Using Evolutionary Computation as a Creativity-Support Tool

Using Evolutionary Computation

as a Creativity
-
Support Tool

Tim Chabuk

University of Maryland

Jason Lohn

Carnegie Mellon University

Derek Linden

X5 Systems

Jim Reggia

University of Maryland

NSF CreativeIT Workshop, January 2009

Pilot

Example: Self
-
Replicating Machines

•

artificial structures/systems that produce copy of themselves

•

why study?

-

insight into biological replication

-

understanding origins of life

-

understanding principles/algorithms

-

potential applications

•

•

•




A B


B


C


t


t
+1

•

historically, most work done with
cellular automata

•

discrete space of cells, discrete time

•

local, parallel computations

Rule set:


A . B C .


B


etc.

Game of Life

Two Primary Classes of Replicators

tape

Tape
control

Construction
control

Construction
arm

Replica
under
construction

John von Neumann (1950’s)


-

29 state cells


-

universal constructor computer


-

complicated


(~100k cells, 100k rules)

Chris Langton (1980’s)


-

8 state cells


-

self
-
replicating loop


-

simpler, implemented

Replication of Loop

t = 80

t = 115

t = 150

t = 6

t = 3


search for simplest, emergence from “primordial soup”,


simultaneous tasks, etc.


A Half Century of Cellular Replicators

•
hand
-
crafted rule sets

•
restricted to these two broad classes

Pan Z, Reggia J, Gao D. Properties of Self
-
Replicating Cellular

Automata Systems Discovered Using Genetic Programming,

Advances in Complex Systems
, 10 (Suppl. 1), 2007, 61
-
84.

Question:

Is automated discovery of novel self
-
replicating structures possible?

Approach:


-

given an arbitrary initial configuration


-

use genetic programming to evolve needed rules

Evolved Replicators

initial structure

evolved rule tree

GP ran quickly on standard PC but how

do the resulting replicators work?

Resulting Replicator Behavior

•

growth, separation of replicants

•

fastest CA replicant ever reported

t

= 2

t

= 1

t

= 0

Another Initial Structure


debris


larger,


repetitive


components




whole new family


of self
-
replicating


configurations

Other Structures?

Evolves same rule table as that

created previously by people?





Replicator Factory:


-

creates rules for arbitrary configurations to replicate


-

rule sets are parsimonious and fast


-

novel replication process

relative to past manual approaches


-

used to study properties of replicators

Pilot: Causally
-
Guided Evolutionary Creativity

•

causality
in human creativity

•

Goal:

guide EC in part using problem specific causal relations

•

Process:


1. specify problem
-
specific causal relations



2. integrate causal influences on evolutionary process



3. validate approach

•

Progress to date:


-

antenna array design as target task


-

derivation of causal relations


-

causal reasoning algorithm


-

integrated with evolutionary process


-

initial experimentation