Artificial Life

1
Artificial Life
and Computer Graphics

• Alexander Bisler
• Fraunhofer IGD
• Fraunhoferstr. 5
• 64283 Darmstadt
• alexander_at_bisler.de

2
Synopsis

• ALife Introduction
• Cellular Automata
• Self-Organization
• L-Systems
• Genetic Algorithms
• Artificial Evolution for CG
• Simulations Instantiations
• Summarized Principles

3

• ALife

4
ALife - Definition

• Biology study of carbon-based life
• "life as we know it"
• ALife study of the dynamics of living systems,
• regardless of substrate
• "life as it could be"
• Substrates abstract chemistries, logical
  networks, cellular automata, abstract ecosystems,
  emulated computers

5
ALife - Definition

• Artificial Life ("AL" or "Alife") is the name
  given to a new discipline that studies "natural"
  life by attempting to recreate biological
  phenomena from scratch within computers and other
  "artificial" media.
• Alife complements the traditional analytic
  approach of traditional biology with a synthetic
  approach in which, rather than studying
  biological phenomena by taking apart living
  organisms to see how they work, one attempts to
  put together systems that behave like living
  organisms.

6
ALife Bio-Logic

• We want a general understanding of living
  systems.
• Living systems are
• complex
• non-linear
• synergistic (whole greater than sum of parts)
• Decompositional approaches (math) are limited.
• Synthetic methods (simulations) are needed.

7
ALife Emergence Self-Regulation

• "The signal feature of life is not the
  carbon-based substrate
• (but) that the local dynamics of a set if
  interacting entities
• (e.g. molecules, cells, etc. )
• supports an emergent set of global dynamical
  structures
• which stabilize themselves by setting the
  boundary conditions
• within which the local dynamics operates."
• Charles Taylor, biologist, UCLA

8
ALife Properties of ALife-Systems

• Synthetic bottom-up, multiple interacting agents
• Self-Organizing global structure is emergent
• Self-Regulating no global/centralized control
• Adaptive learning and/or evolving
• Complex on the edge to chaos dissipative

9
ALife Relevant Domains for ALife

• Individual organisms physiology,behavior,develop
  ment
• Evolutionary biologyecology emergence of
  ecosystems
• Social insect colonies emergence of
  "super-organisms (Ant Colony Optimization)
• Evolutionary Robotics
• MicroEconomics populations of buyers sellers
• Sociology emergence evolution of
  societies
• Traffic emergence of flow/jam patterns
• Abstract computational models logic, chemistry,
  physics

10
ALife Why Study ALife ?

• Understanding emergent phenomena
• Synthetic approaches vs. analytic reductionism
• Chaos, complexity, self-organization
• Biological research
• Controllable reproducible evolutionary
  simulations
• New technologies
• Genetic engineering
• AI multi agent systems, evolutionary computation
• Educational toolkits
• Social system (SimCity)
• Ecosystems (SimLife, SimEarth)

11

• Cellular Automata

12
Cellular Automata

• Discrete system of cells that iterates based on a
  set of rules.
• von Neumann's self-reproductive automaton
• Langtons loop
• Conways Game of Life
• Wolframs classification of 1-dimensional CAs
• Langtons l-parameter

13
Cellular Automata

• John von Neumann
• Invented CAs as a medium for
• Universal Self-Reproductive Automata.
• But his USRA was huge and never fully implemented.

14
Langtons Loop

• Christopher Gale Langton
• Organized A-life I.
• September 21, 1987, Los Alamos.

15
Langtons Loop

• Self-replicating CA (1986)
• 8 states and 179 control rules
• not universal (could only replicate itself)
• A key property of living systems was thereby
  realized in an alternate medium.

22222222 2170140142 2022222202 212 212 212
212 212 212 212 212 21222222122222 207107107
111112 2222222222222
Smallest self-replicating unsheathed loop
(Reggia, 1993) 00 Lgt00
16
Game of Life

• is a 2-dim CA
• invented by John Horton Conway
• (Didnt have a comp when inventing the game!)

17
Game of Life - Rules

• The world is a 2D-grid. Each cell lives or is
  dead .

A cell is born, if it has exactly 3 neighbors.
A cell survives, if it has 2 or 3 neighbors.
A cell dies, if it has less than 2 or more than 3
neighbors.
18
Game of Life Life Forms

• Block
• (stable)

Glider (periodic)
Blinker (periodic)
Superstring
19
Game of Life Life Forms
R-Pentomino (stablizes after 1103 generations)
Rabbits (stablizes after 17331 generations)
20
Game of Life

• Set of simple rules ? complex behavior emerges
• Even a universal computer is possible !

21
1-dim CAs

• Stephen Wolfram
• Mathematica
• A New Kind of Science

22
1-dim CAs - Rules
One generation is shown on a horizontal line.
23
1-dim CAs - Classification

• 256 sets of rules. Wolfram divided the 256 CAs in
  4 classes.
• Class 1 static
• All configurations map to a homogeneous state.

24
1-dim CAs - Classification

• Class 2 periodic
• All configurations map to simple, separated
  periodic structures.

25
1-dim CAs Classification

• Class 3 chaotic
• Produces chaotic patterns (impossible to predict
  long time behavior).

26
1-dim CAs Classification

• Class 4 complex
• Produces propagating structures.

27
1-dim CAs CA-Patterns in Nature

Cone shell markings may be produced by a chemical
analogue of a 1-dim CA.
28
Langton's l-Parameter Classifying CA Rules

• K of possibles states of a cell (2)
• N of cells considered to compute next
  state (3)
• S KN of possible neighborhood states (8)
• KS of possible CA rules (256)
• A rule defines one of K possible next states
• for each of the S possible neighbor states.

29
Langton's l-Parameter

• n of neighbor states that map to a selected
  state, s (e.g. 0)
• (S n) / S, l ? 0, 1
• l near 0 ? most states map to s
• l near 1 ? no states map to s
• l near (K 1)/K ? even distribution of s and
  other next states

30
Langton's l-Parameter

• Langton's l-Parameter for 1-dim CAs
• Langton
• Life exists
• on the edge to chaos.

31
Life on the Edge of Chaos in Nature

• Black Smokers
• Deep-sea hydrothermal vents supporting
  extraordinary ecosystems deep beneath the surface
  of the oceans.
• Temperature differences
• smokers 400C
• deep sea -1C 4C

32

• Self-Organization

33
Self-organization

34
Self-organization
Stuart Kauffman Boolean networks, 1965
The Origins of Order - Self-organization and
Selection in Evolution At Home in the
Universe - The Search for the Laws of
Self-organization and Complexity
35
Self-organization

• Neo-Darwinism
• Evolution chance mutations natural selection
• ? Life is a lucky accident
• ? Complexity takes eons to develop
• Order for Free
• Natural laws of self-organization create ordered
  patterns in complicated systems
• Evolution self-organization drive systems to
  "the edge of chaos", where maximum adaptibility
  is possible(regulation more evolution)
• ? Life is expected ! ? We are At Home in the
  Universe

36
Self-organization General Lessons

• "Much of the order we see in organisms
• may be the direct result not of natural
  selection
• (acting on random variations)
• but of the natural order selection was
  privileged to act on .... Evolution is not just a
  tinkering ....
• It is emergent order honored and honed by
  selection."
• Stuart Kauffman
• Abtract ALife simulations ? key insights into
  ontogeny evolution

37
Self-organization In Biological Systems

38

• L-Systems

39
L-Systems

• Aristid Lindenmayer
• Przemyslaw Prusinkiewicz
• The Algorithmic Beauty of Plants

40
L-Systems Grammar

A system of rules used to model growth and
development of organisms.
41
L-Systems Examples

42
L-Systems Examples

43

• Genetic Algorithms

44
Genetic Algorithms

• John Henry Holland
• Ingo Rechenberg
• Evolutionsstrategien (1973)
• The universe has its own cure for stupidity.
  Unfortunately, it does not always apply it.

45
Genetic Algorithms Basic Concepts

• Both biological and simulated evolutions involve
  thebasic concepts of
• genotype and
• phenotype,
• the processes of expression and
• selection, and
• reproduction with variation.

46
Genetic Algorithms Algorithm

• A population of individuals is chosen at random.
• The fitness of the individuals is determined
  through a defined function.
• Fit individuals are kept and unfit ones are not.
• The new population undergoes the process.
• Individuals are in practice arrays of bits or
  characters.

47
Genetic Algorithms Reproduction with variation

• Perfect copy 10010101 ? 10010101
• ? nothing will ever change
• slightly variations are needed
• Mutation 10010101 ? 10011101
• Crossover String1 10110001 ? 1011 1011
• String2 01101011 ? 0110
  0001

48
Genetic Algorithms Example Ants

• David Jefferson, UCLA
• An ant has to follow (on a 2D-grid) a given trail
  (cells are marked kind of pheromone).
• An ant is simulated by a FSM which is described
  by 450 bits.
• input state of cell in front of the ant
• output turn left/right, go forward, dont move
• The top 10 of a population are taken for the
  next generation.
• Their genes are crossed over and mutated.

49
Genetic Algorithms Example Ants cont.

• The 1st population of 65536 ants is randomly
  created. Most ants dont move at all.
• After 70 generations, most ants follow the
  complete trail.
• ? artificial evolution

50
Genetic Algorithms

• Danny Hillis
• Thinking Machines

Connection Machine
51
Genetic Algorithms Ramps

• Ramps little programs to sort 16 numbers
• Populations of 65536 Ramps
• After 5000 generations, the best Ramps needed
  only 65 exchanges to sort the numbers. Best
  humans solution 60.
• The Ramp-population was stuck in a local maximum.

52
Genetic Algorithms The Red Queen Hypothesis

• Red Queen in Alice in Wonderland Constant
  running is required to remain in the same place.
• Biology evolutionary arms races
• When two populations of different species are set
  against each other, in predator-prey or
  host-parasite relationships

53
Genetic Algorithms Anti-Ramps

• Hillis introduced anti-Ramps into his
  systems. Programs which created test cases
  for the Ramps.They were rewarded for difficult
  test cases.
• Best solution found 61 exchanges
• ? Accelerated evolution thru coevolving parasites

54
Genetic Algorithms Fitness Curve

• Periods of stability are
• shattered by sudden leaps
• in fitness.

55

• Artificial Evolution for CG

56
Artificial Evolution for CG Genetic Images

• Genetic Images is a media installation in which
  visitors can interactively "evolve" abstract
  still images. A supercomputer generates and
  displays 16 images on an arc of screens. Visitors
  stand on sensors in front of the most
  aesthetically pleasing images to select which
  ones will survive and reproduce to make the next
  generation.
• Karl Sims, 1993

57
Artificial Evolution for CG Exploring Parameter
Space

• Procedural models such as fractals and procedural
  texturing allow a user to create a high degree of
  complexity with relatively simple input
  information.
• One method of procedural structure creation
  involves
• a set of N input parameters
• each of which has an effect on a developmental
  process
• which assembles the structure.
• The set of possible structures corresponds tothe
  N-dimensional space of possible parameter values.
  

58
Artificial Evolution for CG Exploring Parameter
Space

• more options added ? more variations
• of input parameters grows ? it becomes
  increasingly difficult to predict the effects of
  adjusting particular parameters and combinations
  of parameters
• An alternative approach is to sample randomly in
  the neighborhood
• of a currently existing parameter set
• by making random alterations to a parameter or
  several parameters,
• then inspect and select the best sample or
  samples of those presented.

59
Artificial Evolution for CG Artificial Evolution

• genotype is the parameter set
• phenotype is the resulting structure
• selection is performed by the user(picking
  preferred phenotypes from groups of samples)

60
Artificial Evolution for CG Evolving 3D plant
structures

• Parameters - used to generate 3-dim tree
  structures - describe
• fractal limits
• branching factors
• scaling
• stochastic contributions, etc.
• Growth rules use 21 genetic parameters and
• the hierarchy location of each segment in the
  tree to determine
• how fast that segment should grow
• when it should generate new buds
• and in which directions.

61
Artificial Evolution for CG Mutation

• Mutating parameter sets (one possibility)
• normalizing each parameter for a genetic value
  between 0.0 and 1.0
• copying each genetic value or gene, g_i,with a
  certain probability of mutation, m
• mutation is achieved by adding a random amount,
  /-d, to the gene
• for each g_i
• if random(.0,1.0) lt m
• then g'_i g_i rand(-d, d)
• clamp or wrap g'_i to legal bounds
• else g'_i g_i

62
Artificial Evolution for CG Mating

• Possible methods
• Crossovers
• Copying each gene independently from one parent
  or the other (fig. right)
• Each gene can receive a random percentage, p, of
  one parent's genes, and a 1 - p percentage of the
  other parent's genes
• each new gene can independently receive a random
  value between the two parent values of that gene

63
Artificial Evolution for CG Forest of "evolved"
plants
64
Artificial Evolution for CG Expressions as
Genotypes

• A limitation of genotypes consisting of a fixed
  number of parameters and fixed expression rules
  is that there are solid boundaries on the set of
  possible phenotypes.
• ? include procedural information in the genotype
• Symbolic lisp expressions as genotypes
• A set of lisp functions and a set of argument
  generators are used to create arbitrary
  expressions which can be mutated, evolved, and
  evaluated to generate phenotypes.

65
Artificial Evolution for CG Evolving Images

• X Y abs(X)
• X abs(Y) X AND Y bw-noise(0.2, 2)
• color-noise(0.1, 2)
• grad-direction( bw-noise(0.15, 2), 0.0, 0.0)
• warped-color-noise( (X 0.2), Y, 0.1, 2)

66
Artificial Evolution for CG Mutation

• Lisp expressions are traversed as tree structures
  and each node is in turn subject to possible
  mutations.
• Any node can mutate into a new random expression.
• Scalar values add some random amount.
• Vector add random amounts to each element.
• Functions can mutate into different functions.
• An expression can become the argument to a new
  random function. Other arguments are generated at
  random if necessary.
• An argument to a function can jump out and become
  the new value for that node.
• A node can become a copy of another node from the
  parent expression.

67
Artificial Evolution for CG Mating

• Two methods were used for mating symbolic
  expressions.
• The first method requires the two parents to be
  somewhat similar in structure. The nodes in the
  expression trees of both parents are
  simultaneously traversed and copied to make the
  new expression. When a difference is encountered
  between the parents, one of the two versions is
  copied with equal probability.
• parent1 ( (abs X) (mod X Y))
• parent2 ( (/ Y X) ( X -.7))
• child1 ( (abs X) (mod X Y))
• child2 ( (abs X) ( X -.7))
• child3 ( (/ Y X) (mod X Y))
• child4 ( (/ Y X) ( X -.7))

68
Artificial Evolution for CG Mating

• The second method for mating expressions combines
  the parents in a less constrained way.
• A node in the expression tree of one parent is
  chosen at random and replaced by a node chosen at
  random from the other parent.
• This crossing over technique allows any part of
  the structure of one parent to be inserted into
  any part of the other parent and permits parts of
  even dissimilar expressions to be combined.
• With this method, the parent expressions above
  can generate 61 different child expressions -
  many more than the 4 of the first method.

69
Artificial Evolution for CG Examples
70
Genetic Algorithms Galápagos

• Galápagos is an interactive media installation
  that allows visitors to "evolve" 3D animated
  forms.
• Karl Sims, 1997

71
Genetic Algorithms Galápagos

• These images show a "parent" in the upper left
  corner, and the remaining 11 are "offspring" from
  that parent.
• Mutations cause various differences between the
  offspring and their parents.

72
Genetic Algorithms Galápagos
73
Genetic Algorithms Morph-Lab

• Morph-Lab is a little Java-applet.
• 16 artificial organisms called BioMorphs.
• Morphs have a genotype and a phenotype.
• Each morph has 16 genes that code its structure
  and color.
• User selects a morph for asexual reproduction.
• This morph and its mutant progeny become the next
  generation.

74

• Simulations Instantiations

75
Simulations Instantiations

• Boids
• Framsticks
• PolyWorld
• AntFarm
• Biotopia
• Avida
• Tierra

76
Boids

• Computer model of coordinated animal motion such
  as bird flocks and fish schools.
• Generic simulated flocking creatures are called
  boids.
• Set of simple rules
• complex behavior emerges
• self-organized flock of birds
• Craig Reynolds, 1986

77
Boids Steering Behaviors

• 3 simple steering behaviors

Separation steer to avoid crowding local
flockmates
Alignment steer towards the average heading of
local flockmates
Cohesion steer to move toward the average
position of local flockmates
78
Framsticks

• The objective of these experiments is a study of
  evolution capabilities of creatures in simplified
  Earth-like conditions.
• They are
• a three-dimensional environment
• genotype representation of organisms, physical
  structure (body) and neural network (brain) both
  described in genotype
• stimuli loop (environment receptors brain
  effectors environment)
• genotype reconfiguration operations (mutation,
  crossing over, repair)
• energetic requirements and balance
• and specialization.

79
Framsticks

• Creatures are made of sticks (limbs).
• Muscles (red) are controlled by a neural network,
  which makes them bend and rotate.

80
Framsticks

• Sticks can be specialized for various purposes
• assimilation (green)
• strength (thickness)
• ingestion (small yellow spots) etc.
• The energy ball on the right is ingested faster
  because of the specialized stick ending.

81
Framsticks

• Creatures may have different receptors.
• The creature shown has a sense of touch (on the
  top) and a sense of equilibrium (glass-like
  cell).

82
Framsticks

• A small creature ("Antelope") attacking another,
  big one ("Spider").

After the collision "Spider" broken apart into
pieces. "Antelope" consumes energy from the dead
body.
83
Framsticks
84
Framsticks
85
Principles

• Emerging behavior
• Self-organization
• Artificial evolution
• Accelerated evolution
• Sudden leaps in fitness

86
.end

• Advances in Evolution

87
References

• Books
• Artificial Life - A Report from the Frontier
  where Computers Meet Biology
• by Steven Levy, ISBN 0679743898
• Introduction to Artificial Life
• by Christoph Adami, ISBN 0387946462
• Self-Organization in Biological Systems
• by Scott Camazine et al., ISBN 0-691-01211-3
• http//www.scottcamazine.com/personal/selforganiz
  ation/SOMain.html
• At Home in the Universe, ISBN 0195111303
• The Origins of Order, ISBN 0195079515
• by Stuart Kauffman
• The Algorithmic Beauty of Plants
• by Przemyslaw Prusinkiewicz, Aristid Lindenmayer

88
References

• Sites
• Alife-intro
• http//www.mikrotron.com/alife/index.html
• http//www.webslave.dircon.co.uk/alife/intro.html
  
• Self-replicating CAs
• http//lslwww.epfl.ch/pages/embryonics/thesis/Cha
  pter3.html
• Conways Game of Life
• http//www.ericweisstein.com/encyclopedias/life/
• Genetische Algorithmen
• http//www.chevreux.org/diplom/node32.html
• Genetic Images
• http//www.genarts.com/karl/genetic-images.html
• Galapagos
• http//www.genarts.com/galapagos/index.html
• Morph-lab
• http//alife.fusebox.com/

89
References

• Persons
• John H. Holland (genetische Algorithmen)
• http//www.evalife.dk/bbase/list_author.php?au_id
  393 publications
• Thomas Ray (Tierra)
• http//www.isd.atr.co.jp/7Eray/
• Craig Reynolds (Boids)
• http//www.red3d.com/cwr/index.html
• Karls Sims (Genetic Images, Galapagos)
• http//www.genarts.com/karl/
• Demetri Terzopoulos (künstliche Fische)
• http//www.cs.toronto.edu/dt/
• http//www.cs.toronto.edu/7Edt/
• John von Neumann (1903--1957) (CAs and more)