This isn't cognitive science, but might be of interest to some anyway:

My view of DNA transcription in the cell has always been fairly simple, in a
way: there are genes of DNA, and after a complex process they instruct the
formation of proteins, which do stuff.  Genes -> proteins.  Nice little
machine.  Oh, sure, my knowledge soup (to use John Sowa's term) also contained
facts about promoter regions of DNA to start gene transcription, and the fact
that one embryonic[2] cell turns into thousands of cell types in a human body,
and that different expression patterns had to be responsible, but I never
integrated those into my simple mechanical picture of genes -> proteins.

Recently I read Gary Marcus's _The Birth of the Mind_, and he used language
which caused a nice little mindquake and reintegration of what I knew.  Every
gene has a promoter region, which can be blocked or activated by proteins,
thus controlling expression.  For example, E. coli's lactase gene has a
promoter which goes forward only if lactose is present and glucose isn't.  So:

IF lactose present AND NOT glucose present THEN make lactase.

In fact, one could say that a gene should really be thought of as the protein
template *plus* the promoter region, as a unit, making the whole thing not
just an instruction for a protein but an IF-THEN rule, which can use
combinations of all three logical primitives (AND, NOT, OR[1]).  And of course
the product of one gene can be part of the condition of itself or another,
giving us loops.

So instead of a computationally simple (but chemically complex)
protein-building machine, I now have a picture of 30,000 interacting IF-THEN
production rules, and a state of up to 30 kilobits (one per whether the
protein is being expressed or not, minus some since some proteins probably
have to be built all the time, possibly plus more than one bit per gene for
reaction rates.)  It's a production system, like SOAR or ACT or Prolog,
probably not big enough to be much of a universal computer (no general memory)
but plenty big enough to be a kick-ass specialized finite-state machine, able
to maneuver around the body's coordinate systems[3], have memory of where it's
been, and figure out which role it's supposed to play based on location and
neighbors.  When people these days talk about cellular 'intelligence', they
could be talking about *this*.

Genes as not a bunch of passive protein templates, but autonomous
conditionally responsive agents.  Cool.

-xx- Damien X-) 

[1] I can't picture how vertebrates do OR.  I'm not sure I can picture how the
lactase rule would work either, chemo-mechanically.  Marcus says plants do OR
pretty simply: they just copy the gene and modify the copied promoter region.
Technically I think that'd be PLUS more than OR -- if both promoters were
satisfied you'd be producing twice as much gene -- but hey, close enough.  He
suggested this might be why plants seem to have more genes than we do.  Don't
know if this is true.

[2] Originally typed as "embyronic".  Byronic embryos?

[3] That's a cool picture as well.  I don't really know how the developing
body actually does it, or whether the system is kept in adulthood.  He does
mention each major body part (like the arm) having its own three-axis
coordinate system.  But for any one part: imagine 3 non-colinear groups of
cells, each pumping out a different chemical.  You've got three gradients, and
our new production-rule cells are complex enough to be set to prefer a
particular value for each gradient, say by controlling the number of
receptors, or the responsiveness to them.  Coordinates are relative: "I want
50% of the A gradient, 25% of the B gradient, and as far away from C as
possible."   Changes in size let everything scale more or less smoothly.