An invited talk at the College of Integrative Sciences General Science Seminar, February 14, 2019, Wesleyan University, Middletown, CT.

1. Swarm Chemistry:
A Decade-Long Quest to
Emergent Creativity in
Artificial “Nature”
Hiroki Sayama
Binghamton University, State University of New York
sayama@binghamton.edu

2. About the Speaker
• D.Sc. in Information Science, University of Tokyo,
Japan (1999)
• Postdoctoral Fellow, New England Complex Systems
Institute (1999-2002)
• Assistant/Associate Professor of Human Communication,
University of Electro-Communications, Tokyo, Japan (2002-2005)
• Assistant/Associate Professor of Bioengineering -> Associate/Full
Professor of Systems Science and Industrial Engineering,
Binghamton University, State University of New York (2006-)
• Associate/Full Professor of Commerce, Waseda University, Tokyo,
Japan (2017-)
• Research areas: Complex systems, network science, organization
science, artificial life/chemistry, evolutionary computation,
complex systems education, etc.
2

3. Artificial Life
(1987-)
• Interdisciplinary field founded
by Christopher Langton
• International Society for
Artificial Life (ISAL; 2001-)
http://alife.org
• Artificial Life journal (MIT
Press)
• ALIFE/ECAL conferences
• IEEE ALIFE, AROB conferences
3

4. “Artificial Life”
• “... is a field of study devoted to
understanding life by attempting to
abstract the fundamental dynamical
principles underlying biological
phenomena, and recreating these
dynamics in other physical media ― such
as computers ― making them accessible
to new kinds of experimental
manipulation and testing.”
• “... In addition to providing new ways to
study the biological phenomena
associated with life here on Earth, life-as-
we-know-it, Artificial Life allows us to
extend our studies to the larger domain of
"bio-logic" of possible life, life-as-it-could-
be.”
• C. G. Langton, "Preface", In C. G. Langton, C. Taylor, J. D. Farmer,
and S. Rasmussen, eds., Artificial Life II, vol. X of SFI Studies in the
Sciences of Complexity, pp. xiii-xviii, Addison-Wesley, 1992.

5. 14 Themes of Artificial Life
(Aguiar et al. 2014)
5
Origins of life Autonomy
Self-
organization
Adaptation Ecology
Artificial
societies
Behavior
Computational
biology
Artificial
chemistries
Information
Living
technology
Art Philosophy

6. Collective Behavior
of Swarms
• Commonly seen in many
animal species
• Fish, birds, insects,
pedestrians
• Self-organization of
macroscopic behavior
out of local
interactions of
individuals
6
© Iain Couzin

7. 7

8. Boids (“Stanley and Stella in:
Breaking the Ice”, 1987)
8

9. Craig Reynolds,
Academy Award
Winner (Scientific
and Engineering
Award, 1998)
For his pioneering
contributions to the
development of
three-dimensional
computer animation
for motion picture
production
9

10. 2006

11. 11

12. 12

13. Artificial Chemistry
•A sub-field of Artificial
Life where models of
artificial chemical
reactions are used to
study the emergence
and evolution of life
from non-living
elements
13

14. Sayama (2006;
from teaching
material at UEC)
14

15. Tenure Pressure as the Mother of Innovation
Mix
multiple
swarms
together
Mix
multiple
swarms
together
Call it
“chemistry”
Call it
“chemistry”
15

16. 16
Sayama(2009)

17. Swarm
Chemistry
http://bingweb.binghamton.edu/~sayama/SwarmChemistry/
17

18. i
• Particles in a continuous open 2-D space
• Kinetic interactions with local neighbors
• No capability to distinguish different types
Model Assumptions
18

19. Behavioral Rules
Cohesion Alignment Separation
19

20. • If no particles are found within local perception range, steer
randomly (Straying)
• Otherwise:
• Steer to move toward the average position of local neighbors
(Cohesion)
• Steer towards the average velocity of local neighbors (Alignment)
• Steer to avoid collision with neighbors (Separation)
• Steer randomly with a given probability (Randomness)
• Approximate its speed to its normal speed (Self-propulsion)
Behavioral Rules (Detail)
20

21. Kinetic Parameters
(Assigned to each particle individually)
21

22. • Similar to those reported in earlier literature
Stationary
clustering
Coherent linear
motion
Amoeba-like
structure
Dispersal
[simulation]
Behavior of Homogeneous Swarms
22

23. Dependence on Kinetic Parameters
• Phase transitions observed
• Effect of c3 not so significant
c1: strength of cohesion
c2:strengthofalignment
c1: strength of cohesion c1: strength of cohesion
c2:strengthofalignment
c2:strengthofalignment
<|v|> : av. absolute vel. <r>: av. distance from CoM
(Results with c3 ~ 50 ± 5)
<v>: average velocity
Coherent linear
motion
Random
swarming
Oscill-
ation
Dispersal
23

24. Behavior of Heterogeneous Swarms
• Simple case: Two-type interaction
+  ??
24

25. Spontaneous Segregation
• Particles spontaneously segregate themselves unless two types share
the same cohesion/separation ratio (c1/c3)
(c1 / c3)A – (c1 / c3)B
LocalhomogeneityH
- Alignment (c2) not so relevant
- c1/c3 determines equilibrium
distance: req = (c1/c3)–1/2
 Segregation is caused by
difference of “personal space”
25
[simulation]

26. Emergent Motion
• Mixing two types may generate new motion that was not seen in either
of them
Linear motion Rotation Oscillation 26

27. Interactive Evolutionary Computation (IEC)
• Powerful tool for designing complex systems and solving complex
problems
“SBART” © Tastuo Unemi “Hunch EngineTM” © Icosystem
27

28. i
Cohesion Alignment Separation
97 * (226.76, 3.11, 9.61, 0.15, 0.88, 43.35, 0.44, 1.0 )
38 * ( 57.47, 9.99, 35.18, 0.15, 0.37, 30.96, 0.05, 0.31)
56 * ( 15.25, 13.58, 3.82, 0.3, 0.8, 39.51, 0.43, 0.65)
31 * (113.21, 18.25, 38.21, 0.62, 0.46, 15.78, 0.49, 0.61)
Recipe
28

29. Hyper-Interactive
Evolutionary
Computation
(Bush & Sayama 2011)
• Evolutionary operators act on
each swarm locally and visually
• Continuous generation changes
• Number of swarms on a
screen dynamically changes
with no predetermined
bound
[demo]

31. 31

32. 32

33. More Patterns Generated by Students
[demo]
Swarms with highest ratings
Swarms with lowest ratings
33

34. Going 3D
(Sayama 2012)
• Straightforward extension of
position/velocity vectors from
2D to 3D
• 3D visualization realized in
plain Java
[Demo]
34

35. Behaviors Robust to Dimensional Changes
2D
3D
2D
3D
35

36. Behaviors Recovered with Minor Parameter
Adjustments
36

37. Do We Really
Need Human
Users?
• No we don’t, if:
• The system can
spontaneously evolve new
potentially innovative designs
and if
• We know what kind of
quantitative properties we
are looking for
37

38. Evolutionary Swarm
Chemistry

39. Making Swarm Chemistry
Evolvable (Sayama 2010,
2011)
• Recipe transmission
between active particles
• Mutation of recipes in
local transmission
• Environmental
perturbation to break
status quo
39

40. Recipe Transmission
1. Active/passive particles
2. Recipe transmission from active to passive
97 * (226.76, 3.11, 9.61, 0.15, 0.88, 43.35, 0.44, 1.0)
38 * (57.47, 9.99, 35.18, 0.15, 0.37, 30.96, 0.05, 0.31)
56 * (15.25, 13.58, 3.82, 0.3, 0.8, 39.51, 0.43, 0.65)
31 * (113.21, 18.25, 38.21, 0.62, 0.46, 15.78, 0.49, 0.61)
97 * (226.76, 3.11, 9.61, 0.15, 0.88, 43.35, 0.44, 1.0)
38 * (57.47, 9.99, 35.18, 0.15, 0.37, 30.96, 0.05, 0.31)
56 * (15.25, 13.58, 3.82, 0.3, 0.8, 39.51, 0.43, 0.65)
31 * (113.21, 18.25, 38.21, 0.62, 0.46, 15.78, 0.49, 0.61)
40

41. Dynamic (Re-)Differentiation
3. Random differentiation at the transmission
4. Occasional re-differentiation
97 * (226.76, 3.11, 9.61, 0.15, 0.88, 43.35, 0.44, 1.0)
38 * (57.47, 9.99, 35.18, 0.15, 0.37, 30.96, 0.05, 0.31)
56 * (15.25, 13.58, 3.82, 0.3, 0.8, 39.51, 0.43, 0.65)
31 * (113.21, 18.25, 38.21, 0.62, 0.46, 15.78, 0.49, 0.61)
97
38
56 31
41

42. Self-Repair Achieved for Free

43. Recipe Transmission Between Active Particles
5. Competition between active particles
97 * (226.76, 3.11, 9.61, 0.15, 0.88, 43.35, 0.44, 1.0)
38 * (57.47, 9.99, 35.18, 0.15, 0.37, 30.96, 0.05, 0.31)
56 * (15.25, 13.58, 3.82, 0.3, 0.8, 39.51, 0.43, 0.65)
31 * (113.21, 18.25, 38.21, 0.62, 0.46, 15.78, 0.49, 0.61)
67 * (216.35, 11.75, 7.7, 0.83, 0.97, 97.31, 0.02, 0.38)
29 * (254.64, 7.28, 7.0, 0.95, 0.11, 22.41, 0.43, 0.31)
13 * (105.4, 3.55, 5.24, 0.34, 0.18, 23.53, 0.39, 0.24)
competition
function
winner:
43

44. 97 * (226.76, 3.11, 9.61, 0.15, 0.88, 43.35, 0.44, 1.0)
38 * (57.47, 9.99, 35.18, 0.15, 0.37, 30.96, 0.05, 0.31)
56 * (15.25, 13.58, 3.82, 0.3, 0.8, 39.51, 0.43, 0.65)
31 * (113.21, 18.25, 38.21, 0.62, 0.46, 15.78, 0.49, 0.61)
Mutations
6. Recipe transmission & mutation
75 * (216.35, 11.75, 7.7, 0.83, 0.97, 97.31, 0.02, 0.38)
29 * (254.64, 7.28, 7.0, 0.95, 0.11, 28.56, 0.43, 0.31)
13 * (105.4, 3.55, 5.24, 0.34, 0.18, 23.53, 0.39, 0.24)
44
67 * (216.35, 11.75, 7.7, 0.83, 0.97, 97.31, 0.02, 0.38)
29 * (254.64, 7.28, 7.0, 0.95, 0.11, 22.41, 0.43, 0.31)
13 * (105.4, 3.55, 5.24, 0.34, 0.18, 23.53, 0.39, 0.24)

45. Competition
Functions
• “Faster”, “Slower”, “Behind”
• “Majority”: The particle surrounded by
more of the same species wins
• Deterministic, probabilistic, relative
(density-based)
• “Recipe length”: The particle with a
longer recipe wins
• Length only, length then majority,
length times majority
45

46. Emergent Ecology and Evolution of Swarms
faster slower behind majority
random
designed
46

47. Introducing
Environmental
Perturbations
• Perturbation I:
Competition function
switched from “majority-
relative” to either “faster”
or “slower” for 50 time
steps in every 5000 time
steps
• Perturbation II:
Competition function
switched from “majority-
relative” to either “faster”
or “slower” for 50 time
steps in every 2000 time
steps only in either left or
right half of the space
47

48. Automatic Identification of “Interesting”
Results (Sayama & Wong 2011)
People are looking for:
Clear macroscopic
structures
Continuous changes
Can “interesting results” be identified automatically
without human intervention?
48

49. Measurement 1: Macroscopic Structuredness
Actual snapshotRandom distribution
Pairwise distance distribution
P(d) Q(d)
Kullback-Leibler Divergence
= 0.1184
49

50. 50

51. Measurement 2: Evolutionary Exploration
Time
Colors that appeared
in this 500-step
period
Colors that appeared
in this 500-step
period
Colors that
appeared in this
500-step period
# of new colors
=151
# of new
colors =96
51

52. 52

53. Overall Results
One condition
successfully
maintained large
values for both
measurements
53

54. Automatically Identified Best Runs
• http://YouTube.com/ComplexSystem/
54

55. But Evolution Slows Down in 3D (Sayama 2012)
3D
2D
55

56. Open-Ended Evolution in
Long-Term Experiments

57. Open-Ended
Evolution
Evolution that keeps producing novel forms and
adaptations with no apparent limit

58. https://www.oreilly.com/ideas/open-endedness-
the-last-grand-challenge-youve-never-heard-of
58

59. Sayama, IEEE ALIFE (2011) Sayama & Wong, ECAL (2011)
Is This Truly Open-Ended?
59

60. Simulate It Much Longer
Detect Objects Automatically
60

61. Automated
Object
Harvesting
(Sayama 2018)
61

62. 62

63. 63

64. Environmental Perturbations
• Competition function switched from
“majority-relative” to either “faster”
or “slower” for T steps in every 2000
steps in either half of the space
T = 50 (short, moderate)
T = 500 (long, severe)
64

65. Detailed Object Tracking (Object Size)
With moderate environmental perturbations
Objects tend
to get bigger
over time
65

66. Detailed Object Tracking (Object Size)
With moderate environmental perturbations With severe environmental perturbations
But it also depends on the
severity of perturbations
66

67. Detailed Object Tracking (Color Diversity)
With moderate environmental perturbations With severe environmental perturbations
Severe perturbations may increase
color entropy of objects
67

68. With Moderate Environmental Perturbations

69. Evolved Under Moderate Perturbations
69

70. With Severe Environmental Perturbations

71. Evolved Under Severe Perturbations
71

72. Automated Generation of
Novel Forms by Evolution
72

73. Conclusions
Artificial life/chemistry study
hypothetical life-like dynamical
systems in creative ways
Simple swarm chemistry
models can demonstrate
surprisingly rich behaviors
Evolutionary swarm chemistry
shows long-term evolutionary
creativity autonomously
Open-endedness in artificial
evolution may be utilized to
achieve creative machines

74. By Graham Moore https://youtu.be/xbNJq90t0Wk
74

75. Thank You
@hirokisayama