Re: poly: Modeling Economic Singularities

 hanson@econ.berkeley.edu (Robin Hanson) writes:
>Peter McCluskey writes:
>> It is easy to imagine how nearly all the benefits of the technology
>>that produces a "singularity" could be unowned, which would significantly
>>reduce the race to be first.
>> A general purpose molecular assembler, if sufficiently easy to replicate,
>>would provide no easy way for a company that developed it to charge for
>>once a fair number are in use, since it would provide the ability to
>>produce nearly unlimited copies of itself.
>
>A lack of ownership would deter investment, but would not stop a race to
>be first with any investment privately worth taking.

 But doesn't it substantially delay the point at which the investments
become worth making compared to when that would be if the investors got
most of the benefits?

>As with your example of galaxy colonization, let me focus your attention on
>the possibility that there would be less capable versions available before
>an easy to program, easy to replicate assembler. We expect investors to
>work with the earliest assemblers that can make them the marginal return,
>even if doing so tends to lock in standards that make easier assemblers
>more difficult to develop and market, and may even use up most of the "sand"
>you hope the better version to build from.

 The assemblers that I had in mind to trigger nearly free computing power
could be specialized enough that they wouldn't be programmable in the
normal sense of programming. There would definitely be many generations of
increasingly easy to program assemblers, but I'm assuming the availability of
machine intelligence will make programming problems much less of a bottleneck.
 What would people be building that would use up much of the world's sand
before assemblers were able to produce computing power?
 As for ease of replication, I think there's a reasonable chance that
the techniques for producing the first assemblers could have mass production
as their normal mode. Some of the solution-based paths to the first assembler
appear to have trouble producing less than a nanomole (6*10^14) of machines
at a time.
 I think the biggest doubts about whether the earliest assemblers would
produce this kind of singularity should revolve more around the processing
that might be required to turn raw materials into something they can use.

>I can't see why you think the liquidity of markets for capital does anythinng
>to reduce the potential for capital bottlenecks. Even if I can trade my
>computer for a fatter water pipe quickly and at low transaction cost, I can still
>need both computers and water inputs into my production process. If computers
>get cheaper and water doesn't, water becomes a bottleneck.

 I'm not refering to markets. I'm talking about the replacement of quite
specialized tools whose location is important with more generic tools,
(e.g. a fully general purpose assembler would replace a wide variety of
special purpose manufacturing machines with one generic type of manufacturing
capital), and increased ability to move things of value around easily.

>There is a sense in which most existing bottlenecks are already programming
>problems. That doesn't make them go away any faster.

 The supply of programmers has been rather stable and inelastic. If there
were little cost to throwing more human-level intelligence at a problem,
most programming problems I know of would be solved much faster.

>There are so many things this scenario leaves out!
>1. How *fast* do assemblers, nanocomputers, nanominers, and software
>"intelligence" get better? Sure *eventually* most bottlenecks may get reduced,
>but how long is eventually?

 I'm imagining one unusually large jump in the importance of assemblers
as they make the transition from primarily research phase to widespread
use. There will still be much trial and error needed to go from the
specialized initial versions to general purpose versions, but I think
each succeeding generation has the potential to be easier to build than
the prior generation because of the improved tools each generation provides.
I also think AI could speed up the generation time by something like an
order of magnitude.

>2. How many kinds of "intelligence" are there, and how sure are you they are
>all limited mainly by CPU cycles?

 I'm assuming that a few dozen uploaded humans or the artificial equivalent
would provide adequate diversity of abilities to accomplish the kind of
changes I have in mind.
 While some approaches to developing AI may produce more specialized results,
I think an evolutionary approach shows some hope of producing a general
enough intelligence (the crude version of this is Drexler's claim that it
would take something like 2 years to simulate all the evolution that has
happened on out planet on a rod-logic computer; I expect something more
directed than this).
 I can also imagine that a number of human brains are scanned with something
like the methods Merkle proposes in http://www.merkle.com/merkleDir/brainAnalysis.html
and their use is being delayed by a few problems with running the simulation,
and that more cpu power is enough to solve those problems.

>4. There really are other inputs besides raw production and computation:
>raw materials, energy, volume, marketing, advertizing, tech support, regulatory
>support, military defense, and figuring out what people will pay how much for.

 Raw materials, energy, and volume appear to be sufficiently abundant that
assemblers which did a moderately efficient job of handling them wouldn't
be limited by them before increasing the availability of goods by several
orders of magnitude.
 The larger the benefit to users of switching to new products, the less
important marketing, advertizing and deciding on the right price are, so
I don't see how they will limit an economic singularity.
 The need for tech support varies widely from product to product. Even
if AI doesn't make it cheap, there will be some products which don't need
it and whose growth rates won't be limited by it.
 Regulatory and military concerns are a big uncertainty. I can imagine
anything from people being largely able to evade them to their being the
main limitation to economic growth.

-- 
------------------------------------------------------------------------
Peter McCluskey          | Critmail (http://crit.org/critmail.html):
http://www.rahul.net/pcm | Accept nothing less to archive your mailing list