SUO: Inquiry Driven Systems
¤~~~~~~~~~¤~~~~~~~~~¤~~~~~~~~~¤~~~~~~~~~¤~~~~~~~~~¤
SUO Work Group,
While rummaging around in my files for stuff that
I felt sure I had written before on modularity
versus primality among sign relations, I came
across this early essay in the direction of
what eventually became my dissertation, on
the subject of "Inquiry Driven Systems".
Because this is a little long, I have
broken it up into bite-sized portions.
¤~~~~~~~~~¤~~~~~~~~~¤~ARCHIVE~SOURCES~¤~~~~~~~~~¤~~~~~~~~~¤
[ Document History:
|
| Subject: Inquiry Driven Systems
| Contact: Jon Awbrey <jawbrey@oakland.edu>
| Version: Draft 11.1
| Created: 1996 Aug 04
| Revised: 1997 Feb 11
| Revised: 2001 Mar 11
| Setting: Oakland University
]
Inquiry Driven Systems
"Inquiry" is a word in common use for a process that resolves doubt
and creates knowledge. Computers are involved in inquiry today, and
are likely to become more so as time goes on. The aim of my research
is to improve the service that computers bring to inquiry. I plan to
approach this task by analyzing the nature of inquiry processes, with
an eye to those elements that can be given a computational basis.
I am interested in the kinds of inquiries which human beings carry on
in all the varieties of learning and reasoning from everyday life to
scientific practice. I would like to design software that people could
use to carry their inquiries further, higher, faster. Needless to say,
this could be an important component of all intelligent software systems
in the future. In any application where a knowledge base is maintained,
it will become more and more important to examine the processes that
deliver the putative knowledge.
Three questions immediately arise in the connection between
inquiry and computation. As they reflect on the concept of
inquiry, they have to do with its integrity, effectiveness, and
complexity. These questions ask whether all such processes dubbed
"inquiry" have anything essential in common, whether any useful parts
of them can be automated in practice, and just how deep is the takedown
needed to reach the level of routine steps. The issues of effectiveness
and complexity will be discussed throughout the remainder of this text,
but the problem of integrity must be dealt with immediately, since doubts
about it may interfere with my ability to exercise this title to "inquiry".
Thus, we must examine the integrity, or well-definedness, of the very idea of
inquiry, that is, "inquiry" as a general concept rather than a catch-all word.
Is the faculty of inquiry a principled capacity, leading to a disciplined form
of conduct, or is it only a disjointed collection of unrelated skills? As it
is currently being carried out on computers today, inquiry includes everything
from database searches, through dynamic simulation and statistical reasoning,
to mathematical theorem proving. Insofar as these tasks constitute specialized
efforts, each of them demands software that is tailored to its individual purpose.
Insofar as these different modes of investigation contribute to larger inquiries,
our present methods for coordinating their separate findings are mostly ad hoc
and still a matter of human skill. Thus, we might question whether the very
name "inquiry" succeeds in referring to a coherent and independent process.
Do all the varieties of inquiry have something in common, a structure or
a function that defines the essence of inquiry itself? I will say "yes".
One advantage of this answer is that it brings the topic of inquiry within
human scope, and also within my capacity to research. Without this, the
field of inquiry would be impossible for any one human being to survey,
because a person would have to cover the union of all the areas that
employ inquiry. By grasping what is shared by all inquiries, I can
focus on the intersection of their generating principles. Another
benefit of opting for this answer is that it promises a common
medium for inquiry, one in which the many disparate pieces of
our puzzling nature may be bound together in a unified whole.
When I look at other examples of instruments that people have used
to extend their capacities, I see that two questions must be faced.
First, what are the principles that enable human performance? Second,
what are the principles that can be augmented by available technology?
I will refer to these two issues as the question of original principles
and the question of technical extensions, respectively. Following this
model leads me to examine the human capacity for inquiry, asking which
of its principles can be reflected in the computational medium, and which
of its faculties can be sharpened in the process. It is not likely that
everybody with the same interests and applications would answer these
questions the same way, but I will describe how I approach them, what
has resulted so far, and what directions I plan to explore next.
The focus of my work will narrow in three steps. First, I intend
to concentrate on the design of intelligent software systems that
support inquiry. Then, I will select mathematical systems theory
as an indispensable tool, both for the analysis of inquiry itself
and for the design of programs to support it. Finally, I plan to
develop a theory of qualitative differential equations, implement
methods for their computation and their solution, then apply the
resulting body of techniques to two kinds of recalcitrant problems:
(1) those where an inquiry must begin with too little information
to justify quantitative methods, and (2) those where a complete
logical analysis is necessary to identify critical assumptions.
The stages of work just described will gradually lead me to introduce the
concept of an "inquiry driven system". In rough terms, this type of system
is designed to integrate the functions of data-driven adaptive systems and
rule-driven intelligent systems. The idea is to have a system whose adaptive
transformations are determined, not by learning from observations alone nor by
reasoning from concepts alone, but by the interactions between these two sources
of knowledge. A system that combines different contributions to its knowledge base,
much less the mixed modes of empirical and rational types of knowledge, will find
that its next problem lies in reconciling the mismatches between these sources.
Thus, we arrive at the concept of an adaptive knowledge-base whose changes over
time are driven by the differences that it encounters between what is observed
in data and what is predicted by laws. This sounds, at the proper theoretical
distance, like an echo of the error-controlled cybernetic system, moreover, it
falls into line with classic descriptions of scientific inquiry. Finally, this
suggests that good formulations of such "differences of opinion" might allow us
to find differential laws for the temporal evolution of inquiry processes.
There are several implications of my approach that I need to emphasize.
Many distractions can be avoided if we guide our approach by the two questions
raised above, of principles and extensions, and if we guard against confounding
what they ask with what they do not ask. The issues that surround these points,
concerning the actual nature and the possible nurture of the capacity for inquiry,
will be taken up shortly. But first I need to deal with a preliminary source of
confusion. This arises from the two vocabularies, the language of the application
domain, which talks about higher order functions and intentions of software users,
and the language of the resource domain, which describes the primitive computational
elements to which software designers must try to reduce the problem. We are forced
to use, or at least to mention, both of these terminologies in our effort to bridge
the gap between them, but each of these languages plays a different role in the work.
In studies of formal specifications the designations "reduced language"
and "reducing language" are often used to discuss the two roles that are
encountered here, that of the "application", "practice", or "target" domain,
on the one hand, and that of the "base", "method", or "(re)source" domain,
on the other. I will be using all of these terms, with the following two
qualifications.
First, I must note a trivial caution. Our sense of "source" and "target"
will often get switched depending on our direction of work. Furthermore,
these words are reserved in category theory to refer to the domain and
the codomain of an "arrow", that is, a function, a mapping, a morphism,
or a transformation. This will limit their use in the above sense to
the more informal contexts.
Now, I must deal with a more substantive issue. In attempting to automate
a fraction of such grand capacities as intelligence and inquiry, it is seldom
that we totally succeed in reducing one domain to the other. The reduction
attempt will usually result in our saying something like this: that we have
reduced the capacity A in the application domain to the sum of the capacity B
in our base domain plus some residue C of unanalyzed abilities that must be
called in from outside the basic set. The residual abilities will then be
assigned to the human side of the interface, that is, attributed to the
conscious observation, common sense, or creative ingenuity of users and
programmers. In the theory of recursive functions, we would say that A
is "relatively computable", given an "oracle" for C. For this reason,
I will often speak of "relating" a task to a method, rather than fully
"reducing" it. A measure of initial success is often achieved when we
can relate or connect an application task to a basic method, long before
we can completely reduce one set of them to the other. The catch will
always be whether the basic set of resources has already been implemented,
or is just being promised, and whether the residual ability has a lower
complexity than the original task, or is actually more difficult.
¤~~~~~~~~~¤~~~~~~~~~¤~SECRUOS~EVIHCRA~¤~~~~~~~~~¤~~~~~~~~~¤
More Tomorrow,
Jon Awbrey
¤~~~~~~~~~¤~~~~~~~~~¤~~~~~~~~~¤~~~~~~~~~¤~~~~~~~~~¤