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

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 Darwins 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 Parents

Offspring C
MIT Design Inquiry

Parents

Offspring

Crossover point
MIT Design Inquiry

Parents

Offspring

Crossover point
MIT Design Inquiry

How do they relate to human processes of design?

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

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

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

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

good genotypes

bad genotypes

MIT Design Inquiry

b a rule 1 a b a rule 2 a a rule 3 a

a a rule 4 b a a rule 5 b a a rule 6 b

b a rule 7 b a a rule 8 a

MIT Design Inquiry

design 1

design 2

design 3

good
{1,12,2,8,5,4,4,2,8,5,7} design 4 design 5

good
{1,2,1,8,2,8,5,5,6,6,8,1} design 6

bad
{3,2,2,6,5,8,2,1,4,4,3,1}

good
{6,4,1,2,8,5,4,2,8,5,3,3} design 7 design 8

bad
{3,4,8,2,8,1,6,5,7,3} design 9

neutral
{2,3,2,3,4,3,5,6,5,1,6,2} design 10

good

bad

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 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

l l

MIT Design Inquiry

Fitnesses for Representation

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

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

g1 gc g2

p2 pc

Phenotypic space P Genotypic space G

C(g1,g2) gc
MIT Design Inquiry

C(p1,p2) pc

P+

p1

p2

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

1
HEDONIC VALUE

Hx
0 n1 n2

Nx

NOVELTY

-1

Punish

Berlynes 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