RE: [Fis] Consilience & patchiness of knowledge

Malcolm wrote:
> Certainly, all communication requires some kind of common
> ground between communicators (which in Dretske, 1981, called
> channel conditions in Knowledge and the Flow of Information).
> It's not possible to question all our background
> assumptions, or even to articulate them all explicitly. But,
> what happens when some of our background assumptions are wrong?

There is an idea in computational learning theory called 'agnostic learning'
(Haussler, 1992). Even if you know that all your theories are wrong, you can
still find one that works best. For example, there is probably no invisible
hand making sure that the relationship between education and economic growth
is linear, but we can nevertheless find which linear function best describes
the relationship. Instead of placing trust into the language we use to
express theories, we place the trust into the *utility* of the theories. By
"language" I mean the common ground, a part of the paradigm: the formal
space of all possible theories; a point in this space is a theory. Such a
perspective needs fewer break-downs as the language continues to develop:
for every problem there are two possibilities: find the best approximation
or develop a new language.

There is a particular division in every community: there are the people that
try to do better with the existing language, and there are rebel boat
rocking people who are proposing new languages. For example, the information
science community is proposing a new language with 'information' inside it,
for example.

> In normal science, Kuhn pointed out that scientists tend not
> to question the background theory so long as it keeps solving
> puzzles. But what happens when normal science ceases to
> solves puzzles? What happens when science is in crisis?
> According to Kuhn this leads to a breakdown in communication.

True. At that point the community splays apart in search of the new
language. But due to the nature of this, the communication is difficult due
to the abundance of mutually incomprehensible dialects. Eventually, one of
them will be better then others (or at least better funded), and others will
abandon and join in. Alternatively, there will be a chain reaction of
consiliences between the dialects with the hope of a new standard paradigm
emerging.

> Following this line of thought, the discovery of new
> consiliences (and the preservation of old ones) is the key to
> developing new common ground that leads to more powerful and
> accurate decision making. More powerful and accurate
> communication does not refer to the increasing the number of
> bits of information (the channel capacity). This by itself
> cannot lead to more successful decision making. It has to do
> with the quality of information,
> not its quantity. That's why something along the lines of Dretske's
> semantic notion of information married with the concept of
> consilience might be the right way to go.

I always view "information" as conditional to some paradigm. Namely, the
information of a certain phenomenon depends on whether I view the process as
linear or as non-linear. The linear view implies a simpler model with less
information (more noise, more Shannon entropy), and the non-linear view
implies a complex model with more information (less noise, less Shannon
entropy). There is an interesting trade-off here, if one considers the
information of the model: a complex model requires more information to
specify than a simple model. The idea of Rissanen's minimum description
length is to combine these two pieces of information: the information needed
to specify the model and the information needed to explain the data with the
model.

Some people think that some particular paradigm is intrinsically true. In a
true paradigm, the information would correspond to physical information. I
cannot deny that they are right, but I can see various kinds of problems
(breakdowns, revolutions) that can arise by putting too much faith into a
particular paradigm. On the other hand, a more relaxed perspective of truth
has the negative side effect of a Babel-like lack of common ground, and the
ensuing problems in achieving synergy and collaboration.

Aleks

--
mag. Aleks Jakulin
http://www.ailab.si/aleks/
Artificial Intelligence Laboratory, 
Faculty of Computer and Information Science, University of Ljubljana. 
_______________________________________________
fis mailing list
fis@listas.unizar.es
http://webmail.unizar.es/mailman/listinfo/fis