=Paper=
{{Paper
|id=Vol-1126/paper4
|storemode=property
|title=Shared Knowledge and Interpretation of Norms in Multi-Agent Systems
|pdfUrl=https://ceur-ws.org/Vol-1126/paper4.pdf
|volume=Vol-1126
|dblpUrl=https://dblp.org/rec/conf/aiia/Antonini13
}}
==Shared Knowledge and Interpretation of Norms in Multi-Agent Systems==
https://ceur-ws.org/Vol-1126/paper4.pdf
Shared knowledge and interpretation of norms in
multi-agent systems
Alessio Antonini
Department of Informatics, Università degli Studi di Torino,
antonini@di.unito.it
1 Introduction
The representation of norms is a central challenge in AI&LAW and in normative multi-
agent systems [1]. Two requirements from AI&LAW are that the representation is real-
istic in the sense of being founded in legal practice1 , and to build more realistic agents.
In particular, the representation of norms has to address the following questions:
1. How to represent the ontological status of norms?
2. How to represent open norms and vagueness?
3. How to represent agent’s interpretation of norms?
In multi-agent systems, norms are usually considered rules that can be violated by
agents. The reasons for norm violation are usually bounded to agents utility evaluation
- gain versus possible fee - but with our proposal we depict a scenario in which norms
are open or vague. Agent’s can interpret norms in different ways, according to their
knowledge but also according with their own interpretation theory of general concepts,
norm goal, etc.
The rest of the paper is organized as follows. In Section 2, we introduce our method-
ology, Section 3 shows the features of norms as social objects. In Section 4, we discuss
the ontological status of norms and their interpretation process. Finally in Section 5 we
present some final remarks.
2 The problem: beyond the rule representation of norms
Part of the current research in AI&Law is based on the strong assumption that norms
can be model as rules and that such rules can be extracted from the legal text only. Thus,
rules are typically considered as given and from this assumption a formal theory of law
in terms of rule reasoning has been given with the formal means at disposal up to now.
Researchers in AI&Law are aware that “law is not a matter of simply applying rules
to facts via modus ponens” [2] but most of the current techniques continue to be based
on the same assumptions [3, 4]. Despite the advances in logic and case-based systems,
“these ideas have not achieved their full potential” [5].
Looking at legal norms in legal practice, norms are not exhausted by rules extracted
from the legislative texts, norms emerge from the whole legislative system and society.
1 In order to answer to the current request of transparency of law processes.
2 Antonini
2.1 Social objects
As methodology in our ontological analysis we use the concept of social object in the
sense of Ferraris [6]. A social object in this sense represents both the physical and
immaterial side of the social constructs, the inscriptions but also a set of shared beliefs.
A social object [6] is a class of entities between concepts and physical objects. A
social object follows the rule Object = Inscribed2 Action that represents the transfor-
mation of an agent’s action into an independent object. Those objects are accessible by
inscriptions, like papers, drawings, digital or human memory, their interpretations are
shared by a group of people and detached from the will of their creators. A social object
is an instance of a common model (like documents) that defines the expected/required
content, a communicative [7] speech act [8], within a context, i.e. within an existing
system of relations and references.
A public and communicative action can create a new social object in each context
it is posed, i.g., each event in witch an act is performed can create a new social object.
A social object is shattered in all participants’ mind, it depends on the memories of
the event (shared) and on the interpretation they elaborate (personal). Social objects
posses both physical and intangible features: the structure of their inscriptions (physical
objects) and the structure of their meaning (mind images and connections of concepts).
For that reason, every social object is unique even if they are instance of known models.
To give an intuition of social object models, we can consider as example the features
of “personal statements”: any personal statement involves several things like the author
mental state, the declaration body, the knowledge shared by the recipes, a time and
place of the statement, the traces left, pragmatic games and social behaviour, the social
status of anyone involved and their mutual expectations.
Agents’ actions creating social objects are communicative and public. Those com-
munications can be represented as graph of relations among entities like common con-
cepts and physical objects [9]. The meaning of those objects is given by their interpre-
tation: the manipulation their graphs using agent’s assumptions.
The references of physical and ideal objects (from now on concepts) in social ob-
jects is represented with relations extracted from their content3 . Concepts are repre-
sented by their use as references to external descriptions or definitions, e.g. domain
ontologies. The distinction between use and definitions catch the meaning of social
objects as instances of shared models: a graph represents the specific use of general
concepts4 .
3 Norms as social constructs
Norms are social objects and legal texts are one of their inscriptions. A legal text is an
accessible reminders (media) leading agents to similar interpretations5 of norm mean-
2 Inscribed stand for written down or recorded in some retrievable media like paper or even
brain memory.
3 E.g. a semantic network of a document text.
4 The technical solution can be found in [9].
5 The use is vague but not necessary the meaning of norms.
Shared knowledge and interpretation of norms in multi-agent systems 3
ing. Norms are connected to agents’ beliefs like common and legal concepts, interpreta-
tions, judgements and common behaviour. The effectiveness of law is bounded to com-
munity’s interpretation of norms and related to common principles/concepts such as “to
not bring harm to someone”6 . Boella and van der Torre [10] showed the dependency
of normative systems to agents’ actions. In detail, they exposed how the (regulative,
constitutive, procedural) aspects of norms requires specific behaviours from agents. For
instance, the regulative effect needs the agents desire to avoid potential sanctions caused
by their actions. Rules emerge from the society in a process of definition of shared goals
and acceptable norms [11]. Those two works show the entanglement between norms,
institutional agents and society but are still based on the same limiting assumption that
norms can be represented as rules, norms are dynamic, vague and part of a greater social
or normative context.
Represent norms as social objects goes a step further in this direction. The notion
of social object allows us to model the requirements mentioned in Section 1:
1. Norms are different than their inscriptions and their life cycle depends only partly
on their inscriptions.
2. As social objects, norms meaning can evolve for other reasons than their inscrip-
tions.
3. Norms can depend to other norms as other social objects.
4. As social objects, norms meaning depend to the related concepts meaning.
5. As social objects, norms depend on the shared beliefs of the agents of the society
(even the very status of norms depends to agents’ agreement).
We want to discuss a methodology to build norm representations that holds those fea-
tures.
3.1 Structure and interpretation of norms
In real life, norms have more than one inscription (copies, versions, public documents,
etc.), several authors (legislators, commissions) for each part or aspect, formal and in-
formal goals, implicit assumptions, implicit and explicit scenarios, explicit and implicit
connections with other norms, related groups that give support (as the parliament)with
formal or informal actions, and a constitutive action (like X count as Norm in Italy)
creating the new norm from the legal texts and the other sources. Norms are complex
objects, made of components, partially defined and bounded to groups’ shared beliefs
and convergent interpretation.
If norm contents depends to agents’ beliefs and goal they are partially inaccessi-
ble and so impossible to represent fully. Furthermore, a norm context and content is
inevitable vague, concepts cannot be universally defined: their are sensible to society,
language and use dynamics. Considering human communication, open concepts and
vagueness are important aspect of communication that do not need to be resolved to
archive mutual understanding. For the sake of norms representation, all a priori at-
tempts of disambiguation of norm vagueness as introduced by legislators makes norms
representations unacceptable by practitioners and applications unreliable.
6 Norms containing general or undefined concepts are called “open norms”.
4 Antonini
3.2 Norm dynamics
A norm representation should catch the real-life dynamics of norms and not just the
dynamic of the legal text [12]. Even the legal text dynamics is more complex than the
life-cycle of legal text7 .
Norm dynamics includes all elements that contribute to norm meaning. For instance,
norms are affected by the current status of the whole normative system and not only
from their legal texts. Those elements have different time so a norm representation
should be incrementally extended. Furthermore, a norm representation cannot be finite
and self-explanatory but a shared starting point for agent’s interpretation [13].
4 The ontological impact of interpretation and vagueness
Interpretation hide always ontological assumptions. For instance, for an interpretation
theory norms evolve or not with society or language. We want distinguish norm as
social social objects to norm meaning as result of agent’s interpretation. We’ll discuss
interpretation in legal practice to give an intuition of norm representation as shared state
of affairs and interpretation as reasoning on shared norm representations.
In order to clarify the differences between ontological status of norms and interpre-
tation, we briefly recall the ontological analysis of the four canonical interpretations of
legal norms [13]:
a) The grammatical interpretation: norm meaning should be found in legal texts.
b) The systematic interpretation: norms meaning emerge from the state of the norma-
tive system.
c) The historical interpretation: norm meaning is within the original context of norms
and in the legislator’s will.
d) The teleological interpretation: norm meaning change with the current state of so-
ciety.
The ontological assumptions in the first two interpretations limits what is part of
a norm representation. The historical and teleological interpretations involve the norm
evolution: how to use the timing of norm components. In order to take in account all
interpretation, the best strategy is to build a widest possible representation and postpone
components selections and time interpretation.
4.1 Norm representation and explicit interpretation
The four interpretations presented in the previous section are all used in legal practice.
An objective representation of norms is required to derive all four interpretations from
the same representation. In general, a norm representation should preserve its original
meaning and all external references considering their timing. A norm can be repre-
sented as a stratified graph, like in Figure 1, in a network with other social objects and
ontologies for common concepts.
7 As text revisions, versions, etc.
Shared knowledge and interpretation of norms in multi-agent systems 5
0 2
C 0
r 2
T1
T 1
G 1
0 n 1
0 n 1
1
r 1
T1
0
G
0 T 1
1
Fig. 1: References between norms n0 and n1 with different timing.
The norm in Figure 1 has a unique identifier (the node n0 ) and its components are
labelled with the creation time8 .
An interpretation as uses external information extracts more than the original con-
tent from a norms. For instance, considering figure 1 the first two interpretation theories
can be implemented as selection of nodes by type (like references). The historical (left)
and teleological (right) interpretations, figure 2, can be calculated selecting the nodes
by time.
0 2 0
C 0
r 2 T 1 C 0
r 2 T 2
1
T1 T1
n 0
0
n 1
1
1
G
0 n 1
0 n 1
1
0
G
0
Fig. 2: Two interpretations of norm n0 in Figure 1 extracted by selection by time
attributes of nodes.
The previous example of interpretation is extremely simplified but even in this in-
terpretation require agent’s parameters: the node type and a time window. Interpretation
mechanism can also include graph traversing and combination of norm graphs.
5 Concluding remarks
In this contribution we depicted a methodology to represent shared knowledge and
agent’s interpretation. In particular, we answered the following questions:
1. How to represent the ontological status of norms? With an incremental representa-
tion of social facts and norm context, and excluding all interpretation assumptions.
2. How to represent open norms and vagueness? It is done avoiding a priori interpre-
tation and representing only the outcome of agent’s actions.
8 To simply the example we annotate time as parameters of nodes avoiding the representation
of time node
6 Antonini
3. How to represent agent’s interpretation of norms? The norm representation is the
ontological status of norms, even if it not so informative it can be interpreted using
agent’s assumptions as parameters of the interpretation mechanisms.
Our methodology requires to define the level of detail of the representation, it is
not exhaustive but can be used also with incomplete information. knowledge graph of
norms with existing tools, like text analysers, it is necessary to map the output of the
tools to a meta-model of norms. That will require, for each tool, the analysis of the
embedded ontological assumptions.
For future work, we want to address the integration between norm representations
of norms and society structures like organizations and roles.
References
1. G. Andrighetto, G. Governatori, P. Noriega, and L. W. N. van der Torre, eds., Normative
Multi-Agent Systems, vol. 4 of Dagstuhl Follow-Ups, Schloss Dagstuhl, 2013.
2. E. L. Rissland, K. D. Ashley, and R. P. Loui, “Ai and law: A fruitful synergy,” Artificial
Intelligence, vol. 150(1-2), pp. 1–15, 2003.
3. G. Governatori and S. Shek, “Business process compliance checker,” in Accepted paper in
the XIV International Conference on Artificial Intelligence and Law (ICAIL2013), 2013.
4. L. Giordano, A. Martelli, and D. T. Dupr, “Temporal deontic action logic for the verification
of compliance to norms in asp,” in Proceedings of the XIV International Conference on
Artificial Intelligence and Law (ICAIL2013), 2013.
5. N. Love and M. R. Genesereth, “Computational law,” in Proceedings in The Tenth Interna-
tional Conference on Artificial Intelligence and Law (ICAIL 2005), pp. 205–209, 2005.
6. M. Ferraris, Documentality: Why It Is Necessary to Leave Traces. Oxford University Press,
2012.
7. H. Grice, Logic and Conversation, p. 2240. Academic Press, 1975.
8. R. Searle, J, Speech Acts. Cambridge University Press, 1969.
9. A. Antonini, L. Vignaroli, C. Schifanella, R. G. Pensa, and M. L. Sapino, “Mesoontv: a media
and social-driven ontology-based tv knowledge management system,” in Proceedings of the
24th ACM Conference on Hypertext and Social Media, (New York, NY, USA), pp. 208–213,
ACM, 2013.
10. G. Boella and L. van der Torre, “Substantive and procedural norms in normative multiagent
systems,” Journal of Applied Logic, vol. 6(2), pp. 152–171, 2008.
11. G. Boella and L. van der Torre, “∆ : The social delegation cycle,” in Proceedings of the 7th
International Workshop on Deontic Logic in Computer Science (DEON 2004), pp. 29–42,
2004.
12. A. Antonini, G. Boella, and L. van der Torre, “Beyond the rules representation of norms:
norms as social objects,” in Proceedings of Rules 2013 Conference, 2013.
13. A. Antonini, C. Blengino, G. Boella, and L. van der Torre, “Norm dynamics: institutional
facts, social rules and practice,” in Proceedings of SOCREAL 2013 Workshop, 2013.