1. Agent Architecture for
Simulating Norm Dynamics. Part I
Rosaria Conte
rosaria.conte@istc.cnr.it
LABSS (Laboratory of Agent Based Social
Simulation), Roma, ISTC-CNR
2. Outline
How norms emerge? Conventions
But spontaneous equilibria are not always desirable…
1st simulation model
A more general notion is needed
EMIL-A: A cognitive norm-based architecture
Emergence and immergence
Mental representations
How tell norms
When is EMIL-A needed?
2nd simulation model
Why comply?
Towards a theory of norms internalization
3rd simulation model
Conclusions
3. Q How do norms emerge?
Q From which type of
agents? Questions
Q How necessary is norm
enforcement? Punishment is
essential in the evolution of
norms (Bowles and Gintis,
1998; 2003; Axelrod,1986 ;
etc.)
Norms are generally based on
enforcement
Usually complied with based
on strategic reasoning
Still moral education aims at
fostering compliance for the
sake of norms as ends in
themselves
How is this possible? Which
mental processes are needed
to make norms happy?
4. Norms in the behavioural
sciences
Norms are
universally present in all human societies (Roberts, 1979; Brown, 1991; Sober
and Wilson, 1998);
ancient: highly elaborated in all human groups, including hunter-gatherers and
groups that are culturally isolated.
ubiquitous. governing all activities, from mate choice to burial
Impactful: on welfare and reproductive success.
Nonetheless (or consequently?), norms break down in too specific notions
Archipelago norm includes at least
Conventions
Social norms
Laws
5. Conventions 1/5
From analytical philosophy (Lewis 1969),
social sciences derived a conventionalistic
view of norms as
spontaneously emerging
behavioral regularities
based on conditioned preferences
enforced by sanctions
For Lewis, conventions solve problems of
coordination,
When different equivalent solutions are
available,
But agents must converge on one such solution
Which is then arbitrary
Example: telephone line falling
Who is calling back?
6. Conventions 2/5
Why such a convention did never
establish?
It seems to crash with a norm of
equity…
But this does not solve problems of
coordination…
Exercise: other exs?
9. Conventions 5/5
In real
scenarios,
agents
may not
converge • Or they
at all may
converge
on pareto-
suboptimal
equilibria…
• Let us
simulate a
congestion
game
10. Strategies
• Unconditioned
• Aggressive: Hawks -> always
GOAHEAD,
• Cooperative: Doves -> STOP if
orthogonal agents approach
crossroad, else GOAHEAD
• Conditioned
• Left-watchers: if orthogonal coming from left
approach crossroad STOP, else GOAHEAD
• Right-watchers: dual of LW
16. Conclusions
How force a
desirable
solution?
Rather than a moral religious
behavioural social
notion legal
We need an
inlcusive notion
of norm that
Does justice to What is common to them?
its mandatory
force
17. A general notion
A norm “is a presribed guide for conduct which is
generally complied with by the members of society”
(Ullman-Margalit, 1977).
In our theory,
Norms spread because
and to the extent that the
corresponding normative prescriptions
spread as well
(Conte et al., 2007)
18. What is a normative
prescription?
A command that pretends to be adopted
for its own sake, because it ought to be
observed (Conte et al., 2009)
Ideally, norms are adopted for their own sake
Sub-ideally, norms are adopted because of
external enforcement
Norms’ felicity requires ideal reasons for
compliance.
19. Emergence implies immergence
S
EMIL project results: o
• To allow norm emergence ci
• agents need internal et
mechanisms and mental y
representations allowing
norms to affect their
behaviours.
M
• For a theory of immergence
see Castelfranchi, ; Conte et i
al., 2007.
• EMIL’s major outcomes
• Conte et al. (2011) Minding
n
Norms, OUP
• Xenatidiou and Edmonds d
(2011) A Dynmic View of
Norms, CUP.
20. Emergence implies immergence
EMIL project results:
S
• To allow norm emergence
• agents need internal o
mechanisms and mental ci
representations allowing et
norms to affect their y
behaviours.
• For a theory of immergence
see Conte et al., 2007.
M
• EMIL’s major outcomes
• Conte et al. (2011) Minding i
n
Norms, OUP
• Troitzsch and Gulyas (2011)
EMIL-S: Smulating norm
innovation, Wley
• Xenatidiou and Edmonds d
(2011) A Dynmic View of
Norms, CUP.
21. What are mental
representations?
States of the mind
triggering and guiding Gee, I thought Hey, do
behaviours that p’. you know
Could it be the that p?
Subsymbolic (eg., neural
same?
networks)
Symbolic: representations
of the world that can be
compared and
manipulated by the
agents while
Reasoning
Solving problems
Planning
Taking decisions
22. Two main functions
Epistemic: agents keep their
representations as close as Mind
possible to the world
Belief, knowledge, evaluation, etc.
Pragmatic: agents try to make the World
world as close as possible to their
representations
Goal, intention, motivation, etc.
Mind
How?
By means of planning and acting.
Lets go back to classic cybernetic World
circuits….
23. The TOTE unit (Miller et al.,
1960)
TEST: perceived ws
compared with
wanted ws; If
discrepant
OPERATE: apply action
TEST: perceived ws
compared with
wanted ws; If
coincident
EXIT
24. Norm-based mental
representations
N-beliefs
N-B1, general form N-B: there is an obligation, forbearance, permission on
a given set of agents to perform a given action.
N-B2, pertincence N-B: I am a member of the set of agents interested by
the norm.
N-B3, enforcement N-B concerning positive or negative sanctions
consequent to compliance or violation.
N-goals: a goal relativised to at least N-B1.
N-G1 N-adoption: want to act as prescribed, as long as and because this is
prescribed
N-G2 N-invocation: want others to form NBs
N-G3 N-defence: want others to comply with N
N-G4 Sanction: want violators be punished.
N-intentions: NGs chosen for execution.
25. Norm-based mental
representations
N-beliefs
N-B1, general form N-B: there is an obligation, forbearance, permission on a given
set of agents to perform a given action.
N-B2, pertincence N-B: I am a member of the set of agents interested by the norm.
N-B3, enforcement N-B concerning positive or negative sanctions consequent to
compliance or violation.
N-goals: a goal relativised to at least N-B1.
N-G1 N-adoption: want to act as prescribed, as long as and because this is
prescribed
N-G2 N-invocation: want others to form NBs
N-G3 N-defence: want others to comply with N
N-G4 Sanction: want violators be punished.
N-intentions: NGs chosen for execution.
28. Epistemic component
Vc=N-threshold
Vc=8
> vc
LTM
(CandidateN-Bel “It
N-bel:It is prohibited to smoke is prohibited to
smoke”)
W < vc
N-Board M
x smoke Prohibition y
Agent x Agent y
29. To practice 1/2
Vc=N-threshold
At time T1 Vc=8
LTM
(CandidateN-Bel “It
is prohibited to
smoke”) +
W
N-Board M
x ? ? y
Agent xi Agent y
30. To practice 2/2
Vc=N-threshold
At time T1 Vc=8
LTM
(CandidateN-Bel “It
is prohibited to
smoke”) -
W
N-Board M
?
x ? ? y
Agent xj Agent y
31. Epistemic component
LTM
N-board (norms arranged for salience)
N-bel1:general S
It is prohibited to smoke in public places m Norm salience
o measures how
N-bel2:pertinence. It concerns me ki operative NP is
n (perceived to be by
N-bel3: enforcement. Violators get a fiine g group members).
Signaling
(visibility)
Source Transgression
(Cred. Norm salience rate
& legitimacy Sanctions (pr. &
severity
Norm invocation
Norm's effect
33. Emergence of norms in artificial
populations
(www.emil.istc.cnr.it )
Artificial wikipedia (Emde and Troitzsch,
2008)
Traffic scenario (Lotzmann et al., 2008)
Microcredit (Lucas et al., 2009)
Multicontext world (Campennì et al, 2010)
models available at
http://mass.aitia.ai/applications/emil
Norm òatency
33
34. The Use of Norm Recognition
Module:
Effects on the Environment
35. Objectives
Lets compare
Norm recognizers
Social conformers
in a world in which agents leave traces of
their actions in the environment
Do they make a difference?
36. The Agent 1/2
Each Agent is provided with:
1. a Normative Board;
2. a double-layer architecture;
3. a vector of possible behaviors.
38. The Model 1/2
Agents
try to be compliant with surrounding environment;
follow preferred color (if switched on);
Social Conformers
tend to assimilate others’ preferences (to a certain speed)
Norm Recognizers
form normative beliefs and goals
All randomly move in the world (if they do not follow preferred colors)
color the patches with one of three possible colors:
Red
Black
Gray
39. The Model 2/2
Gray is more environmentally suitable than black and red: if
agents, in a portion of the world with lots of black and red
patches, color patches gray, they perturb the environment less
than would be the case otherwise (red if most patches are black
and vice-versa)
What is the relationship between environmental
responsiveness (color of patches) and norm compliance
(follow the salience of normative beliefs to choose the
action to be performed)?
40. Concluding Remarks
Social Conformers:
Rarely converge on one color
Sometimes GRAY with Uphill switched on
Norm Recognizers:
No case where the result is different from GRAY (they converge
very clearly on gray)
Mixed Populations:
More the population is composed by norm recognizers, more the
result tends to GRAY (small markers indicate mixed populations
– 50%)
41. Why?
As soon as the norm immerges, NR bring it around:
They compare it with current state of the envirnment
If conflict (2 cases out of 3), they act GRAY (to perturb
environment as little as possible)
Instead, SC act GRAY 1 out of 3, whether
they prefer gray and follow it
they modify their preference according to others’
It is the normative belief that generates compliance
42. First conclusions
While regularities can emerge in
populations of simple agents
“Prescribed guides of conduct” emerge
while immerging in the mind of rich
cognitive agents endowed with the
capacity to represent and adopt
prescriptions.
Immergence precedes emergence: Never smoke
Norms compete in the mind before
competing in society. Don’t smoke at work
Norm latency: it takes time before Don’t smoke
norms surface. Candidate norms In public
may never surface!
43. First conclusions
While regularities can emerge in
populations of simple agents
“Prescribed guides of conduct” emerge
while immerging in the mind of rich
cognitive agents endowed with the Don’t smoke
capacity to represent and adopt In public
prescriptions.
Immergence precedes emergence: Never smoke
Norms compete in the mind before Don’t smoke at work
competing in society.
Norm latency: it takes time before Don’t smoke
In public
norms surface. Candidate norms
may never surface!
44. For discussion
• When are simple architectures (say SC) fit?
• Which real-world setting does 2nd simulation
model refer to?
– Which actions
– Which norms
– Which domain?
• How about
– Evolutionary scenario
– Envirnmental policy