Using a network representation at different levels (social embedding and information), we have built a framework for social agents where reasoning is explicitly represented. We used abstract argumentation and argumentative theory of reasoning to build agents that exchange information through simulated dialogues. We demonstrated that our approach is, in principle, sufficient to reproduce two macro behaviors embedded in Granovetter’s theory, i.e., the tendency to inclusion of weak ties and a competitive advantage for non-isolated caves. Our framework explicitly models agents reasoning capabilities and can be applied to socially embedded and interacting agents. Our proposal: - represents a way for qualitative approaches to fit ABM formal requirements; - envisages possible new grounds for crossfertilization between computer sciences

1. A NEW FRAMEWORK FOR ABM BASED
ON ARGUMENTATIVE REASONING
SIMONE GABBRIELLINI AND PAOLOTORRONI
{SIMONE.GABBRIELLINI;PAOLO.TORRONI}@UNIBO.IT
1
1giovedì 19 settembre 13

2. INTRODUCTION
• Argumentative approach to ABM
• Agents are socially embedded
• Agents exchange information by means
of simulated dialogues.
• We argue that this approach:
• Allows a simple representation of
actor’s reasoning,
• helps filling the gap between scholars
that represent explicitly agent’s
reasoning and scholars that do not.
2
2giovedì 19 settembre 13

3. FIRST STREAM
• Models of the first stream share at least
two common features:
• a network representation to mimic
social embeddedness;
• a preference for mathematical, game
theoretical or evolutionary computing
techniques.
• It encompasses model of human
hierarchies, trust and norm evolution,
cooperation, cultural differentiation and
collective behaviors, opinion dynamics
and so on and so forth...
3
3giovedì 19 settembre 13

4. SECOND STREAM
• The second stream focuses on
how social agents should reason
• These models rely on formal
logics and BDI frameworks to
represent agent opinions, tasks
and decision-making capabilities.
• It encompasses models of trust,
cognitive representations and
norms evolution and evaluation,
and so on and so forth...
4
4giovedì 19 settembre 13

5. ARGUMENTATIVE AGENTS
• Our approach is grounded on well
established theories from social, cognitive,
and computer science...
• Many degrees of freedom:
• Different embedding structures
• Different trust models
• Different ways of processing information
• The only pivotal point is to
represent information and
reasoning with computational
abstract argumentation.
5
5giovedì 19 settembre 13

6. EMBEDDED AGENTS
• The basic idea of embeddedness
comes from Granovetter’s hypothesis,
w h i c h s t a t e s t h a t o u r
acquaintances are less likely
to be connected with each
other than our close friends...
• Granovetter called this type of bridges
“weak ties”, and demonstrated their
importance in permitting the flow of
resources, particularly information,
between otherwise unconnected
clusters - i.e. small worlds...
6
6giovedì 19 settembre 13

7. EMBEDDED AGENTS
• We use a “caveman graph” to represent a
situation where clusters are maximally dense...
• We then allow for two kind of structural
settings:
1. each “cave” is disconnected from the others,
thus agents can interact within their own
cluster only;
2. a random number of bridges is added
between caves, thus agents can interact
occasionally with members of different
caves. Even if our mechanism does not
guarantee that all the caves become
connected, on average the resulting
networks exhibit small-world network
characteristics...
7
1
2
7giovedì 19 settembre 13

8. AGENT REASONING AND
INTERACTION
• According to Mercier & Sperber’s
argumentative theor y of reasoning,
reasoning developed as a “tool” to
convince others by means of
arguments exchanged in dialogues
• Let’s see a brief sketch of how it works:
• Every time a bit of information is
received, an addressee checks if the
information is consistent with what she
knows
• If this is the case, the bit is included in
addressee’s knowledge, otherwise, she has
to react to avoid cognitive dissonance
8
8giovedì 19 settembre 13

9. AGENT REASONING AND
INTERACTION
• A d d r e s s e e f a c e s t w o
alternatives:
• either to reject the new
information (addressee does
not trust the source enough),
and possibly rebut;
• or to accept the new
information (addressee trusts
the source) and revise his/her
own beliefs;
9
I claim x
because y
???
9giovedì 19 settembre 13

10. AGENT REASONING AND
INTERACTION
• Both addressee and source
may have to revise their own
opinions while involved in such
a turn-taking interaction, until:
• after revising, they both
agree;
• they decide to stop arguing
because they do not trust
each other...
10
10giovedì 19 settembre 13

11. ABSTRACT ARGUMENTATION
• An Argumentation Framework (AF)
is defined as a pair ⟨A,R⟩, where:
• A is a set of arguments
• R is a binary attacks relation over
arguments, R ⊆ A×A, with α →
β ∈ R interpreted as “argument
α attacks argument β.”
• In other words, an AF is a network
of arguments, where directed links
represent attack relations
11
11giovedì 19 settembre 13

12. ABSTRACT ARGUMENTATION
• D1: My government cannot
n e g o t i a t e w i t h y o u r
government because your
government doesn’t even
recognize my government (a).
• D2: Your government doesn’t
recognize my government
either (b).
• D1: But your government is a
terrorist government (c)...
12
a
b
c
12giovedì 19 settembre 13

13. SEMANTIC EXTENSIONS
• Abstract argumentation analyzes
AFs by means of semantics, i.e.
set of rules used to identify
“ c o h e r e n t ” s u b s e t s o f
arguments.
• Semantics may range from very
credulous to very skeptical ones.
• Each coherent set of arguments,
according to the correspondent
s e m a n t i c s , i s c a l l e d a n
“extension” of A.
13
a
b
c
13giovedì 19 settembre 13

14. SIMULATED DIALOGUES
• A simulated dialogue D starts with an “invitation
to discuss” from A (source) to B (addressee), by
picking a random argument β in her own
extensions.
• IF: B evaluates β as coherent with his own AF ,
dialog stops: they already “agree”
• ELSE: if β is not included in any of B’s extensions, B
faces an alternative:
• if he trusts A, he will revise his own opinions
by adding the new information and
recalculating his semantic extensions
• if he does not trust A, he either will quit the
discussion or, if he can, rebut α→β and wait
for A’s reaction
• [So on and so forth until exit conditions...]
14
attack
invitation
14giovedì 19 settembre 13

15. AGAIN ON AGENT
REASONING
• As a framework, several points are left open:
• we do not commit to any specific
argumentation semantics
• we do not commit to any specific opinion
revision mechanism.
• We also assume that agents rely on a trust
model.Arguably, a realistic model of trust would
to take into account the authoritativeness, rank
and social status of the interlocutor.
• To date, our dialogue model is orthogonal to
trust, we define trust thresholds statically but
different trust models can be accommodated in
the future.
• Furthermore, in our model information is either
accepted or rejected...
15
15giovedì 19 settembre 13

16. ABM AND ARGUMENTS: NETARG
16
16giovedì 19 settembre 13

17. MODEL SCHEDULING
• At each time step, each agent A is asked
to start a dialogue with one of her
neighbors B, extracted at random, who
could be restricted to the same cave or
not, depending on the presence of
bridges.
• The probability to “argue” with members
of the same cave is higher than with out-
cave neighbors, according to the fact that
bridges are less activated than strong ties.
• A picks one random argument in her
extensions and addresses B
• B replies following the simulated dialogues
procedure already sketched
17
17giovedì 19 settembre 13

18. OUTCOME MEASURE
• The opinion revision process gives
raise also to a polarization effect at
the population level
• Polarization occurs when the
population divides into a small
number of factions with high internal
consensus and strong disagreement
between them.
• We measure the level of polarization
Pt as the variance of the distribution
of the AF distances dij,t, i.e. the AF
distance between agents i and j at
time t...
18
18giovedì 19 settembre 13

19. EXPERIMENTAL RESULTS
19
19giovedì 19 settembre 13

20. EXPERIMENTAL RESULTS
20
20giovedì 19 settembre 13

21. EXPERIMENTAL RESULTS
21
21giovedì 19 settembre 13

22. CONCLUSIONS
• Using a network representation at different
levels (social embedding and information),
we have built a framework for social agents
where reasoning is explicitly represented.
• We used abstract argumentation and
argumentative theory of reasoning to build
agents that exchange information through
simulated dialogues.
• We demonstrated that our approach is, in
principle, sufficient to reproduce two macro
behaviors embedded in Granovetter’s
theory, i.e., the tendency to inclusion of
weak ties and a competitive advantage for
non-isolated caves...
22
22giovedì 19 settembre 13

23. CONCLUSIONS
• Our framework explicitly models agents
reasoning capabilities and can be applied to
socially embedded and interacting agents.
• To the best of our knowledge, abstract
argumentation has not been used for social
simulation.
• Our proposal:
• represents a way for qualitative
a p p r o a c h e s t o fi t A B M fo r m a l
requirements...
• envisages possible new grounds for cross-
fertilization between the social and
computer sciences...
23
23giovedì 19 settembre 13

24. THANKYOU FORYOUR
ATTENTION!!!
24
24giovedì 19 settembre 13