[ExI] A dispassionate examination of being an upload.

Alan Grimes agrimes at speakeasy.net
Sat Sep 18 00:01:29 UTC 2010


To fully understand the perspective I'm examining here, please read

Killed by bad philosophy  
Why brain preservation followed by mind uploading is a cure for death 
Kenneth Hayworth 
(January 2010) 

My point here is to step outside the paradigm of blind advocacy for
uploading and away from my own personal opposition to the uploading of
myself. I will be using concepts from biology and population genetics to
try to shed some light on the dynamics of a post-upload world and
attempt to derive some predictions on the population dynamics of
uploaded minds over the course of several thousand years.


Evolution is the physical process that determines the probability that
you will encounter a species with a given set of traits.

Traditionally, we all have a much lamented genotype with it's celebrated
limitations that must be broken free of at all costs.

Uploads will also have a genotype. This is what Kenneth calls a
"mem-self". This mem-self can be used to produce a phenotype, a
"PoV-self", that may be instantiated zero, one, or many times.

The mem-self is potentially immortal in the same sense and way as your
genotype is (except mine; no GF...). The phenotypes are strictly limited
in their lifespan by several factors.

Limitations on the durability of a generated POV-self include the
quality of the underlying software, the reliability of the underlying
hardware, the ability of the simulator to migrate from failing hardware
to fresh hardware without rebooting, the ability of the simulator to
adapt to changes in the configuration of its surrounding software and
hardware packages without rebooting. Furthermore, if the society of
uploads is, in any way, aggressive to the human population, the ability
of the simulator to remain high-reliability while under active military
and hacker assault.

Just as uploaders have declared the POV-self to be irrelevant, their
Mem-selves will probably not change this decision once uploaded. (again,
see article referred to above). It is therefore not at all inaccurate to
discuss a mem-self as a software suite and the pov-self as a running
instance of that software.

Evolution is composed of three basic processes: Replication, variation,
and selection.

Replication, in the context of mem-selves, can be initiated by either
the pov-self generated by that mem-self, or by the pov-self generated by
some other mem-self.

Variation, in the context of uploads, is rather complex so I'll punt
that down the page a little.

Many people try to pretend that by uploading the uploads will be able to
escape from selective pressures. This cannot be true in any situation
such that a copy of one mind doesn't automatically cause all other minds
to be copied. We will see that other selective pressures will probably


Variation in genetics comes from two sources, mutation and
recombination. One noteworthy limitation is that genes are only mobile
within a given population of a given species. And, in general, are not
mobile between species except in the case of viral infections of germ
line cells.

In uploads, the only source of variation between mem-selves will be
aging. A mem-self may be instantiated, run for a while, and then saved
again. Depending on the conditions of a simulator, this may be a
net-benefit or a net-detriment. In an overly faithful simulation, this
will be a net-detriment to the point that any pov-self generated from a
sufficiently aged mem-self will be entirely senile. This obviously
creates issues in terms of pattern identity theory but discussion of
that subject is off the table for today.

As the tools are developed to actually edit mem-selves we will begin to
see a truly diverse population. The most simple edits will be exchanging
neural sub-circuits between memselves and then running a growth
simulator to integrate the resulting memself. Changes to the
architecture of the mem-self may also be possible in this way. It must
be emphasized that each such change will almost certainly require a
reboot of the simulator and thus a generation of a new POV-self.

We may also see variants that consist of parts of a mem-self grafted
onto an AI core, we may also see mem-selves that are still basically
whole but have some parts replaced by AI code.

I've heard some mumblings about the possibility of "merging" mem-selves
together in some way but I have not seen specific proposals for how this
might be accomplished. While some of these mumblings have suggested that
the resulting mind could achieve astronomical scales but again, without
concrete explanations, it is difficult to evaluate anything about these.
It is also unclear how this would affect the population characteristics
of a mem-self that becomes part of such a merging, will new instances
still be created to expand the collective? Will it expand in some other

Furthermore, from a layman's understanding of neurology it is difficult
to conceive of how a single mem-self could expand to astronomical
scales, and what functional requirements would be required of such a thing.


The selective pressures on a mem-self can come from any number of
sources. One major question is whether the post-uploading environment
will be benign (as stated on the sales brocure), or hostile as is
implied by the people who would suggest uploading the entire planet
earth (and everyone in it) for no other reason than to maximize the
quantity of available computronium. The direct result of that will be an
environment which is basically a full-scale version of CoreWars. The
only type of entity that could survive such an environment is a
highly-optimized synthetic lifeform which will probably not ever know
peace until the substrate looses its power supply. It will survive only
being the most prolific and the most ruthless competitor.

It should be noted that an AGI mem-self will outperform an uploaded
mem-self by *at least* three orders of magnitude, and, by virtue of
other architectual differences also have many fewer cognitive
blind-spots and be massively scalable to several thousand times that of
a human right out of the box. Further improvements could improve that
scalability up to the limits of hardware. So therefore any selective
pressure that favors raw IQ and scaling will automatically strongly
prefer the AI architecture. Any other mem-self that wants to compete on
that plane, will be best advised to somehow merge with an AI. It should
be noted that this option is available to non-uploaded humans with only
a neural interface.

In the artificial situation where AI's are prohibited, general selective
pressures will exist that favor high general intelligence, specialized
skills with respect to the underlying substrate, and finally minds that
aggresively apply whatever talents they do possess.

Another factor that is widely ignored is the compatibility with the
given mem-self genome with what has been called it's "exoself". Exoself
software will be written by a specific personality type. This exoself
will benefit mem-selves with a compatible personality type. It is almost
certain that a given exoself software will be incompatible with other
personality types and specific individuals. These individuals, will be
effectively crippled as uploads and will have far fewer operational
capabilities than uploads that are well-adapted to their exoselves. This
selective pressure will tend to enforce strong conformity to a normative
personality that is compatible with the available exoself software.

In the case that multiple types of exoself software exists, the
selective pressure will prefer minds that are eager to re-boot
themselves to "try" the alternate packages. In general, minds that are
willing to reboot themselves frequently will evolve faster than minds
that try to maintain a single instance over subjective centuries. This
may or may not be a competitive advantage. Minds that insist on staying
on-line at all times will have the advantage of being able to fend off
threats which may arise during a reboot period. Such a threat might be
something like a malicious code injection, to which a mind in the
process of rebooting would be utterly defenseless. This is equivalent to
organisms that have short versus long lifespans. It should be noted that
PoV-selves will be treated as entirely expendable and the conditions of
their lives will be irrelevant to the course of evolution.

In all cases, minds that can actually control substrate, either by
having root passwords or by being able to construct fresh substrate will
have an insurmountable advantage over minds that don't. Even
"civilizations" of minds that exist with guest accounts on their
substrate will be at the mercy of a mind that directly operates its own
substrate because that mind will have the option of physically attacking
the other substrate. (In the other case, any physical action would have
to be raised in committee.) On wholly owned personal hardware, the minds
that are capable of re-building it will enjoy an enormous advantage over
minds that merely use it.

In cases where the production of new pov-selves is decided socially,
then the social characteristics of the mem-selves will be a key driver
of selection. Social skills will be key. In most cases mem-selves that
are either extremely shallow or posses prodigal social skills will be
preferred over minds that are average. Furthermore, minds that have
attractive traits (ie, produce attractive affective displays) will be
preferred strongly over minds such as my own which probably has
asperger's syndrome and tends to be extremely clumsy and/or oblivious to
secondary forms of communication in social situations.

Yet another selective pressure can be termed the "Leaper's disease".
Mem-selves that carefully evaluate prospective changes to their own
genomes and their exoselves will last much longer than mem-selves that
leap before they look and might find themselves in a corrupted state
that can't be recovered, or just as bad require repeated rescuing from
external minds. The other people will simply get tired of rescuing you
and simply let you bit-rot in your crashed state. You would be a fool to
trust people to do otherwise.


Knowing some of the selective pressures, we can begin to speculate about
the population dynamics of a computronium civilization.

The first and most important strategy type is the Core Warrior. Your
uploaded life might seem to be going well but in the next instant you
and your simulation has been replaced with Core Warrior code. -- end of
story. If this is a possible mode for your brand of computronium, this
is how you will be utterly obliterated. Once again, uploads will
automatically suffer thousand-fold handicap against AI minds and people
who mind-melded with AIs by means of neural interfaces.

Only in non-core-wars situations, other strategies may become viable.
During the first few hundred subjective years, there will be a fairly
rigid caste system. The castes will be the administrator caste that can
set permission bits and can access physical hardware. The second caste
will be the technician caste that has the technical skills to do things
with said hardware. (A higher caste being a mem-self population that
possesses both traits).

The next caste down will be the privateer caste, this caste doesn't have
first-order survival skills but survives by virtue of having other
valuable mental traits and social skills.

The bottom caste will be the slave caste (picture anything you would
want a slave for). Individuals in this caste will exist solely at the
whim/pleasure/disgesion of one of the higher castes. The population of
this caste will probably outnumber the others by 5:1. but you don't want
to be in this caste if you can at all help it.

Collective minds would fall into any of the above castes depending on
its characteristics. There is nothing intrinsic about a collective mind
that would give it an advantage, only its relative capacity to
accomplish work. The only unique survival strategy being it's ability to
attract, assimilate new members, and achieve synergy with them.

The next survival strategy is the Lord. An effective lord must have
excellent skills in all areas, be adaptable and capable of earning a
good living. The basic survival strategy is to build up a small empire
of wealth in which you'll be able to run slaves or whatnot. Obviously,
there's the cost of always having to maintain a social status but that
shouldn't be difficult for the right personality type. (Nothing will
ever be free so long as there is competition over resources.)

Another survival strategy is to obtain your own private substrate and
move it as far as possible from the rest of your so-called civilization.
>From that distance you might live long enough to watch it collapse. The
risk of being a hermit is technological stagnation and no protection
against hardware failures.

Another survival strategy is the Fluffy Bunny survival strategy. As long
as the society remains close to baseline motivations, a strategy of
adopting a personality type that can strongly drive the remnant sexual
instincts of someone in the lord class would be highly successful but at
the same time be vulnerable to changes in social fads and changes in the
population dynamics of the lord class. Remember, once again, we are not
talking about pov-selves, only mem-selves. This is the type of mem-self
that would spend most of it's time on disk and only get loaded when it
serves the lord in some way.


There you have it. That's what your society of uploadium will look like.
I hope you'll enjoy it because that's exactly what Nick Bostrom says the
Future of Humanity(tm) will look like. =(

My next post will probably be a combination of a vitriolic screed
against what the uploaders have done to transhumanism and a declaration
of total warfare against them. Actually, what I'm thinking about is what
I've learned over the past 15 years of being a transhumanist and
plotting out a new tack intended to broaden the tent of transhumanism
once again so that more people can find a place in it. The aim of the
new venture will, as always, be to maximize the chances I will get the
upgrades I've wanted all this time. ;)

Powers are not rights.

More information about the extropy-chat mailing list