One goal might be to provide a mechanism for adaptability during the lifetime of particular organism. Our immune system for example adapts over the course of our lifetime as it encounters new and different pathogens but it does this not through direct self DNA modification of existing cells but rather through the "directed" creation and selection for new cells which code for the new pathogens. In humans, selection has favored the devlopment of immune systems which have the ability to create massive genetic varability in certain cells in certain ways at certain times.
This may be off-topic, but if one goal of DB is to over time put in place physics which can favor increased organism complexity, multibot, multi-cellular organisms in particular, then I think that what we may want isn't direct DNA self-manipulation within a single bot but rather better mechanisms to direct mutation within specific areas of the genome at reproduction time.
Imagine a complex mutlibot where different cells have specialized. The outermost "hide" bots convert most of their energy to poison, one end specializes in getting rid of waste, the other has a cluster of tie feeders which pass energy to the rest, two bots on long ties serve as eye stalks passing back information to a cluster of central nervous system cells which do the calulations for binocular vision and tells the locomotion bots what to do... but I digress. :)
Say some of the hide bots are getting eaten. Something has evolved resistance to the specific poison we use. So, the bot reproduces new cells to replace the old ones which now poison a different memory location in the attacker. Problem solved. Bots have died and been replaced but the organism as a whole lives on.
We could do this today, explicitly code the hide-bot DNA codule to be able to use a variety of different posions and explicitly pass to the offspring which one to use, a sort of intelligent design in this one area, but we have no good way to tell it to do other things - make more posion than it was before, start growing a shell or making slime, use two layes of posionous cells instead of one, etc. I'm not talking about directed DNA manipulation but rather directed mutation in narrow, specific ways. We want the ability for a parent, when reproducing, to say "mutate here. try something different here, in this part of the genome, leave the rest alone". We don't have that today.
Of course this necessitates a mutation locality mechanism, so that the mutation rates of different parts of the genome can be selected for and potentially directed by parents. If the mutation rate was part of the genome and parents could specifiy it at reporduction time, that would be a type of directed DNA modification.
I think I would rather see work in this area, on mutation locality, than on a generic DNA self modification as a means to provide richer mechanisms for single (albeit multi-cellular) organisms to adapt during their lifetime.
Man, I cannot seem to type a short post these days...
-E