From: Anders Sandberg (asa@nada.kth.se)
Date: Mon Mar 11 2002 - 06:06:15 MST
On Mon, Mar 11, 2002 at 12:03:24AM -0800, Samantha Atkins wrote:
>
>
> Anders Sandberg wrote:
>
> >Complex problems sometimes have simple solutions, but it is very rare.
> >As a conjecture I would say that the solution to a problem usually
> >requires roughly the same algorithmic complexity as the causes of the
> >problem. This is why it is so easy to solve an engineering problem - the
>
>
> But often the "cause" is relatively simple once it is found
> compared to the complexity of the effects. And often the cause
> is much simpler to address. I don't believe that an engineering
> solution will fix some of the most pressing problems. But I
> also don't believe it necessarily or even generally takes equal
> complexity to solve them.
Do you really think there is a simple cause behind poverty, crime or
racism? And if there are such simple causes, do you really think they
would be simple to address?
My conjecture that solutions might require equal complexity to solve
complex problems is of course just a conjecture. I don't know if it is
true. But it seems to fit with my experience.
> >laws of physics are fairly simple, and the specifications usually not
> >too complex. Social problems on the other hand are extremely complex,
> >involving many autonomous agents controlled by complex minds that create
> >complex interaction structures all the time. This does not mean social
> >problems are unsolvable, but that usually partial solutions or solutions
> >that we know are worse that the imagined global optimum will have to do.
>
> Unless you significantly change the nature of the agents and the
> context of their interaction.
Changing the nature of agents is often not ethically acceptable or
possible, and in many cases such changes would be highly complex and
individual. Similarly it is hard to change interactions from the
outside. It is easy to say that if every racist person were changed to
not hate/fear/loathe other people the problem would go away, but it is
not a solution to the problem - a solution would involve a recipe for
action, and that is something entirely different when we are dealing
with humans.
> >The best solutions are those that harness the inherent complexity of the
> >system itself to regulate it. Instead of imposing a simplistic order
> >from the outside, they allow a complex order to grow from inside.
>
> All other things being equal, this is true. But how long do we
> expect enough other things to be equal? If enough of the
> fundamental context changes then a complex order may develop
> alright but the result is not the original system regulating
> itself necessarily. The original system has been transformed.
Yes. Transformation is often a good method. If winning is hard, change
the rules.
> >A supergenius able to solve the "problem" of the state of the world
> >would need a mind with a power equivalent to a sizeable fraction of the
> >world. Its solutions may very well be utterly innovative and ingenious
>
> Again, not necessarily. Complexity can and often does grow out
> of only a hand full of active elements.
But finding these active elements is often computationally intractable.
There is likely not that many genes controlling ant social interactions,
but figuring out how these interactions actually work (and getting the
ants to build square anthills instead) is a major undertaking for us
human superintelligences - we have godlike resources from an ant's
perspective, but even these are not enough to quickly solve the problem
(I think it will be solved, but it will take a while). And that was just
dealing with ants.
I also don't think human behavior, especially when in groups with
culture and dynamic institutions, is factorable into a handfull of
active elements.
> >vastly more powerful entity, the sheer mass of complexity in the world
> >places some huge demands on it.
>
> It depends on how you grok the complexity and what is being
> attempted relative to that complexity.
Groking is about finding a very compressed representation of the world;
true groking would be the shortest possible representation. But many
things in the world are not algorithmically compressible. It might be
simple in principle to understand the geometry of a tree as a few basic
rules, but there is also an incompressible part of the individual unique
differences from these rules (twigs that were snapped by a passing
animal, the slight deformation due to a now gone treehouse). It is even
worse with the human mind: there might be less than 30,000 genes
involved in growing it, but the 1e11 neurons emerge in an emergent
process that is very sensitive to the environment and later experience -
I would be surprised if it was very compressible. Similarly, I have a
hard time believing the human psychology built on this basis would be
very compressible.
The end result is that I don't think groking the complexity of entire
societies is feasible, not even for SIs unless they are of a
ridiculously high order.
-- ----------------------------------------------------------------------- Anders Sandberg Towards Ascension! asa@nada.kth.se http://www.nada.kth.se/~asa/ 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 2.1.5 : Sat Nov 02 2002 - 09:12:55 MST