« more IQ and Populations | Gene Expression Front Page | Muslim kill Muslim? »
September 01, 2003

Blind Watchmakings

long-horned beetles

from the Insect Company
(click for larger view)

blind watchmakings
Blind Watchmakings
from Richard Dawkins
(click for larger view)

I recently discovered The Insect Company website, which has fascinating photo galleries of beautiful and interesting insects.  [ via Boing Boing ]  An example is shown at right; Cerambycidae are long-horned beetles, and this gallery shows the variations from different countries.  I was awestruck by these wonderful examples of Darwinism in action; for me this was a religious experience.

In paging through these photos, I was reminded of the amazing software Richard Dawkins wrote to accompany his 1986 classic, "The Blind Watchmaker".  (If you have not read this book, then STOP, do not pass go, and immediately order it.  You will thank me.)

Chapter 3, Accumulating Small Change, is an in-depth exploration of a synthetic organism-producer Dawkins developed to try out the ideas behind the book.  This is a Macintosh application which generated "biomorphs", 2D black and white organism-like configurations of pixels which were generated algorithmically from a set of variables ("genes").  From any biomorph "children" are generated by mutating the variables.  You then select which children survive to generate children themselves, and thereby "breed" generation after generation of evolving biomorphs.

BTW, I just ran the Mac application again.  This 17-year-old program still runs!  (OSX emulating OSn and PowerPC emulating 68000.)  Pretty nice GUI, despite being black and white, and awesome functionality.

An example of biomorphic evolution is shown at right.  Each of these "organisms" differs from the previous by a single mutation in one of the "genes".  The visual similarity to the beetles is profound, and to my mind not coincidental.

There is one big qualitative difference between the beetles and the biomorphs; the beetles are naturally selected, while the biomorphs are not.  In each generation of beetles the fittest survive to have offspring.  The variation among beetles from different countries presumably reflects different environments (food, predators, habitat, weather, etc.).  In each generation of biomorphs the program user performs the selection, using morphological similarity to actual organisms as a measure of "fitness".

Or based on visual similarity to some other target; when the book was first published Dawkins offered a $1,000 prize for anyone who could "breed" an image of a chalice, "the Holy Grail".  To his surprise, a Caltech student claimed the prize within a year.  Subsequently a new prize of $1,000 was offered for breeding an image of a human, but this has not to my knowledge been claimed.

Biomorphs are generated from sixteen variables ("genes"), each with a range of 20 values ("alleles").  There are thus 16^20 possible biomorphs.  Mimicking biology, one of the genes controls the magnitude of mutation which can occur in one generation (variation in alleles), and another the range (number of genes which mutate).  These genes can of course themselves mutate, so that some biomorph populations are relatively stable from one generation to the next, while another might vary wildly.  The capacity of the program to surprise you from one generation to the next will, er, surprise you.  Fans of Stephen Gould will also note the "hopeful monster" mode, in which an entirely new biomorph is randomly generated!

Great stuff.  What is most amazing is that evolution has resulted in creatures sophisticated enough to generate algorithmic models that mimic evolution!


Posted by ole at 11:25 PM

Maybe you could comment on how this relates to Stephen Wolfram's theories. He is quick to show how simple algorithms can generate complex and interesting patterns. Do you know if Dawkins and Wolfram are really making the same point?

Posted by: eric f at September 2, 2003 06:56 AM

Stuart Kauffman also does things like this. He's very ambitious and his work is quite demanding. I cannot judge it but it's fascinating.

Posted by: zizka at September 2, 2003 10:13 PM

shouldn't we be bit more rigorous in our analogy? Pixel generating instructions and actual genes are a bit different. Clearly the phenotypical changes in the insect schematic are not simply cosmetic, i.e. a change in antennae length is the work of a number of different processes.
So I would say, to my mind they are merely coincidental.

Posted by: cutler at September 8, 2003 07:13 PM