The Institutional Stance in Agent-based Simulations

Giovanni Sileno, Alexander Boer and Tom Van Engers

Leibniz Center for Law, University of Amsterdam, Amsterdam, The Netherlands

Keywords:

Normative Systems, Story Animation, Institutions, Agent-roles, Agent-based Simulation, Jural Relations,

Legal Knowledge, Non-compliance, Multi-Agent Systems.

Abstract:

This paper presents a multi-agent framework intended to animate scenarios of compliance and non-compliance

in a normative system. With the purpose of describing social human behaviour, we choose to reduce social

complexity by creating models of the involved agents starting from stories, and completing them with back-

ground theories derived from common-sense and expert knowledge. For this reason, we explore how an

institutional perspective can be taken into account in a computational framework. Roles, institutions and

rules become components of the agent architecture. The social intelligence of the agent is distributed to sev-

eral cognitive modules, performing the institutional thinking, whose outcomes are coordinated in the main

decision-making cycle. The institutional logic is analyzed from a general simulation perspective, and a con-

crete possible choice is presented, drawn from fundamental legal concepts. As a concrete result, a preliminary

implementation of the framework has been developed with Jason.

1 INTRODUCTION

In the last decades, software engineering has been

moving away from machine-oriented views of pro-

gramming towards concepts and abstractions more

closely reﬂecting the way in which humans conceive

the world. Multi-agent systems have been introduced

to develop complex services, typically strongly char-

acterized by interactions between entities, but, differ-

ently from other distributed systems, they often inte-

grate concepts derived from philosophy or psychol-

ogy, like the Belief-Desire-Intention paradigm. Fur-

thermore, multi-agent systems are successfully used

for simulations in economics, sociology, and biology

(in that case the term ABM - Agent-based modelling

is preferred). However, although since a few years

there has been an active research interest on norma-

tive multi-agent systems, there is no equivalent de-

velopment about ABM simulations from a legal per-

spective. Although there are interesting overlaps with

these studies there is an intrinsic, and important, dif-

ference of perspective. With the ﬁrst approach, con-

ception, we design and create artefact systems, which

are largely under our control, trusted, and highly pre-

dictable (for example, electronic institutions). With

the second approach, interpretation or discovery, we

interpret the relevant events against a background of

See for example in (Batten, 2000; Duffy, 2006).

systems, in our case mostly of human origin, some

of which we trust to work in certain ways, and some

others we only guess at. The second is the arena

where ABM actually plays a role, and where this

study should be placed.

2 AGENTS, INSTITUTIONS,

ROLES

A (soft) Evolution. In the computational world, all

starts (and ends) with imperative commands in the

form of instructions. Instructions can be grouped in

procedures. Data passes from numeric to string, uni-

ﬁed in arrays, structures, and ﬁnally instantiated in

objects, elements of classes deﬁned by data struc-

tures and methods handling that data. Objects do not

perform any action independently, they can only be

used

. An agent is instead a proactive entity (agere

in Latin means to do, to achieve, to lead). The basic

type of agents is the behavioural agent: it has appar-

ent autonomy, executes a plan composed by actions.

The next step in this evolution is brought by inten-

tional agents, that show autonomy, proactiveness, re-

activeness and social ability. These four character-

A practical deﬁnition in object-oriented programming

may be different.

255

Sileno G., Boer A. and Van Engers T..

The Institutional Stance in Agent-based Simulations.

DOI: 10.5220/0004257602550261

In Proceedings of the 5th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2013), pages 255-261

ISBN: 978-989-8565-38-9

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

istics are the references of a very important human

capability: mentalization (Fonagy and Target, 1997).

Humans are able to mentalize, i.e. to create an inter-

nal representation of an external agent, in a way that

its behaviour can be predicted and, according to the

intentional approach, explained in terms of attitudes,

for example using concepts like beliefs, desires and

intentions

. This is an innate exercise of modelling

for humans, and its importance comes from the fact

of being commonly used as representation of reality.

As a matter of fact, humans tend to mentalize not only

individuals, but also communities, cultures, organiza-

tions, nations.

Instructions, Habits, Rules. Every human individ-

ual or group, having a goal, sets the more adequate

means, or better, those considered as the more ade-

quate means to achieve it. The means/end relation-

ship creates generally an instruction of conduct of that

type: “If you want to reach the goal A, you have to

do the action B”. All these instructions could be very

different in aim, content, scope of validity, or subjects

involved. Moreover not all the instructions are estab-

lished in the previous form (to do), but also in a qual-

iﬁcation or state discriminant form (to be). If the for-

mer type requires the capability of executing tasks and

in a larger sense of planning, the latter subsumes the

existence of another process: the situation (or state)

recognition. Anyway, all instructions have something

in common: they are propositions that aim to inﬂu-

ence the behaviour of individuals and/or groups, lead-

ing their actions towards certain objectives more than

others. In a social agglomerate, similar instructions

could be part of common knowledge, or cultural con-

ventions, social habits and social rules. Only effective

legal norms become part of the social rules.

Choice of Conduct. Recognition of the present

state, and then planning tasks towards a goal are parts

of a typical problem solving cycle (Breuker, 1994),

that could be seen as the main core of the agent activ-

ity. The choice of conduct is then a sort of preference

order between possible plans of actions, considering

eventually associated economic costs or some poten-

tial side-effects. But how are norms related to the

choice of conduct? (Neumann, 2010) has found in the

normative multi-agent literature two main philosoph-

ical approaches: deontic and consequentialistic. The

ﬁrst considers norms as cognitive objects. A deontic

conceptualization of norms typically emphasises that

norms are in itself a reason for action. On the other

The intentional stance has been introduced by Dennett

(Dennett, 1987).

brute

reality

institutional

reality

brute fact

institutional fact

}

institutional

rule

constitutive

rules

normative

rule

constitutive

facts

Figure 1: Institutional perspective.

side, with a consequentialistic approach, agents con-

sider the possibility of breaking the norm as one of

the alternative choices. Stated differently, the latter

treats norms as behaviour regularities, and, for this,

they may be described for example with game theo-

retic models

. The two different perspectives result

in two potential attitudes towards norms; our target

ABM framework should integrate both of them.

Institutions. In (Searle, 1969) an interesting point

is traced about the difference between normative (or

regulative) and constitutive rules. The former regu-

lates existing forms of behaviour. For example, the

rules of polite table behaviour have been introduced

later than the eating activity. Eating existed and ex-

ists anyway, with or without these rules. But, at the

contrary, the rules of playing chess have created the

possibility of playing chess. Searle extended these

examples. “The institutions of marriage, money and

promising are like the institutions of baseball and

chess in that they are systems of such constitutive

rules or conventions.”

Generalizing, we can say that an institution is an

intentional social collective entity (Boer, 2009), de-

ﬁned by certain rules and some institutional facts. It

is collective and intentional, simply because a group

of people recognizes and intends its existence. This

concept of institution uniﬁes games, social informal

norms and legal norms

From a game theory point of view, a social rule elim-

inates certain strategies for each of social convention the

agents, and thus induces a sub-game. “For a given agent,

a social law presents a trade-off; he suffers from loss of

freedom, but can beneﬁt from the fact that others lose some

freedom.” (Shoham and Leyton-brown, 2009)

According to the institutional perspective, law is an

institution whose purpose is to create normative order via

formalization (MacCormick, 1998). Legal facts are de-

pendent on legal norms and observed through legal norms.

Law’s ‘truth’ is different from reality, psychological or so-

cial truth. “In court proceedings, the ‘truth’ of the facts is

ultimately determined not by criteria employed in empirical

sciences but by those provided by procedural and substan-

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

256

A general institutional perspective has been visu-

alized on Fig. 1. Two realities coexist: a brute reality

and an institutional reality. Filtered by the constitu-

tive rules associated to the institution, only few of all

real facts determine an institutional fact. Stated dif-

ferently, these real facts are events triggering a change

in the institutional reality. They will be called consti-

tutive facts, and typically are created through a consti-

tuting act. Institutional rules create new institutional

facts from existing institutional facts. Finally, norma-

tive rules convert institutional facts into instructions

of behaviour, and if the choice of conduct of the agent

goes accordingly, they perform an actual normative

action.

Institution as Agents. As pointed out in (Boer and

Van Engers, 2011) externalization of rules and social

organization is philosophically defensible, and often

an efﬁcient solution from a system engineering point

of view. However, for ABM simulation purposes, this

is not a natural choice. During a theory construction

session with legal experts, it is more natural to think

of the agent as provided with an internalized artefact,

of arbitrary complexity, embedding the social factor.

Following the terminology used in (Chu, 2011), we

consider that there is no institution “an sich” in the

brute reality. It exists only in the form of conceptual-

ization, internal to each agent. The collective ﬂavour

of this type of entity is given by a similar represen-

tation in each individual of the community, or, better,

may be regarded as an instance of emergence. At a

macro-level, institutions arise, evolve and function as

patterns of social self-organization, which goes be-

yond the conscious intentions of the individual hu-

mans involved, but still, it is generated by a micro-

level, individual behaviour.

From the agent’s point of view, to be comply-

ing with an institution means to have a representa-

tion of the present institutional state (rules and facts)

and behave accordingly with the normative rules gen-

erated by reasoning with it. This is a speciﬁc form

of the equation T hinking = Knowledge + Reasoning

(Kowalski, 2010). This kind of modularization un-

veils an architectural possibility: to consider the insti-

tution itself as a cognitive agent, dividing the model

in separate but interconnected sub-models.

Scenarios and Roles. Humans continuously learn

about the social institutions around them. The law

tells us that a sale involves a buyer and a seller, that it

deploys an offer and an acceptance, and that it leads

tive legal norms” (Tuori, 2006). This observation could be

extended also to informal institutions.

rule 1

institution A

role X

scenario

agent

role Y

role Z

rule 2

rule 3

institution B

institution C

Figure 2: Architectural components.

to an obligation on the buyer to pay, and an obligation

on the seller to deliver. It does not entail that the seller

and buyer feel harmed if the other party does not meet

their obligations, it does not tell what they will do if

that happens, it does not tell when an offer will be

acceptable, but we do have expectations about these

things. In fact, analyzing a certain context, we pro-

vide a plausible (at least, according to us) intentional

model of the involved agents, using our knowledge of

the domain, analogies with other experiences and our

mentalization capability.

In doing this, we use prototypes of normal be-

haviours (on which we will plausibly base our own

behaviour, at the place of that agent), and also pro-

totypes of the behaviours of non complying, faulty

agents. Almost without noticing, we are not talk-

ing any more about individual agents. Identifying the

contextual common patterns between individual cases

(and their rationality), we are writing down typical

scenarios (as patterns of social interaction) with the

typical (social) roles that agents play in them. Such

roles are associated to certain beliefs, plans (result-

ing in actual actions) and goals. Nothing forbids an

agent to play simultaneously several roles (Fig. 2)

and viceversa.

3 INSTITUTIONAL THINKING

Constitutive and institutional rules could be expressed

in the form of normative conditionals (Sartor, 2006):

i f CONDIT IONS then

CONCLUSION. Conditions

are constitutive and institutional facts. The conclu-

sion is a new institutional fact. Institutional facts con-

cern institutional entities or jural relations. Every new

institutional fact typically draws a new conﬁguration

of the jural relations between the agent entities exist-

ing in that institution. This new conﬁguration is a re-

sult of the modus ponens or forward reasoning applied

through some deontic rules. We have taken as a start-

ing point the fundamental legal concepts elaborated in

TheInstitutionalStanceinAgent-basedSimulations

257

(Hohfeld, 1917), completed and further developed in

(Sartor, 2006), but another choice could be possible

as well.

obligative

right

obligation

no-right privilege

correlative

opposite

action

power

subjection

disability immunity

correlative

opposite

Figure 3: The two Hohfeldian squares: directed obligations

and power concepts.

Directed Obligations and Permissions. Simplify-

ing, the ﬁrst Hohfeldian square deﬁnes the directed

obligations between two actors, and it introduces four

different deontic concepts: obligative-right, obliga-

tion, privilege and no-right (Fig. 3). An obligative-

right, strictly deﬁned, is one’s enforceable claim

against another (progressive form). Privilege is one’s

freedom from the claim of another one (regressive

form). Like the cited author, we added to these

concepts the permissive-right, different to obligative-

right. This is in fact not completely reducible to the

idea of a directed obligation.

Power. The second Hohfeldian square refers to the

concept of power, with four new concepts: action-

power, subjection, disability, immunity (Fig. 3).

Action-power consists in the agent’s power to deter-

mine a normative position involving a certain agent

performing a certain action. On the contrary, disabil-

ity deﬁnes the impossibility (no-power) to achieve a

certain normative position.

Following the analysis found in (Vatiero, 2010),

we extend their use also to not strictly legal aspects.

In the case of the buyer and seller, if the buyer is com-

pletely dependent to the seller because he is the only

one in possession of the good he wishes, he is actu-

ally in a subjection position towards the seller for his

pricing choices. On the other hand, the seller is in a

power position and can eventually raise the price. For

the buyer, the only way to avoid this subjection is to

stop wishing that good (which is not always possible),

or ﬁnding another seller with a better price.

Generalizing this example, an action (in the form

of an investment) of the agent may bring about a

change of his institutional position. At the same time,

contextual factors may determine an endogenous en-

forcement of power: the fact that the buyer needs that

good and that the seller is the only one to sell deter-

mines the power positions between the two agents,

and this is not inﬂuenced by the agents’ consciousness

of these positions. However, renewing the Bacon’s

knowledge is power, an agent that is able to recognize

the contextual state, and knows how to change it ac-

cordingly to his own interests, has the action-power to

modify the power positions in the institution.

Coordination and Competition Dimensions. The

agent behaviour associated to an institution could be

generally ascribed along two main axis: (social) co-

ordination and (economic) competition. Simplifying,

the ﬁrst covers the rules of the “game”, the second the

knowledge about how to “play the game” within the

rules. The rules of fair competition are particularly

meaningful. They are introduced in order to avoid

that one player attains a position too strong compared

to the others, limiting or blocking de facto their pos-

sibilities of action and resulting in a failure or degen-

eration of the institution.

Practical Normative Indications. In order to have

a successful normative action, the jural relations re-

sulting from the institution should have some con-

sequence on the agent’s behaviour. Focusing on the

decision-making cycle, only four practical normative

indications are important for the agent: you can

you can not *, you must *, you do not have to *.

In general, * may be related to actions (to do),

or states (to be), or results (to bring about). The

state subsumes a state recognition, i.e. the possibil-

ity that a certain group of perceived/believed condi-

tions counts as a certain state. For this reason, deﬁn-

ing a fact as a reiﬁed statement about something, a

state/situation is a second order fact, requiring a qual-

iﬁcation. The bring about requires a plan toward the

objective/result, and a result can be both a fact or a

state.

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

258

constitutive facts

institution

practical indications

you can *

you can not *

you must *

you do not have to *

constitutive facts

to be brought about

request for change

how to be allowed to *

how to be allowed not to *

how to be obliged to *

how to be obliged not to *

forward-chaining

backward-chaining

institution

Figure 4: Forward and backward reasoning in institution.

Institutional Backward Reasoning. The process

analyzed until now, from constitutive facts to practical

indications, is a forward-chaining procedure (condi-

tion to conclusions). It is interesting to consider also

the other way around, the backward-chaining (con-

clusion, if condition): which conditions have to be

brought about to reach a different institutional conﬁg-

uration? More answers might be possible to the same

question, in the form of alternative sub-goals aiming

to the same objective (Fig. 4).

If the logic describing rules and jural relations

is consistent, forward and backward chaining rela-

tions can be automatically inferred one from the other.

However, here we emphasize the possibility to sim-

ulate in such a framework also a wrong conceptu-

alization of the institution, conﬂicting rules, faulty

logic, using for example a non-monotonic approach

and handling separately the forward and backward re-

lations

Normative Plans. Put in a simple form, in order to

be complying, an agent should: (a) check, before per-

forming an action, if he has the permission to do that,

and if not, evaluate to drop the plan involving that ac-

tion, (b) add as goal the performance of any new obli-

gation. On the other side, in order to be socially reac-

tive in a context in which non-complying behaviours

are possible, an agent should: (c) be able to recognize

not complying events or situations

, (d) have a main-

tenance plan with a monitoring purpose, (e) react to

the occurrence of not complying events.

In this context can has only the meaning of being al-

lowed to (and not, for example, of being able to).

Here we are considering * deﬁned always in a positive

form, but it is easy to demonstrate that the negative forms

can be reduced to one of these. Furthermore, all four po-

sitions could actually be expressed using only one functor,

nevertheless we prefer to use the proposed expressions, as

they are more pertinent to focus on behaviour.

Starting from the famous Wason’s selection task, sev-

eral studies have been conducted in behavioural psychology

with the purpose of assessing the heuristics involved in hu-

man reasoning. Part of these results, relevant for our scopes,

are described in (Cosmides and Tooby, 2008).

Using the terms previously deﬁned, an event is a fact,

observed or acknowledged by the agent, and a situation is a

state, in the sense of fact of facts.

Problem Solving Cycle of the Agent. From the

point of view of the modeller, it is worth to con-

sider the possibility to associate different attitudes

towards norm to each institutional component. Fol-

lowing the consequentialistic approach the agent will

consider plans containing violations of norms. Fol-

lowing the deontic approach he will abolish any plan

that includes non-complying actions or results. In

both cases, some kind of meta-reasoning about plans

has to be implemented. For this reason, this inves-

tigation would be part of the more general conﬂict

resolution problem for norm vs desire conﬂicts, to be

analyzed altogether with plan vs plan (when multiple

alternative plans are possible to reach the same goal),

desire vs desire conﬂicts (caused for example by con-

ﬂicting goals between simultaneous roles), norm vs

norm (between institutions) and rule vs rule conﬂicts

(for an internal institutional conﬂict)

. From a gen-

eral point of view, we are not aiming for autonomous

agents that create their own plans from ﬁrst princi-

ples, but that evaluate and choose between existing

plans according a given rationality.

4 PRELIMINARY

IMPLEMENTATION

An implementation of the framework has been devel-

oped for Jason, a MAS platform based on an exten-

sion of the AgentSpeak(L) programming language

(Bordini et al., 2007). The source code is available on

our site

and consists in different modules handling

the creation of institutions, the institutional logic, the

institutional communication and a coordination so-

lution that follows a deontic approach towards nor-

mative indications. At the moment, two examples

are included, not presented here in details for space

constraints. The ﬁrst one describes a mythological

story of non-compliance (Achilles avoiding the Tro-

jan war): an example of how a story could be ani-

mated, starting from a narrative, interpreting it, and

adding background theories derived from common-

sense knowledge. The second one is an institutional

model of the sale process, as described in common

Many of these conﬂicts are consequence of the distribu-

tion of the the institutional thinking to multiple entities, thus

requiring adequate coordination functions. However, this is

also a good representation of our daily experience about the

choice of conduct, when we have to take choices between

our social roles expectations (for example as members of a

family, researchers, citizens, etc.).

Jason is based on logic programming and implements

a BDI architecture for cognitive autonomous agents.

http://justinian.leibnizcenter.org/inst jason

TheInstitutionalStanceinAgent-basedSimulations

259

law/the Dutch civil code, backing several buyer/seller

scenarios.

5 CONCLUSIONS AND FURTHER

DEVELOPMENTS

The intent underlying our research is to (re)connect

normative (including legal) reasoning with other

forms of reasoning. Particularly, we are interested in

the role that (legal) norms play in social structures and

how norms inﬂuence human behavior in those struc-

tures.

In the current approach, typical strategy decision

problems for a given game do not take explicitly into

account the possibility of the player to behave avoid-

ing a rule, or forcing the interpretation of the rule to-

ward its interest, if the regulator (consciously or not)

left some ambiguity. The second case is not so com-

mon in games, but in the case of legal order, lacunae

of law are practically unavoidable and within limits,

desirable. This is part in fact of the human collec-

tive adaptation and social reasoning capabilities. In

that way, humans question rules, both directly or in

an involuntary way (for example, in the case of lack

of knowledge or misunderstanding of the rule) and de-

termine with their actions if social rules are successful

or not in their normative intentions. Furthermore, hu-

mans do not play ever a single “game” at once. In a

broader sense, humans are always players of several

games simultaneously, or to put it differently, they are

agents concerned in the same moment by many dif-

ferent institutions, sometimes conﬂicting, created by

habits, social rules and legal order.

In the present paper we propose a framework that

aims to take explicitly all of this into account. Our ob-

jective is a partial alignment of the representations of

law with actual social structures and existing imple-

mentations of law. Descriptions of those are present

for example in legal narratives, in the form of court

decisions or anecdotes by legal experts, where a con-

structed theory, at least partially, is explicitly stated.

Thus, using our framework, models of agents or roles

involved in a social scenario could be animated, out-

lined from a story, enriched with knowledge from

experts and/or referring to the sources of regulation,

with the possibility of integrating game-theoretic be-

havioural theories. As operative result, such a sim-

ulation would furnish a support to understand the so-

cial (institutional) dynamics: validating the domain of

conceptualization of the experts, making predictions,

suggesting improvements to regulations.

Along with this paper, a preliminary implementa-

tion has been developed, using an existing multi-agent

system platform. Although successful, this experi-

ence showed the necessity of creating (or extending)

a platform with an explicit ABM approach in order

to attain a full computational deployment of the pro-

posed framework. This is one of the directions of our

future research.

REFERENCES

Batten, D. F. (2000). Discovering Artiﬁcial Economics.

Westview Press.

Boer, A. (2009). Legal Theory, Sources of Law and the

Semantic Web. IOS Press.

Boer, A. and Van Engers, T. (2011). An Agent-based Legal

Knowledge Acquisition Methodology for Agile Pub-

lic Administration. ICAIL 2011: The Thirteenth In-

ternational Conference on Artiﬁcial Intelligence and

Law, June.

Bordini, R. H., H

ubner, J. F., and Wooldridge, M. (2007).

Programming multi-agent systems in AgentSpeak us-

ing Jason. John Wiley & Sons Ltd.

Breuker, J. (1994). Components of problem solving and

types of problems. A Future for Knowledge Acquisi-

tion, 867:118–136.

Chu, D. (2011). Complexity: against systems. Theory in

biosciences, 130(3):229–45.

Cosmides, L. and Tooby, J. (2008). Can a General Deontic

Logic Capture the Facts of Human Moral Reasoning?

How the Mind Interprets Social Exchange Rules and

Detects Cheaters. In Sinnott-Armstrong, W., editor,

Moral psychology, pages 53–119. MIT Press, Cam-

bridge.

Dennett, D. C. (1987). The Intentional Stance. MIT Press,

Cambridge, Massachusetts, 7th edition.

Duffy, J. (2006). Agent-based models and human subject

experiments. Handbook of computational economics,

2(05):949–1011.

Fonagy, P. and Target, M. (1997). Attachment and reﬂective

function: their role in self-organization. Development

and psychopathology, 9(4):679–700.

Hohfeld, W. N. (1917). Fundamental legal conceptions as

applied in judicial reasoning. The Yale Law Journal,

26(8):710–770.

Kowalski, R. A. (2010). Computational Logic and Human

Thinking: How to be Artiﬁcially Intelligent. Number

November. Cambridge University Press.

MacCormick, N. (1998). Norms, institutions, and institu-

tional facts. Law and Philosophy, 17(3):301–345.

Neumann, M. (2010). A classiﬁcation of normative archi-

tectures. Simulating Interacting Agents and Social

Phenomena, 7:3–18.

Sartor, G. (2006). Fundamental legal concepts: A for-

mal and teleological characterisation. Artiﬁcial Intel-

ligence and Law, 14(1):101–142.

Searle, J. R. (1969). Speech acts: An essay in the philosophy

of language. Cambridge University Press.

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

260

Shoham, Y. and Leyton-brown, K. (2009). Multiagent sys-

tems: Algorithmic, game-theoretic, and logical foun-

dations. Cambridge Univ Pr.

Tuori, K. (2006). Self-description and external description

of the law. No Foundations, 2:27–43.

Vatiero, M. (2010). From W. N. Hohfeld to J. R. Commons,

and Beyond? Journal of Economics, 69(2).

TheInstitutionalStanceinAgent-basedSimulations

261