[extropy-chat] Singularity Drugs

[Robert Bradbury]

On 4/15/06, John K Clark <jonkc at att.net> wrote:
>
> Even after 4 billion years of effort Evolution never came up with
> supersonic
> flight, or plants that produce light, or a wheel you could see without a
> microscope, or even fire. Evolution is a lousy inventor, it's just that
> until is stumbled upon brains it was the only inventor.


John, can you *prove* that evolution never came up with supersonic flight?
Falcons in a dive for example achieve relatively high speeds (~100 mph???).
Just because we don't see it as an current characteristic in Nature doesn't
mean that it wasn't invented at some point in time (it doesn't violate any
laws of physics -- it just requires very high strength materials -- perhaps
lightweight bird bones coated with a spider silk fiber matrix).   As there
is little survival advantage associated with supersonic flight it is not
surprising that it hasn't evolved -- that doesn't mean that it couldn't
evolve if you created an environment which gave distinct survival advantages
to creatures able to attain supersonic speeds.

If you think about the history of the human creation of supersonic
capabilities it was developed primarily to destroy other humans (or defend
oneself against them)).  It isn't exactly needed to kill woolly mammoths,
bears, wolves, etc. for example.  Spears and guns solved those problems
quite well.

It is reasonable to view the brain as an extension of "natural" evolution
which in humans has the primary advantage of "lightweight" (and therefore
rapid) mutation and selection.  Imagination (manipulation of virtual
realities) allows one to create, modify and select good ideas before one has
to pay the cost (in materials and time) of the real-world testing that such
ideas entail [1].  A good example would be Edison's invention of the light
bulb -- according to the story he tried something like 1000 different types
of filaments before he found that carbonized thread worked effectively.  The
invention process would have been much faster if he hadn't had to actually
build the light bulbs to test the ideas.  Mutation and selection (evolution)
is much faster when you can do it virtually esp. if you have the means of
applying selection criteria which match real-world physics and/or relatively
fast computational capabilities.

One thing that people are missing which relates to our other discussion
about the development of drug resistance is the ability to simultaneously
evolve along multiple vectors.  It is very hard for natural evolution to
realize that it has to change 3 genes simultaneously to get around the
selection function imposed by the HIV "triple cocktail".  To get a protein
structure which is resistant to the drugs which also does its intended job
may require changing multiple amino acids (and therefore DNA bases) within
those genes simultaneously.  Where humans have an advantage, because
imagination is relatively cheap (and computer "thought" time even cheaper)
is that they can follow paths which have no immediate benefit and view the
entire multi-dimensional space of evolving towards a goal.  For example it
was realized some complex problems involving systems of linear equations
(minimax problems) could be solved by proceeding along multiple sequential
vectors until the best solution was found.  The only problem is this doesn't
work if the solution space includes optimal solutions which cannot be easily
reached from the initial starting point.  The Simplex Algorithm [2] was
invented in 1947 to solve the "simple" cases (and natural evolution can be
viewed as following a similar scheme) but it wasn't until the late '70s and
'80s that the Ellipsoid Algorithm [3] and Interior Point Methods [4]
provided relatively low cost methods for exploring the entire phase space of
solutions. [5]

Now, with respect to the brain and "smart drugs" one of the fundamental
problems you have in communication rates is diffusion time across the
synapses in the brain.  Neurotransmitters are small molecules but they are
larger than they probably need to be (and thus diffuse more slowly).  So it
would be better if you could engineer the brain to use "lightweight"
neurotransmitters.  NO comes to mind as a lightweight signalling molecule
which the body already knows how to produce and detect (though it can cause
damage to proteins).  Drugs might be able to impact the quantity of specific
neurotransmitters released (or increase their availability such as  SSRI
drugs do) but it is going to take genome engineering to change the
neurotransmitters to "lightweight" (faster) molecules.  You could change the
signal transmission architecture completely but that would inventing a
completely different electrochemical computational system.

With respect to the learning rates of young people, you have to realize that
young people have more neurons and many more synaptic connections than those
who are older.  The rapid learning process involves the deletion of synapses
and the death of neurons which are not "useful" (i.e. those which are
exercised at some level).  The learning one does as an adult is a much
slower process that involves slow neuron replacement and tuning of the
existing neural network (changing the relative strengths of various synaptic
connections).  Think of it as the difference between casting a bronze statue
(molten metal assuming a crystallized state where there was nothing
previously) and the finishing and polishing of the statue after you take it
out of the mold.  If you want to get back to the rapid state of "form"
creation you have to remelt the bronze (regenerate an unpatterned neural
network) and it isn't easy to do that without losing the pattern which is
already present. [6]

Robert

1. William Calvin's books have very good discussions of these concepts.
2. http://en.wikipedia.org/wiki/Simplex_method
3. http://en.wikipedia.org/wiki/Ellipsoid_method
4. http://en.wikipedia.org/wiki/Interior_point_method
5. For those of you unfamiliar with these terms think of it as "You are
standing at the bottom of the Grand Canyon and want to reach the highest
peak in North America -- *without* having to climb each and every peak so
you can see the next peak which is higher.  The Simplex Algorithm would
result in your scaling the highest peak that was nearest to the Grand Canyon
(probably someplace in S. Nevada, perhaps somewhere in the southern
Sierras).  The Elipsoid Algorithm and Interior Point methods effectively
move a virtual plane through all of the peaks in N. America until it is left
touching only Mt. McKinley.  That tells you to go directly to Alaska and
avoid wasting time elsewhere.
6. I miss Anders.  :-(
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20060416/094ce0f6/attachment.html>