Eric Duchon. - John Boccio Website

rucksackbulgeAI and Robotics

Dec 1, 2013 (3 years and 8 months ago)

67 views

Evolution and

Complex
Structures:

Simulated Evolution Hints
at Features?


Eric Duchon

March 17, 2008


Complex Structures


Darwin:

To suppose that the eye
with all its inimitable
contrivances for adjusting the focus to different
distances, for admitting different amounts of light, and for
the correction of spherical and chromatic aberration,

could have been formed by natural
selection, seems, I freely confess, absurd in
the highest degree.

Even today, it is not clear how
many of the complex structures
in Nature evolved.

The Eye

How do genetic mutations
create more complex eyes
without intermediate steps
destroying their advantages?

Arguments For Simulation


Fossil records not complete enough to track
emergence of complexity


Lab experiments limited by number of generations
and by ability to track mutations through generations



Computer simulations allows exact tracking of
mutations


Limited by computer resources and a simplified
model

Computer Models


Evolutionary simulations are usually modified cellular
automata. Although not useful for directly modeling
biological systems, they can offer support for
suspicions and theories. In particular, work with Avida
has elucidated how complexity can arise.

Digital Organisms


The genome is a circular sequence of instructions (26
possible)


Energy: received single instruction processing units
(SIPs) relative to the rest of the organisms


Rate of errors when replicating the genome


0.175: an instruction to be copied is switched for another


0.05: single instruction is deleted or added


Environment determined by what merited additional
SIPs

Competition and Fitness


Competition was introduced by assigning additional
computational time to organisms which demonstrated
logical functions


The SIPs an organism received was proportional to
the product of genome length and computational
merit.

Reading a Digital Genome

Locating Complexity


Computational merit was assigned on the basis of
complexity of the genome required to produce the
logic function.


With the possible instructions, NOT and NAND were
the easiest to create while EQU was the most difficult
(it required at least 19 instructions). So to investigate
complexity, the emergence of the EQU operation was
tracked.

Case Study: A genotype with all
operations


This genotype achieved all
logical operations. Not all
the mutations were
advantageous, as seen on
top right. However, even the
deleterious mutation that
knocked out the NAND
function was essential for
forming EQU in the next
replication.


Conclusions


Support for Darwin’s general idea that
complex structures evolve from simpler
ones.


A reasonable demonstration of the
usefulness of cellular automata?

More Generally,


Out of 50 populations, 23 gained EQU.


The final genomes ranged from 49 to 356
instructions, so tendency to larger genomes.


Median of seven of eight simpler functions
already apparent before EQU.


The mutation to EQU caused 20 of 23
genotypes to lose at least one simpler
operation.


But when only EQU was rewarded, no
populations evolved that trait.