How do genetic algorithms relate to their
biological origins?
How do they relate to human processes of
design?
Why is there power in this metaphor?
How can they be used to extend the
capabilities of the designer?
John Gero
Professor of Design Science
University of Sydney
Visiting Professor of Design and Computation
MIT
MIT Design Inquiry
How do genetic algorithms relate to their
biological origins?
l
l
l
l
l
Separation of genetic material (genotype
[representation]) from organism (phenotype
[design])
Expressing genotype as organism
Organism carries genotype and reproduces
genotype using ‘genetic’ processes of crossover
and mutation
Darwin’s natural selection uses fitnesses of
organisms in their environment to improve the gene
pool
GA is a simple model of this process
MIT Design Inquiry
Genetic processes
Crossover Points
A
B
Parents
Offspring
C
D
MIT Design Inquiry
Parents
Offspring
A
A
B
A
B
C
B
B
C
B
A
B
Crossover point
MIT Design Inquiry
Offspring
Parents
A
A
B
A
A
C
B
B
C
B
B
B
Crossover point
MIT Design Inquiry
How do they relate to human processes of
design?
l
l
l
Can map genetic representation onto a
computational representation of a design; can map
phenotype onto a interpretable view of a design
Humans work on single or few designs at a time/
genetics works on a population of ‘designs’ in parallel
Humans can be seen to “search” design spaces - this
is one interpretation of what GAs are doing.
MIT Design Inquiry
Why is there power in this metaphor?
l
l
l
l
l
l
l
Guaranteed improvement - Darwinian
evolution
Large scale search
Blind search
Fitness can be human evaluation
Fitness can change over evolutionary time
Can produce complexity
Can produce unexpected results
MIT Design Inquiry
How can they be used to extend the capabilities
of the designer?
l
l
l
Creativity through genetic engineering
Novel designs through extending genetic
crossover
Novel designs through different fitnesses
MIT Design Inquiry
Genetic Engineering and Creative
Design
l
l
l
Background
l
genes, genotype, phenotype, fitness
Connecting genes to performance in fitness
Emergent gene clusters ˛ evolved genes
MIT Design Inquiry
Total Population
“bad”
“good”
x
x
x
x
••
••
x
x
“good” genotypes
“bad” genotypes
MIT Design Inquiry
b
b
a
rule 1 a b
a
a
rule 2 a
rule 3
a
a
a
rule 4 b a
a
rule 5
a
a
b
b
b
a
rule 7 b
rule 6
a
a
MIT Design Inquiry
rule 8
a
design 1
design 2
design 3
good
good
{1,12,2,8,5,4,4,2,8,5,7}
{1,2,1,8,2,8,5,5,6,6,8,1}
design 4
design 6
bad
{6,4,1,2,8,5,4,2,8,5,3,3}
{3,4,8,2,8,1,6,5,7,3}
design 8
good
{3,2,2,6,5,8,2,1,4,4,3,1}
design 5
good
design 7
bad
neutral
{2,3,2,3,4,3,5,6,5,1,6,2}
design 9
bad
design 10
neutral
bad
{3,1,8,5,5,6,4,6,1,1,3,3} {1,6,4,2,7,3,4,8,6,1,6,2} {6,4,1,2,3,4,5,2,1,7,4} {2,3,7,5,1,2,8,3,1,6,2,1}
Composite building block A
{2,8,5}
MIT Design Inquiry
MIT Design Inquiry
MIT Design Inquiry
MIT Design Inquiry
MIT Design Inquiry
MIT Design Inquiry
MIT Design Inquiry
MIT Design Inquiry
MIT Design Inquiry
MIT Design Inquiry
MIT Design Inquiry
Mondrian
Genotype Form
l
l
In form of a tree
Each node has four variables
direction of rectangular split (4 values)
fraction of the split (15 values)
colour of split area (10 values)
line width (3 values)
l
l
l
l
MIT Design Inquiry
Fitnesses for Representation
l
l
l
l
l
offset between actual and required positions of
dissection lines
number of lines with correct line width,
normalised
number of correct colour panels, normalised
number of lines assigned, normalised
number of unassigned lines, normalised
MIT Design Inquiry
Genetically Engineered Mondrian
Genetically Engineered Mondrian
Genetically Engineered Frank Lloyd Wright Windows
MIT Design Inquiry
Flondrians
l
l
l
Mondrian painting ˛ genetically engineered
genes: M-genes
Frank Lloyd Wright windows ˛ genetically
engineered genes in same representation: Fgenes
“Flondrians” are the genetic product of mating
M-genes with F-genes
MIT Design Inquiry
MIT Design Inquiry
MIT Design Inquiry
How Many Designs Are There and
Where Are They?
MIT Design Inquiry
Genetic crossover as an interpolation
p1
p2
g1
pc
gc
g2
Phenotypic space P
Genotypic space G
C(p1,p2) Æ pc
C(g1,g2) Æ gc
MIT Design Inquiry
P+
p1
p2
P
MIT Design Inquiry
Interpolation
MIT Design Inquiry
MIT Design Inquiry
(interactive Genetic Art III)
MIT Design Inquiry
Image detection
(a)
(b)
MIT Design Inquiry
Modelling Interest
Reward
HEDONIC VALUE
1
Hx
0 n1
n2
Nx
NOVELTY
-1
Punish
Berlyne’s model of arousal based on novelty using Wundt curve
MIT Design Inquiry
Different novelty functions
MIT Design Inquiry
Different novelty preferences
N=0
N=1
N=2
N=3
N=4
N=5
N=6
N=7
N=8
N=9
N=10
N=11
N=12
N=13
N=14
N=15
N=16
N=17
N=18
N=19
MIT Design Inquiry