Emergence
- From: "jimbromer@xxxxxxxxx" <jimbromer@xxxxxxxxx>
- Date: Sat, 11 Aug 2007 09:58:24 -0700
The Wikipedia entry for Emergence http://en.wikipedia.org/wiki/Emergence
is one of the best articles that I have read on the subject.
I believe in strong emergence, that many qualities and relations of a
complicated system only become apparent as the system is developed.
(I am not interested in talking about some abstruse conjecture about
the fundamental-nature-of-the-universe, or someone's simplistic
derivative of fundamental theory of some sort, what I sometimes call a
fuddamental, but in how we might write a better AI program.) I think
it should be obvious that learning is emergent, that as you learn
something new about a mental model, and incorporate it into what you
had previously known about the system, that this new information can
produce emergent understanding. Various methods of simplification can
be used to further study the system that is being considered, but
quite often many of the steps of a process of simplification must
themselves be inducted to handle the new information.
There are many systems whose qualities can be examined -as if- they
were reducible but which are not necessarily reducible. What I mean
is that when you do examine a part of a more complicated system as if
it were logically reducible to its parts, you will frequently obscure
some significant features and relations of the system and emphasize
others. Furthermore, the human use of reductionism is a creative and
imaginative process that may require new information to be inducted
into the logical system that is being considered.
My point is that knowledge must be learned and therefore from a
practical standpoint some aspects of knowledge must be emergent. And
just because a person can simplify some aspects of a complicated
subject matter it doesn't mean that all knowledge can be reduced to
individual components and every aspect of knowledge is not necessarily
reducible in the truest logical sense of the term. Furthermore, some
knowledge does not qualify as being in strong logical form and yet it
can be used effectively none the less. This suggests that strong
logical reductionism is not the same as simplification and that
understanding is not solely dependent on strong logical methods.
Other methods must be in use as well.
When information is inducted into a logical system (or a coherent
system) it may contain complex references that themselves
'contain' (or could be understood to be strongly correlated with)
effects and relations that are not initially considered. When the
references to the 'parts' of a complicated system are analyzed, a
human being may discover that he knows other things about the
reference that he did not initially consider. If his mental model of
the system is strongly logical (it need not be) then he would have to
induct the new information that he just recalled into the model to
incorporate its logical relations. This process could lead to an
emergent understanding that had been previously lacking. However, I
do not believe that the human being typically makes strong logical
models, I feel that he relies on coherent models which use a variety
of methods to create overlapping models about some subject matter.
This means that he may not be fully aware of some complicated aspects
or relations of the models that he creates to explain the 'world' (the
'world' of the subject matter being considered), and yet he may still
use these models to effectively anticipate and explain details about
it.
I feel that the concept of emergent capabilities in an AI programs is
valid. The idea that an AI program might begin to exhibit abilities
that it did not initially seem to possess seems like an obvious aspect
of a program that is capable of learning. But the argument about
strong emergence goes a little further here, in that it seems to
suggest that the program might actually develop some mysterious powers
that could not be reducible. That is not what I mean. The
reductionist critic of AI might argue that the AI program could not
acquire potential abilities that it did not always possess (i.e. at
least as potentials) and that any power that an AI program did acquire
could be logically reducible to their component parts. I do not
completely agree but my true feeling about this view is that it is
just misses the point. While a state of a running computer program
(the state of the running program as if it were suspended at some
point) can be analyzed into logical components, an AI program should
be designed to make references to complicated concepts (or other kinds
of data relations) that may not be so readily soluble to insightful
logical deductions. However, in spite of the fact that external
references may contain complicated hidden relations, an AI program
still might be able to analyze and make intelligent estimates about
the nature (of some) of these hidden relations based on partial
information it has about them.
But if the reductionist wants to make his stand on the claim that a
computer program could not possibly acquire a power if it did not
always possess the potential, then I will leave it to him. My
interest is in creating a better AI program in the near future and the
question for me is how do I do that. An attempt to better comprehend
of the nature of inductive knowledge and emergence seems like the
natural next step.
Jim Bromer
.
- Prev by Date: Re: The driving force of evolution
- Next by Date: Re: The driving force of evolution
- Previous by thread: Symbols
- Next by thread: Is evolution an intelligent process?
- Index(es):
Relevant Pages
|
