Evolution essay

This is an "active essay" demonstrating some ideas about evolution,
originally expressed by Richard Dawkins in his book _The Blind
Watchmaker_.  Here we're going to recapitulate his argument but with
the processes described set in motion on the computer, as can't be
done on the printed page.  (This was inspired by Ted Kaehler's active
essay in Smalltalk on the same subject.)

Many people feel that evolution by natural selection can't account for
the exquisitely complicated life we see around us, because natural
selection is powered by chance, while complicated life is too
improbable to arise by chance.  It's true that mere randomness cannot
produce complexity, and we'll see that truth in the demo below; but
natural selection isn't *just* chance.  So let's try to generate a
complex structure by random chance, then see what else is needed.

Our target is a string of letters, and we'll consider it more
`organized' or `complex' the closer it comes to "METHINKS IT IS LIKE A
WEASEL".  We'll name our `mere chance' process Drifter, and our later,
better process Evolver.  They both start out empty:

  def    goal METHINKS IT IS LIKE A WEASEL
  def Drifter                             
  def Evolver                             

We'll fill them with random picks from 'alphabet':

  def alphabet  ABCDEFGHIJKLMNOPQRSTUVWXYZ

(There's an extra space before the "ABC..." because the goal contains
spaces as well as letters.)

Here's how Drifter works: we pick a random position within Drifter:
  let i random-index Drifter	#11

then a random element of the alphabet:
  let r random-element alphabet	#D

and then set the character at the position to the selected element:
  do text-set Drifter i r

Hit Alt-O to see it working -- the "def Drifter ..." line above should
pick up a letter.  (It might just replace a space with the same space,
since that is a possible, though unlikely, `mutation'.  If you keep
hitting Alt-O it should change, though.)  The Evolver will also
update, but the machinery behind that appears below.

Now for a display of the fitness of our two organisms, defined as the
count of genes correct (i.e. positions that match the goal):

  let comparison         compare goal Drifter	#>>>>>>>>=>><>>=>>>>=>=>>>>>>
  let Drifter-fitness    count "=" comparison	#4

  let comparison         compare goal Evolver	#>>>>>>>>=>>=>>=>>>>=>=>>>>>>
  let evolver-fitness    count "=" comparison	#5

So, watching this, we can see that Drifter's fitness drifts around
1/27 of the maximum fitness, and is horribly unlikely to get close to
the maximum -- getting the maximum would be like rolling a die 28
times in a row and calling each result in advance.  (Only worse, since
a die only has 6 sides, compared to 27 `sides' here -- 26 letters plus
the space.)  Even a nation of gamblers couldn't expect that much luck
in the lifetime of the universe.

So let's try to make something better with the Evolver.  This version
will only pass mutations that don't lower the fitness.

First we create a child as a copy of Evolver with one letter changed
at random:
  let child           copy Evolver      	#                  J         
  let i               random-index child	#18
  let r               random-element alphabet	#J
  do text-set child i r

And we compute the child's fitness:
  let comparison      compare goal child	#>>>>>>>>=>>=>>=>>><=>=>>>>>>
  let child-fitness   count "=" comparison	#5

Finally, we replace the Evolver with its child unless the child's
fitness is lower than its parent's:
  let selected   notless child-fitness evolver-fitness	#true
  when selected  do replace Evolver child

Evolver's fitness now tends to maximum within around a thousand steps.
This shows that a process with random mutations can produce
extraordinarily improbable complex results if improvements are
*cumulative*.  The problem with Drifter is that any improvements in
one generation are forgotten in the next.

This can't be considered a realistic model of evolution; most
obviously, it has one particular goal wired in from the start, where
life has no goal.  All we've done here is refute the creationists'
claim that random mutations can't produce improbable results in a
reasonable time; they can, if the changes can accumulate.  This
creationist argument would make sense if biologists were saying that,
for example, one day an ape gave birth to a human, fully formed.  But
nobody says that; we must have had a long lineage shading
imperceptibly from near-apes to near-humans, with each change along
the way being reasonably likely on its own.


Some things to try: bear n offspring, selecting just one per
generation, with fitness-proportional selection.  Vary the mutation
rate.  Look for the error catastrophe.