Re: genetic engineering was: a new thread

From: Anders Sandberg (
Date: Fri May 12 2000 - 18:45:50 MDT

=?iso-8859-1?q?Dan=20Adams?= <> writes:

> What if we ask how we can create dynamic DNA rather
> than allowing ourselves to fall down the slippery
> slope of ethical considerations attached with
> prescribing physical conditions to our progeny?

Sure, that would be ideal. But it is also rather tricky. The dream
would be something like a system where a certain plasmid with a patch
for the genome is introduced into the body, inserted into the right
place in the genome and replicated between the cells until all have
been changed. Doesn't sound entirely impossible, but darn hard to
implement well - likely a basic "operating system" for the plasmids
would have to be introduced using germline engineering so that cells
accept the plasmids when they encounter them, and to keep them
person-dependent (you don't what your latest body upgrade to infect
others, after all). This still doesn't do any major morphological

> Without making any appeal to a non-corporeal
> information existence, I can envision a
> semi-autonomous process (governed by a deep
> understanding of genetics) whereby not only are
> negative factors (e.g. potential disability and
> disease) precluded within the genetic programming, but
> only the best (read, "really good") solutions are
> found. All of this however without any hardwiring on
> the human designers part.

Many factors have unclear value, both because it is not obvious if
they are negative or positive (bloodtype A or B?), they can have
different effects under different environmental circumstances (like
various heart-disease linked genes), some have both good and bad
effects (blood clots when you are young but no heart problems when
old) and the value of some are highly subjective (blue eyes?). If you
can set a value function and simulate the results of changes in the
genome well, then you can do a GA search for an "optimum", but it
seems to be a highly underconstrained problem in a very
high-dimensional search space (~100,000 or more dimensions) where the
fitness evaluation step will be computationally very heavy.

> The field currently known as genetic programming uses
> a highly dynamic matrix of factors which allow it to
> seek out a near-global optimum in a search space
> within highly constrained parameters. The idea behind
> it (Holland, 1975) was originally derived from
> biology. Now that it's been abstracted and perfected,
> let's use it in biology again.

GAs are definitely not abstracted or perfected. In fact, Holland's
schema theory has been widely criticed for being too simplistic, and
there is currently no other all-encompassing theory for GAs (lots of
research, though). In applications, GAs are just one of many
optimiziation methods - that they are inspired by biology doesn't make
them better (a claim often heard about neural networks too). They have
their uses, but work best on certain kinds of fitness
landscapes. Hopefully metaevolution has led to the human genome
becoming suitable/the mammalian fitness landscape becoming searchable
using GAs, but there is no proof of this.

> This time to produce the next generation of people. You can design
> better people without definite specifications of what they must be
> and without falling into the dangerous "eugenics" zone...

Hmm, exactly how does this not fall into eugenics? If you do it on
adults (euthenics) then it is clearly not eugenics, but otherwise it
is just a version where humans do not make detailled decisions about
the genome.

I think an overall optimization procedure is possible to deal with
obviously negative genes and the unpleasant effects of certain gene
combinations (say a novelty seeking dopamine receptor gene and genes
leading to lower levels of self control and addiction resistance), but
there will still be a huge amount of variability.

Anders Sandberg                                      Towards Ascension!                  
GCS/M/S/O d++ -p+ c++++ !l u+ e++ m++ s+/+ n--- h+/* f+ g+ w++ t+ r+ !y

This archive was generated by hypermail 2b29 : Thu Jul 27 2000 - 14:11:10 MDT