This document summarizes a presentation on self-organization of society. It discusses how social fragmentation, disagreement and extremism can emerge from decentralized interactions between individuals seeking conformity and homophily. Three recent papers are summarized that show how social networks can become polarized through adaptive dynamics, how enhanced information gathering can intensify disagreement, and how behavioral diversity among individuals can allow for both cultural diversity and network connectivity in society. The key messages are that individual and collective outcomes may not align, and behavioral heterogeneity presents opportunities for diverse yet cohesive social outcomes.

1. Self-Organization of Society
Fragmentation, Disagreement, and How to Overcome Those
ACSOS 2020 Washington D.C. / Online August 20, 2020
Hiroki Sayama (Binghamton University & Waseda University)
sayama@binghamton.edu

2. Emergence
and Self-
Organization
2
Source:
https://en.wikipedia.org/wiki/Complex_system#/media/File:Complex_
systems_organizational_map.jpg (diagram created by the speaker)

3. Self-Organization ☺
Decentralized information processing
Robust, resilient function and self-
maintenance
Autonomous learning and self-adaptation
3

4. Self-Organization 🙃
Absence of easy-to-find
leverage points
Strong self-preservation and
self-constraints
Groupthink, habituation,
institutionalization
4

5. Self-Organization 🙃
of Society
Segregation (Sakoda 1971;
Schelling 1971; Hegselmann 2017)
Polarization (Conover et al. 2011;
Morales et al. 2015)
Inequality (Price & Feinman 1995)
Preferential attachment (Price
1965; Barabasi & Albert 1999)
5
Source: Morales et al. 2015 Source: Wikipedia

6. Modeling Society as
a Self-Organizing
System
6

7. Agenda: Three Recent Papers
Sayama, H. (2020) Extreme ideas emerging from social conformity and homophily: An adaptive
social network model. Proceedings of the 2020 Conference on Artificial Life (ALIFE 2020), MIT Press,
pp. 113-120.
Sayama, H. (2020) Enhanced ability of information gathering may intensify disagreement among
groups. Physical Review E, 102, 012303.
Sayama, H. & Yamanoi, J. (2020) Beyond social fragmentation: Coexistence of cultural diversity and
structural connectivity is possible with social constituent diversity. Proceedings of the International
School and Conference on Network Science (NetSci-X 2020), Springer, pp. 171-181.
7

8. 1. Fragmentation and
Extremization
Sayama, H. (2020) Extreme ideas emerging from social conformity and homophily: An adaptive social network model.
Proceedings of the 2020 Conference on Artificial Life (ALIFE 2020), MIT Press, pp. 113-120.
8

9. 9
This Photo by Unknown Author is licensed under CC BY-SA
This Photo by Unknown Author is licensed under CC BY-ND
This Photo by Unknown Author is licensed under CC BY
This Photo by Unknown Author is licensed under CC BY-ND

10. 10
This Photo by Unknown Author is licensed under CC BY-SA
This Photo by Unknown Author is licensed under CC BY-NC
This Photo by Unknown Author is licensed under CC BY-SAThis Photo by Unknown Author is licensed under CC BY-SA-NC

11. 11
Are people trying to
escalate their ideas
and actions
intentionally?
Or could it be just self-
organization in which
they try to conform to
each other?

12. Adaptive Networks Dynamical networks whose
states and topologies co-evolve,
often over similar time scales
(Gross & Sayama 2009; Sayama et al. 2013)
12

13. An Adaptive Network Opinion Model
13
Social conformity Random fluctuation
Homophily Attention to novelty
Node state
Edge weight
Social norm

14. Homophily and Attention to Novelty
14

15. 15
Live Demo

16. Parameter Sweep Simulations
• Network size:
• Parameter values:
• t = 0-100, Dt = 0.1
• Node states & edge weights are initially random
• Five independent runs for each parameter combination
(except for n = 1000)
16

17. Social Fragmentation Transition
17
(Kozma & Barrat 2008)
(Centola et al. 2007)

18. Outcome Measures
• Average edge weight low high
• Number of communities high low
• Modularity of community structure high low
• Range of average community states high low
• Std. of average community states high low
18
StructureIdeas

19. 19
n = 1000
Results
(Structure)
c h a qh qa

20. 20
Results
(Ideas)
c h a qh qa
n = 1000

21. Regression Analysis of Effect Sizes
21

22. Summary
• Homophily promotes escalation of ideas
even if people just seek social conformity
• Attention to novelty makes society
connected but homogeneous
• A paradox:
Society produces diversity when
individuals seek sameness;
society reaches sameness when
individuals seek diversity
22

23. 2. Disagreement
Sayama, H. (2020) Enhanced ability of information gathering may intensify disagreement among groups. Physical Review
E, 102, 012303.
23

24. Widening
Disagreement
24Sources: Pew Research Center 2014, Disrn 2020, Pixabay

25. How Does Our Enhanced Information
Gathering Ability Affect Disagreement?
25

26. 26Source: Wikipedia (Bruno in Columbus / Public domain)

27. PDE Model
27
Diffusion Aggregation
Perception
kernel
(Inspired by Keller & Segel 1970; Horstmann 2003/2004; Boi et al. 2000; etc.)

28. PDE Model
28
Cross-correlation between P
and kernel function g
‫׬‬ < 0

29. Perception Kernel
29
G(P) converges to local
gradient with m, s → 0
Represents enhanced
information gathering ability

30. Not-So-Live
Demo
30

31. Space-Time
Visualization
31
time

32. Linear Stability Analysis
32
?

33. Linear Stability Analysis (Result)
33
If this is positive
for some w,
perturbation will
grow

34. Effect of Perception Kernel
34
How does the shape of g(y)
affect Q(w)?

35. as a Function of m and s
35
As m and s becomes larger, the characteristic value of w becomes smaller (= distance between peaks greater)

36. 36
m
s

37. Peak Distance as a Function of m and s
37
Surface: Analytically
obtained lower bound
Blue dots: Numerical simulations

38. Scenario 1: Techno-
logical Advancement
38
Initially m, s = 0.5 for t = 0~20
Then linearly increased up to
5.0 by t = 200

39. Scenario 2: External
Intervention
39
Same scenario as the
previous one up to t = 100
Then m, s are forcefully reset
to 0.5 (e.g., government bans
social media)

40. Summary
• Widening disagreement among groups may
be caused by broader perception kernel
• External intervention may not help relieve
already existing disagreement
• A paradox:
The more information people
have access to, the more severe
their disagreement can get
40

41. Fragmentation or homogenization
Widening disagreement with no way
back
41

42. 3. So, Are We All
Messed Up?
Sayama, H. & Yamanoi, J. (2020) Beyond social fragmentation: Coexistence of cultural diversity and structural connectivity
is possible with social constituent diversity. Proceedings of the International School and Conference on Network Science
(NetSci-X 2020), Springer, pp. 171-181.
42

43. Not Quite:
Individual Diversity
Ideas
Backgrounds
Behaviors
Knowledge
Skills
Personalities
…
43

44. Introducing Behavioral Heterogeneity
44
What if the agents are behaviorally heterogeneous?
Earlier models typically assumed behaviorally
homogeneous, identical agents

45. Cultural Integration Model (Yamanoi & Sayama 2013)
Each agent has an m-dimensional cultural state
45
Culture of community B
Cultural space (e.g., m=2)
Culture of community A
Cultural distance

46. Cultural Spread on an Adaptive Network
A node chooses info source using
link weights as probs.
◦ May pick a random node (1%)
It may “accept” the source’s
culture based on cultural distance
46

47. Link Weight Updates
47
acceptancerejection

48. Live Demo
48

49. Behavioral Parameters
• Cultural tolerance d
• Rate of cultural state change rs
• Rate of edge weight change rw
49

50. Parameter Sweep Simulations
Initial condition:
Two culturally distinct groups connected by random ties
Parameters d, rs, rw: mean = 0.5, s.d. = 0 ~ 0.5
100 replications of simulation for each parameter value
50

51. Outcome Measures
<CD>: average cultural difference
b/w original two subnetworks
<SPL>: average shortest path length
• (<CD>, <SPL>) = (high, high)
=> social fragmentation
• (<CD>, <SPL>) = (low, low)
=> loss of cultural diversity
51

52. Results (d)
52
High s.d.-of-d areas showed
(<CD>, <SPL>) = (high, low)
behavior that was not
previously reported

53. Diverse Yet
Connected
Society
53

54. Results (rs , rw)
54

55. Parameter Interactions on <CD>
55

56. Parameter Interactions on <SPL>
56

57. Regression Analysis of <CD>
57

58. Regression Analysis of <SPL>
58

59. Regression Analysis (Combined)
59

60. Summary
• Diversity of cultural tolerance helps
coexistence of cultural diversity and
structural connectivity
• Not just tolerance, but diverse levels of
tolerance are needed
• A paradox:
Behavioral heterogeneity makes
diverse society more connected
than homogeneity does
60

61. Conclusions
61

62. Take-Home Messages
• Society can show non-trivial self-organization
• Fragmentation
• Production of extreme ideas
• Disagreement
• Connectivity
• Individual actions and collective outcomes may not align
• Behavioral heterogeneity presents opportunities
62

63. Key Question to Consider
•What do we want?
• Unity and homogeneity?
• Consensus formation?
• Conflict resolution?
• Efficient collective decision making?
• Diversity?
• Novel discoveries and innovation?
63

64. In This With-COVID Society
• Our inter-personal communications are now almost
entirely online 🙃
• System infrastructure has tremendous impact on how
society will self-organize in the coming decades
• Risks, vulnerabilities, opportunities
• Counter-intuitive connections between individuals and the
collective must be considered in system design
64

65. Acknowledgments
Junichi Yamanoi (Waseda University)
This work is supported by:
◦ Waseda University Research
Promotion Division
◦ JSPS KAKENHI Grant 19K21571
◦ Max Planck Institute for the Physics
of Complex Systems Visitors
Program
65

66. Thank You
@hirokisayama
66