ory which was earlier oscillating in the two-dimen-

sional field between the two opposite poles (human

and nonhuman) leaving after every sway the two-

https://www.spiiras.nw.ru/en

dimensional hybrids, now acquires the third pole and

the third dimension: nonhuman intelligent objects-

subjects in their interactions with humans and non-

intelligent nonhuman actors. Hence the emerging de-

mand for integration of the actor-network paradigm

with the intelligent systems research, multi-agent sys-

tems studies, knowledge engineering, ergonomics

and other related fields of research and applications.

We foresee the trajectory of evolution of the actor-

network approach from the descriptive theory created

(and further revised) by Latour, Callon, Law and

other ANT protagonists, through its formalization and

integration with other relevant methods of AI and

agent-based research, toward its eventual conversion

into a full-fledged applied tool for modelling and sim-

ulation of socio-technological systems. Starting pav-

ing this way we have demonstrated that the actor-net-

work theory provides new semantics for some core

concepts of the multi-agent systems theory (Iskan-

derov et al., 2020), and suggested to use the elements

of applied semiotics and logics of action (TI, SAL) to

formalize some basic concepts of the actor-network

theory.

Semiotics is considered the ideological nucleus of

ANT, the whole theory being viewed as the newest

phase of evolution of semiotics toward its object-ori-

entedness. Therefore, in our move to assemble basic

formal definitions here we follow the semiotic route

of conceptualization: sign – actor – actor-network.

Iskanderov, Y. and Pautov, M.

Agents and Multi-agent Systems as Actor-networks.

DOI: 10.5220/0008935601790184

In Proceedings of the 12th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2020) - Volume 1, pages 179-184

ISBN: 978-989-758-395-7; ISSN: 2184-433X

179

Sign s can be formally defined as a set of the four

components (Kiselev et al., 2018): <n, p, m, a>, where

n is the name of sign s; p is the portrait (image) of sign

s corresponding to node w

(s) of the causal network

on the portraits (W

); m is the meaning or “sign sig-

nificance” (Kiselev et al., 2017) of sign s correspond-

ing to node w

(s) of the causal network of meanings

); a is the ascription synonymous with attribution,

individual sense or “personal meaning” (Kiselev et

al., 2017) of sign s corresponding to node w

(s) of the

causal network on the ascriptions (W

). R

defines the

relations on the set of signs; Θ defines the operations

on the set of signs based on the fragments of the

causal networks where belong relevant sign compo-

nents. Tuple of the five elements <W

, W

, R

, Θ>

represents the semiotic model of actor A

(Kiselev et

al., 2018). We assume that actor A

has a predeter-

mined goal G

) not achievable by his own efforts

due to the existing obstacle(s). In this situation actor

can either abandon the goal or try to achieve it by

taking an alternative route (detour) through mediation

of other actor(s): A

… (human or nonhuman, in-

telligent or non-intelligent) (Shirokov, 2019). To-

gether they can either strive to achieve the initial goal

) or choose alternative goals (G

…). Return to

the initial goal G

is only one virtual scenario in the

set of alternative scenarios (Figure 1) (Latour, 1994;

Shirokov, 2019).

Figure 1: Translation (Latour, 1994).

Thus actor A

together with the mediating actors

… accepted by A

after negotiation and trans-

formation form a network which, in turn, is trans-

formed by A

(Callon, 1991). Such network of heter-

ogeneous actors is called actor-network (AN). For-

mally speaking, all actions in an actor-network are

distributed on a set uniting actors with human intelli-

gence (humans), nonhuman actors with artificial in-

telligence (AI actors), and nonhuman actors without

intelligence (other artificial/technological and/or nat-

ural objects), i.e. AN

∪ AN

, where AN

is the set of human actors, AN

is the set of AI actors,

is the set of other natural and artificial (techno-

logical) actors (Iskanderov et al., 2020). Any actor A

can be involved (and in the majority of cases is

involved) in multiple actor-networks {AN

}, or:

A∈⋂

. The following core concepts were formu-

lated in ANT:

Generalized Ontological Symmetry: Heterogene-

ous actors share the same capacity for agency (Bal-

zacq, 2016). Equivalence of heterogeneous actors in

their interplay within actor-network AN should mean

that the same criteria and terms are equally applied to

the technological and natural actors on the one hand,

and the socio-cultural actors on the other hand

(Bencherki, 2017). This vision represents a revolu-

tionary paradigm shift from differentiation between

the agency of intelligent (human and AI) actors and

that of non-intelligent actors. The latter were totally

deprived of agency in older paradigms.

Actor-Network Dualism: ANT rejects the dualism

that tends to separate the social (human) from the ma-

terial (nonhuman). Every individual actor of an actor-

network is considered a network acting together with

this actor. ANT is the theory of actors as networks.

As Latour’s famous maxim goes: “faire c’est faire

faire” – when one acts, others proceed to action

(Bencherki, 2017). Network is a work done by actors,

i.e. by entities who act or undergo an action (Latour,

1996).

Nebular oppositions revised: ANT reconsiders

some fundamental relations and metrics on the net-

works, e.g. (Latour, 1996):

- Far/Close. Physically close elements (when dis-

connected) may appear extremely distant from each

other if we analyze their connections, and vice versa

(cf. Latourian metaphor: “an Alaskan reindeer might

be ten meters away from another one and they might

be nevertheless cut off by a pipeline of 800 miles that

make their mating forever impossible”) (Latour,

1996).

- Large/Small. A network is never “larger” than

any other one. It can only have a more complex to-

pology.

- Inside/Outside. A network is its own border. A

network in ANT has virtually nothing external.

2 TRANSLATION IN

ACTOR-NETWORKS AND ITS

CORRESPONDENCE WITH

PROCESSES IN MAS

Operation of translation in actor-networks is defined

as a delegation of powers of representation from a set

of actors (actor-networks) to any particular (black-

boxed) actor or actor-network in a particular pro-

gramme of actions: A=T(A

,…,A

), where Т is the

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

180

translation of actors A

,…,A

to А. In other words,

actions of actors A

,…,A

(translants) are brought

into being or expressed through representative A act-

ing on behalf of the entire actor-network. “A trans-

lates B” means A defines B. It does not matter

whether B is human or nonhuman, a collectivity or an

individual (Callon, 1991). Operation of translation

equalizes actor-network actions in various space-time

areas and various meta-levels of presentation (e.g.

when behavior of an actor-network is translated

through textual intermediaries: graphs, diagrams, al-

gorithms, formulae etc.).

It appears convenient to use the formalism of TI-

logic of action (Blinov et al., 1991; Von Wright,

1967) to describe the status shift of an actor-network

in the process of translation in the following standard

format:

[A]T([B]I[C]), (1)

where [A] is the initial status of actor(s), [B] is the

next status when translation of the actor(s) is success-

ful, [C] is the next status when translation fails. If we

suppose that all network situations are limited to two

fundamental actor statuses [A] and [B], then the for-

mula:

[~A]&[B]T([A]&[B]I[~A]&[B]), (2)

(like any other of the 64 formulae of this type) reflects

one of the possible “complete” translations. The fol-

lowing two formulae are derivable from (2):

[~A]T([A]I[~A]) (3)

[B]T([B]I[B]), (4)

meaning that the translation: (a) has resulted in situa-

tion [A] and (b) has not “destroyed” situation [B] (i.e.

network situation [B] was allowed to persist). Modal

operator M was introduced in (Von Wright, 1967) as

follows:

M([A]&[B]T([A]&[~B]I[~A]&[B])) (5)

This formula in terms of actor-networks states that the

translation from the initial situation can prevent “de-

struction” of [A] and “destroy” [B], but if the transla-

tion fails the shift from the initial situation of [A] and

[B] may end up in “destruction” of [A] and conserva-

tion of [B]. A successful translation generates a

shared space, equivalence and commensurability. It

aligns. A failed translation means that the players are

no longer able to communicate (Callon, 1991). An ac-

tor-network starts to form as soon as at least three ac-

tors A,B,C are joined together by intermediaries.

There are two possible elementary translation config-

urations (Figure 2) (Callon, 1991):

(a) A  B  C

(b) C

Figure 2: Elementary translation configurations in actor-

networks: (a) complementarity and (b) substitutability (Cal-

lon, 1991).

The first is the transitive configuration of comple-

mentarity: if B=T(A) and C=T(B) then C=T(A). The

second is the binary configuration of substitutability:

C=T(A,B). These two elementary configurations join

together to form longer chains of translations. All

complex networks are built out of these two basic

building blocks (Callon, 1991).

Potential of successful translation or “translatabil-

ity” of an actor is determined by its prescription. Pre-

scription index P(A)∈[0,1] of actor A is a fuzzy esti-

mate of possible actions of actor A from the view-

point of other actors in actor-network AN. More for-

mally, the more complete and determined knowledge

actors AN\A of actor-network AN have in regard to

actor A the higher the value of index P

AN\A

(A). The

less prescribed actors are more easily translatable in

the interest of others, than more rigidly prescribed

ones (Cordella et al., 2003). For any actors A

and A

P(A

)<P(A

) => τ(A

)>τ(A

), where τ(A

)∈[0,1] is the

“translatability” of actor A

: a quantitative metric to

measure ability of A

to be translated.

Any translation is in principle reversible. Irrevers-

ibility of translation means its ability to resist re-

versed translation (retranslation) and competing

translations. The more numerous and heterogeneous

the interrelationships in an actor-network the greater

the degree of network coordination and the greater the

probability of successful resistance to alternative

translations (Callon, 1991).

The process of translation in actor-networks can be

presented as a tuple of consequent operations:

T=<P,I,E,M>, where P is problematisation, I is inter-

essement, E is enrolement and M is mobilization

(translation, in turn, is considered the first phase of

the mediation metaprocess M=<T,C,RBB,D> by

some authors (Latour, 1999), where T is translation,

C is composition, RBB is reversible black-boxing and

D is delegation).

Problematisation: Problematisation is the first

step of translation, because according to Latour every

action wants to solve a problem (Fischer, 2017).

Problematisation can be defined as something that is

indispensable and where one or more key actors try

to define the exact nature of the problem as well as

Agents and Multi-agent Systems as Actor-networks

181

the roles of other actors that could fit with the pro-

posed solution (Silic, 2015).

Problematisation embeds what MAS studies de-

fine as “commitments” of an actor postulated as the

necessity of a chain of actions performed by this actor

towards a predetermined goal in the interests of the

community of actors (Gorodetskii et al., 2010). The

commitments are viewed as pledges to undertake a

specified course of action providing a degree of pre-

dictability so that actors can take the (future) activi-

ties of others into consideration when dealing with in-

terdependencies of actors, global constraints or re-

source utilization conﬂicts (Jennings, 1993).

Interessement: Synonymous with interposition

this stage of translation process represents a group of

actions where an actor tends to impose and stabilize

identities of other actors determined at the problema-

tisation stage. To make a group of actors “interested”

means to create a “virtual device” (Callon, 1986),

which can be placed between this group and all other

entities who strive to re-identify this group. In other

words, А

makes A

“interested” when it breaks or

weakens all links between A

and group

,…,A

of n-2 actors who may tend to liaise

with A

. Figure 3 schematically demonstrates the pro-

cess of interessement of actor B by actor A in actor-

network AN={A,B,C,D,E} (Callon, 1986).

Figure 3: Interessement of actor B by actor A in actor-net-

work AN={A,B,C,D,E} (Callon, 1986).

Successful outcome of the interessement confirms

(more or less completely) the efficiency of the prob-

lematisation and supposed actor alliances (such het-

erogeneous alliances are often viewed as human-non-

human quasi-objects or hybrids in ANT texts). Be-

sides, this stage tries to break all competitive liaisons

and build a system of alliances within actor-network.

At this stage the socio-technological communities are

formed and fixed.

This stage of the translation process embeds “con-

ventions” discussed in MAS studies. According to the

MAS principles conventions fix the conditions of ful-

filment/rejection of obligations by an actor (Go-

rodetskii et al., 2010). “All coordination mechanisms

in MAS can ultimately be reduced to (joint) commit-

ments and their associated (social) conventions” (Jen-

nings, 1993). The interessement goes further: in terms

of MAS it is driven by the actor’s intention to

weaken/break other actors’ commitments/conven-

tions and thus create new conventions with them to

achieve a particular goal.

Enrolement: The core function of the enrolement

is the determination and coordination of roles of ac-

tors aiming at creation of a steady network of alli-

ances (Silic, 2015).

This stage of translation is also interconnected

with the MAS concepts where one of the forms of ac-

tor obligations is the role accepted by or assigned to

an actor (Gorodetskii et al., 2010). The notion is that

actors have general roles to play in the collective ef-

fort, and by using knowledge of these roles the actors

can make better interaction decisions.

This notion can be explicitly manifested in an or-

ganizational structure, which defines roles, responsi-

bilities and preferences for the actors within a coop-

erative society, and thus in turn defines control and

communication patterns between them (Durfee,

1999).

Mobilization: Through step-by-step appointment

of representatives and establishment of a series of

equivalences heterogeneous actors are moved and

then “reassembled” at a new place/time. This stage

completes translation and certain actors start acting as

representatives (delegates) of other actors (Callon,

1986). The “evolution indicators” were introduced in

(Latour et al., 1992) to measure the progress of mobi-

lization:

= A

n-1

+ A

, (6)

where S

is the number of associated elements at step

n of the translation; A

n-1

is the number of allies re-

tained from the previous step; A

is the number of the

newly “recruited” actors.









(7)

where IN

is the negotiation index. High value of IN

indicates that the project represented by actor-net-

work must be extensively renegotiated (Latour et al.,

1992).

The mobilization concept of ANT can signifi-

cantly invest into understanding of the team behavior

of actors in MAS (where it is considered as something

more than just a set of coordinated individual actions

of the actors) (Gorodetskii et al., 2010). The mobili-

zation concept suggests new semantics for transla-

tion-like effects discussed in the framework of MAS

studies where a collection of actors needed to accom-

plish a task frequently includes humans who have del-

egated tasks to the [nonhuman] actors and/or humans

who will be performing some of the work, and

(hence) it is essential that the functions being offered

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

182

by the actor communication language be common

across the language of intelligent [nonhuman] actors

and the language that people will use to communicate

with them (Cohen et al., 1995). The new vision to-

wards agency as a mere effect of interaction of heter-

ogeneous actors regardless of their nature and inher-

ent intelligence (or absence thereof) introduced in

ANT (generalized symmetry principle) may thus help

reconsider and enrich coordination scenarios dis-

cussed in the agent-oriented methods.

3 CONCLUSIONS

This paper introduces the actor-network paradigm

and discourse to the world of agent based modelling.

Some core concepts of ANT (e.g. generalized sym-

metry principle) represent a brand new vision towards

agency and interactions in heterogeneous multi-agent

communities. Correlation between certain ANT and

MAS concepts makes their potential assemblage (en-

riched with approaches and formalisms provided, in-

ter alia, by applied semiotics and action logics) a pro-

spective tool for use in agent based models of socio-

technological systems, including but not limited to:

intelligent logistics; global business networks; com-

plex research, engineering, industrial and construc-

tion projects; urban and regional governmentality; in-

formation security management systems; human-

nonhuman communities encapsulated in space sta-

tions, human-nonhuman interactions in multimedia

arts and many others. As an object-oriented semiotic

tool actor-network theory provides an approach to the

analysis of connections between the information in

the form of texts and meta-texts (documents, con-

tracts, messages, scripts, protocols etc.) circulating in

a socio-technological system, and situations caused

by the texts and meta-texts. This approach in our

opinion has a good potential in data and information

security studies (Iskanderov et al., 2019): investiga-

tion of dependencies between the level/quality of pro-

tection of the texts moved across the network and the

network situations caused by protected/unchanged

texts on the one hand, and deliberately or arbitrarily

changed texts on the other hand. Unfortunately the

space limits of this paper do not allow for more de-

tailed discussion which will necessarily be continued

in future texts. Our paper aims at bringing attention

of the AI researchers, MAS theorists, human-machine

systems engineers, ergonomists, knowledge engi-

neers, logistics specialists and broader research com-

munity to the actor-network paradigm and its applied

potential in socio-technological systems research.

REFERENCES

Latour, B., 1994. On Technical Mediation: Philosophy, So-

ciology, Genealogy // Common Knowledge. Vol. 3. №

2. P. 29–64.

Latour, B., 1996. On Actor-Network Theory. A Few Clari-

fications Plus More than a Few Complications // Soziale

Welt. 1996. Vol. 47. P. 369–381.

Latour, B., 1999. Pandora’s hope: Essays on the reality of

science studies. Cambridge, MA: Harvard University

Press.

Callon, M., 1991. Techno-economic networks and irrevers-

ibility. In J. Law (Ed.), A sociology of monsters (pp.

132-161). London: Routledge.

Callon, M., 1986. Some Elements of a Sociology of Trans-

lation; Domestication of the Scallops and the Fisher-

men of St.Brieuc Bay // Power, Action and Belief / J.

Law (ed.). L.: Routledge, 1986. P. 196-223.

Silic, M., 2015. Using ANT to understand dark side of com-

puting – Computer underground impact on eSecurity in

the dual use context. University of St Gallen. Institute

of Information Management. January 2015.

Cordella, A., Shaikh, M., 2003. Actor-network theory and

after: what's new for IS research. In: European confer-

ence on information systems, 2003-06-19 - 2003-06-21.

Gorodetskii, V.I., Karsayev, O.V., Samoylov, V.V.,

Serebryakov, S.V., 2010. Applied multiagent systems

of group control. Scientific and Technical Information

Processing. 2010. T. 37. No.5. C. 301-317.

Jennings, N., 1993. Commitments and Conventions: The

Foundation of Coordination in Multi-Agent Systems,

The Knowledge Engineering Review, 8 (3), 1993, С.

223-250.

Durfee, E., 1999. Distributed Continual Planning for Un-

manned Ground Vehicle Teams. AI Magazine

20(4):55-61, Winter 1999.

Cohen, P., Levesque, H., 1995. Communicative Actions for

Artificial Agents, In Proceedings of ICMAS’95, 1995,

С. 65-72.

Latour, B., Mauguin, P., Teil, G., 1992. A Note on Socio-

Technical Graphs. Social Studies of Science (SAGE,

London, Newbury Park and New Delhi), Vol.22 (1992),

33-57.

Fischer, E., 2017. Socio-Technical Innovations in Urban

Logistics: new Attempts for a Diffusion Strategy. 16

Conference on Reliability and Statistics in Transporta-

tion and Communication, RelStat’2016, 19-22 October,

2016, Riga, Latvia. In Procedia Engineering 178

(2017) 534-542.

Balzacq, T., 2016. A theory of actor-network for cyber-se-

curity. European Journal of International Security, p.1-

tional Studies Association 2016.

Kiselev, G.A., Panov, A.I., 2018. Sign-based Approach to

the Task of Role Distribution in the Coalition of Cog-

nitive Agents. SPIIRAS Proceedings. 2018. Issue 2(57).

ISSN 2078-9181 (print), ISSN 2078-9599 (online), pp.

161-187. DOI 10.15622/sp.57.7

Shirokov, A.A., 2019. The Politics of Explanation and

Strategy of Description of Bruno Latour: How to Write

Agents and Multi-agent Systems as Actor-networks

183

In-fra-reflexive Texts. RUSSIAN SOCIOLOGICAL

REVIEW. 2019. VOL. 18. NO 1. pp. 186-216.

Blinov, A.L., Petrov, V.V., 1991. Elements of Logic of Ac-

tion. NAUKA Publishers. 1991. ISBN 5-02-008150-

7Von Wright, G.H., 1967. The Logic of Action – a

sketch // The logic of decision and action. Pittsburgh.

P. 121-139.

Deleuze, G., Guattari, F., 1993. A Thousand Plateaus. Min-

neapolis: University of Minnesota Press. ISBN 0-8166-

1402-4.

Iskanderov, Y., Pautov, M., 2020. Actor-Network Ap-

proach to Self-organisation in Global Logistics Net-

works. In: Kotenko I., Badica C., Desnitsky V., El Baz

D., Ivanovic M. (eds) Intelligent Distributed Computing

XIII. IDC 2019. Studies in Computational Intelligence,

vol 868. Springer, Cham

Bencherki, N., 2017. Actor-Network Theory. In Craig Scott

& Laurie Lewis (eds.), The International Encyclope-dia

of Organizational Communication. New York, NY:

Wiley.

Iskanderov, Y., Pautov, M., 2019. Security of Information

Processes in Supply Chains. In: Abraham A., Kovalev

S., Tarassov V., Snasel V., Sukhanov A. (eds) Proceed-

ings of the Third International Scientific Conference

“Intelligent Information Technologies for Industry”

(IITI’18). IITI'18 2018. Advances in Intelligent Systems

and Computing, vol 875. Springer, Cham

Kiselev, G.A., Panov, A.I., 2017. Synthesis of the Behavior

Plan for Group of Robots with Sign Based World

Model. In: Ronzhin A., Rigoll G., Meshcheryakov R.

(eds) Interactive Collaborative Robotics. ICR 2017.

Lecture Notes in Computer Science, vol 10459.

Springer, Cham

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

184