[ExI] [tt] Identity thread again

Rafal Smigrodzki rafal.smigrodzki at gmail.com
Mon Apr 6 06:05:29 UTC 2015


On Fri, Apr 3, 2015 at 8:20 PM, Robin D Hanson <rhanson at gmu.edu> wrote:

>  On Mar 26, 2015, at 12:07 AM, Rafal Smigrodzki <
> rafal.smigrodzki at gmail.com> wrote:
>
>   ### Here I am not so optimistic - the ease of making copies means the
> substrate is likely to be crowded. Also, there will be a trade off between
> spawning independently goal-oriented processes (copies) and running
> non-sentient optimization and search processes. A single conscious mind
> could elect to use available resources for e.g. exhaustive searches of
> solution spaces to different problems. Or it could choose to make many
> copies of itself. Which strategy would spread would depend on the fitness
> payoffs: If non-sentient processes (=supercomputing) produce valuable
> excludable goods (i.e. intellectual property or new physical resources)
> that you can protect from thieves and sell for more energy/matter than can
> be bought from the labor of copies, then the supercomputing strategy would
> predominate. Otherwise, breeders would swamp the substrate.
>
>
>  In a competitive world the ratio of computer hardware spend running
> minds like humans to "non-sentient optimization and search processes”
> depends on the relative effectiveness of the later compared to the former.
> But I don’t see what high levels of theft have to do with it. If there is a
> lot of theft then how that changes things depends on what kinds of things
> are easier to steal, and what kinds of abilities are better at doing and
> preventing theft. If both are equally easy or hard to steal, and if the
> efficient ratio of minds like humans and non-sentient processes is the same
> for doing and preventing theft as it is for other useful production, then
> we wouldn’t expect more theft to change that relative ratio.
>

### I am assuming that there is a trade-off between the spawning copies and
supercomputing new solutions - if you spawn copies, ceteris paribus, you
are less able to supercompute. Assuming an identical amount of starting
resources (matter+energy)x time, the breeder can never find the solutions
accessible to the supercomputer, simply because the resources are fungible
and expending them towards breeding precludes other achievements. So there
is a difference in the relative effectiveness between the two strategies in
terms of finding solutions to problems, especially over large time
horizons. The complexity of thinking is cumulative, breeding is not.

In a world without theft (where theft may include direct invasion of the
computing substrate, not just IP theft), the supercomputer is not at risk
of losing resources it gains through making inventions over long periods of
time. Given the law of comparative advantage, a supercomputer survives
under a wide range of values for payoffs from supercomputing vs. labor of
copies. It would not lose over time if it could buy enough resources to
sustain itself, and it would overtake the world if it could buy enough
resources to grow.

You are correct, introduction of theft would not immediately change things
under the assumptions you spelled out - i.e. supercomputers and breeders
are suffering equal losses and making equal gains per unit of time due to
theft. However, for this to be the case, you would need a surprising parity
between a number of processes - e.g. breeders stealing from supercomputers,
supercomputers stealing from breeders, the efficiency of conversion of
stolen goods into breeding, the efficiency of converting stolen goods into
supercomputing, and perhaps others that I can't think about now. Therefore
I would expect that the existence of theft would change the relative
fitness of breeding vs. supercomputing. Of course, it is conceivable that
supercomputers might be better at theft than breeders. In the living world
there are situations where theft by supercomputers vastly wins over
breeders - we are the supercomputers, and we feed on massive numbers of
stupid beings. In other situations however, small stupid things eat us
alive. I cannot begin to predict which conditions would obtain in the
computational substrate. We have insufficient data. It might be useful to
think through the details of conceivable strategies of growth, breeding,
resource acquisition, and try to tie them to what we know about the physics
of computation and the mathematics of resource control (i.e. cryptography,
game theory) but this is an endeavor beyond the scope of an ExI post.

Let me just elaborate on the world without theft. New resources of matter
are won and divided between breeders and supercomputers depending on their
relative ability to take over unowned resources. Let's assume that the
amount of matter/energy available would remain constant after all unowned
resources are taken over, and there is an initial ratio of mass of
supercomputers to breeders.  How it would change over time would now depend
their relative ability to produce exchangeable units of value. Please note
that if the breeders are only interested in increasing their numbers, they
would not trade with supercomputers - there would be nothing they need that
supercomputers could offer, and under these assumptions the ratio would
remain constant until the death of the universe. However, over time there
would be a change in the internal state of breeders vs. supercomputers.
Breeders, being less able to generate new solutions, would effectively
remain in a form of stasis, unchanging in numbers or internal complexity
over long periods of time. This is approximated by a steady-state bacterial
culture fed constant amount of feed - it just stays there with minor
variations as new mutations sweep through the population. On the other
hand, supercomputers would accumulate change over long periods of time -
they would keep finding new solutions, exploring new mathematics - by not
using resources to generate conscious thought they would innovate. If the
innovations produced new resources (e.g. a magical gate to an NP-complete
oracle), the ratio of computation carried out by supercomputers to the
computation by breeders might increase.

I leave it for the reader to introduce intellectual property trade and
theft into this world and see how it works out under different assumptions
about their relative impacts on breeders and supers.

Rafal
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20150406/e2a10988/attachment.html>


More information about the extropy-chat mailing list