an rg theory of cultural evolution
DESCRIPTION
An RG theory of cultural evolution. Gábor Fáth Hungarian Academy of Sciences Budapest, Hungary in collaboration with Miklos Sarvary - INSEAD, Fontainebleau, France. mental representation. shared. evolving. Sociology defines: „ Culture is the sum of knowledge, - PowerPoint PPT PresentationTRANSCRIPT
An RG theory ofcultural evolution
Gábor FáthHungarian Academy of Sciences
Budapest, Hungary
in collaboration with
Miklos Sarvary - INSEAD, Fontainebleau, France
What is Culture?
Sociology defines:
„Culture is the sum of knowledge,
beliefs, values, norms and behavioral patterns
represented by a group of human beings
and transmitted from one generation to the next.”
mental representation
shared evolving
Cultural identity as a clustering problem
unique individuals
humanity
similarity threshold
culturessubcultures
Two possible approaches
• How can cultural diversity emerge despite our fundamental biological similarities?
• How can cultural coherence emerge despite our fundamental biological/economic/ environmental differences?
Axelrod’s theory of culture (1997)
Outline
1. Mental representation:
RG approach for bounded rationality
2. Heterogeneous interacting agents:
Spontaneous ordering of mental reps.
Mental representation
Objective (physical) reality Representation in the mind
Bounded rationality = representation error
Mental representation (2)
COMPLEXITY
AC
CU
RA
CY
Optimal representationfor given complexity
Sub-optimal representations
- - - - - - - - - - B o u n d e d r a t i o n a l i t y - - - - - - - - Superrationality
Concepts
Cognitive science:
Human mind is a feature detector („pattern recognizer”, „filter”)
We only perceive the part of reality which we have a concept for.
e.g.: chess concepts
Model of bounded rationality
Choice among decision alternatives
Mental model on concepts+
Concepts as feature detectors
Microscopic variables describing decision alternatives
Behavior
Perception
Mental representationin MIND
Bounded rationality in chess
xa x
a
Objectivepayoff
(super rationality)
Estimatedpayoff
(bounded rationality)
How to evaluate a move ?
Microscopicattributes
of alternative(board configuration)
Concepts
Value of alternative in
decision context(value of move)
Weak pawn
Pinned piece
Positionaladvantage
decisioncontext
(adversary) possiblemove
payoffof move
decisioncontext-2
(adversary-2)
decisioncontext-3
(adversary-3)
Bounded rationality in general
xa x
a
Objectivepayoff
(super rationality)
Estimatedpayoff
(bounded rationality)
How to evaluate a decision alternative ?
Microscopic attributesof alternative a = {a1,…,aD}
ConceptsConcept ConceptConcept
decisionalternative
payoffof alternative
payoff of alternative in decisioncontext - 1
payoff of alternative in decisioncontext - X
1 D
1 KFixed
Fixed
Fixed
Linear
Linear
Linear world approximation
Microscopic attributesof alternative a = {a1,…,aD}
ConceptsConcept ConceptConcept
payoff of alternative in decisioncontext - 1
payoff of alternative in decisioncontext - X
1 D
1 K
Linear
Linear
context dependentpreference vectors
mental weights
concept vectors
Representation error
Assume that attributes are -correlated:
How to choose the mental representation (v(x) and d)
to minimize the error?
(we assumeK andd(x) fixed, ||=1)
Representation error (2)
Assuming v(x) (mental weights) are fast variables we get
How to choose the concepts {,2,…,K} to maximize utility?
Def: Agent’s utility
with
and World matrix
Principal Component Analysis
Theorem (Principle Component Analysis)
Error is minimal if are the K most significant eigenvectors of W.
This is the PCA problem:
Cultural profile
Def:
Agent’s cultural profile
=
PCA-defined K dim -subspace of D dim
World
Determines what/how the agent can understandpredict communicatemeanbehave…
Connection with DMRG
xd
d,x
xd
,x d
xd
d,x
22
x xd d
d,x
x xd d 'dd ' dd '
x
d x
v x with d
v d x
Trunc. Error v Re p. Error
Density matrix World matrix
Superblock target state
Renormalized superblock ground state
Block {d} Environment {x}
Superblock
Renormalization
Heterogeneous agents
Heterogeneous preferences
World matrices differ Wi ≠ Wj
Cultural profiles (-subspaces) differ
Tensor order parameter (à la de Gennes):
Eigenvalue structure measures cultural ordering!
Perfect order:
Perfect disorder:
Order parameter
How to measure cultural (subspace) coherence?
Interactions
Understanding/predicting other agents is advantageous
Reality = Individual + Social
strength of social interactions
Sj projects onto agent j’s -subspace
Mean field:
Agent’s utility:
DynamicsBest response dynamics
co-adaptation to natural & social environment
0i iW (t) W h O(t 1)
i (t)
O(t)
PCA
Population average
RG
iter
atio
n cy
cle
RG fixed point: Nash equilibrium (no incentive to deviate)
Phase transition
Fixed point properties for
heterogeneous agents with unbiased random preferences
Wi0 = Wishart distributed
h < hc: disorderedh > hc: ordered
Spontaneous ordering in1st order transition
Analytic results
Disordered solution loses stability at hc
hc can be calculated using
1st order perturbation theory and RMT (Wishart)
For K << D=Xcomplexity of world
capacity of agents
critical social coupling strength
Phase diagram
Agent intelligence K
Str
engt
h of
soc
ial i
nter
actio
ns h
Unbiased random population D = fixed
DisorderedNo Culture No Language
OrderedCoherent Culture, Language
Cultural explosion ~50,000 years ago
Summary
Culture: Ordering of mental representations (Concepts)
Bounded rationality: Mental rep. should be accurate and simpleRG agents: Sorting / truncating the degrees of freedom
IterativelyFixed points
Spontaneous ordering with jumpsas mental abilities improveas interactions strengthen
Archeological evidence: „Cultural explosion”
G. F
ath
and
M. S
arva
ry,
nlin
.AO
/031
2070