ONT Re: Inquiry Driven Learning Environments
o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o
IDLE. Note 3
o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o
1.3. Current Approach to AI
Of course, I cannot list all of the components of human intelligence.
After we build Data, we can let him and Lore argue it out. But I can
indicate what I think are three fundamental modes by which intelligent
systems, humans included, are able to come to grips with the problems of
understanding that their environments demand of them. For the time being,
I believe that reasonable progress in developing these capacities can be
expected to result from the persistent application of current resources.
I think that developing intelligent systems which can fully integrate
this short list of generic components would be more than enough work
for several years, and would just about "satisfice" -- for now.
If we take a cue from the origins of the sundry words gathered
about "intelligence", it may appear that the Ancients already
understood the fundamental importance of what we Post*Moderns
call "representation", "inference", and "search" for the sake
of intelligent understanding, but more than likely all of that
business about "gathering among" and "selecting between" refers
to the process of "picking out" letters in reading. The lexical
roots collected here probably indicate an original analogy between
the game of hunting after meaningful and edifying lexemes in reading
and the work of gathering edible legumes and sorting out the grain from
the chaff and the tares ("gramma" = "letter", "writing", "a small weight",
"tare weight", "grain"). It is interesting that the Latin gleanings refer
mainly to reading or to passive inference while the Greek senses suggest
the kinds of added meanings that might be involved in synthetic efforts
to gather thoughts and to select ideas for arrangement in speeches and
lectures. Maybe this echoes the true historical circumstance that the
Romans acquired and compiled their earliest lessons in reasoning from
Hellenic sources and codes ("to compile" = "to plunder", "to pillage",
"to pile together in a heap"). But etymologies can be a tangled lot
of booty in their own rites.
Before discussing my list of three components, I need to
point out a single theme that will serve to unify most of
the goals that I presently hope to achieve in AI. One of
the things that we use our intelligence for is to help us
understand complex phenomena in the world -- this includes
both the natural world that we never made and the worlds of
our own nature (psychological, social, economic, political,
scientific, and technological) that we understand even less
for having made them. This is where I think that we find
a task that cries out for assistance. Specifically, we
need more intelligent software for dealing with complex
dynamic systems. Because intelligent systems are also
complex systems, there is a certain amount of recursive
self-application lurking in this domain and looping all
around it. Whether this helps or hurts I don't know yet.
Perhaps a little of both. But it seems certain that the
next few years will intensify a demand that we tackle two
of the most intertwined but still untouched quandaries of
intelligent software engineering: First, the qualitative
analysis of complex dynamic systems, second, the dynamic
modeling of complex qualitative systems (that is, where
the primary data of the realm either initially develop
or necessarily remain in qualitative form, constrained
by relations of logical or declarative type). It seems
clear that we will not be able to rise above the decimal
and binary dust of our previous ventures into these fields
without substantially augmenting our powers of qualitative
logical analysis, as aided by the evolution of intelligent
reasoning tools. Next in order of further ado, I need to
explain two features that define the scope of my present
conceptual framework and my current approach to pressing
problems in AI.
First, I see AI as a task of extending human capacities for
intelligent functioning, not merely a business of trying to
simulate the status quo. This attitude was manifest in the
"Intelligence Amplifier" rationale of several early workers
in AI and it often resorts to the analogy of the telescope
("intelliscope"). Of course, the reasoning goes, we need
to discover the essential principles of the faculty that
we wish to extend, but those principles must have their
implementations in materials and modalities well beyond
the original model, at least, if this extension is to be
anything but trivial. On the other hand, nobody working
in this vein is driven to put themselves completely out
of work as intelligent creatures. It is not a problem
to people of this persuasion if we leave our own human,
all too human selves unsimulated, but simply inhabiting
a hopefully more interesting and/or comfortable niche at
the core of all that we artifice. The aim is to serve as
the undischarged homunculi of our own recursive extensions.
For this reason, the components of intelligence that I take
as high-priority tasks for AI are not necessarily, at least,
not merely, those which are central to our characteristically
human way of understanding the world around us but those which
would reward us with the greatest non-redundant gains from being
extended by artificial means.
Second, it seems natural to use a topological metaphor to talk
about the issue of where such extensions are most advantageous.
The "boundary" is any interface or any perimeter of our human
intelligence where human limitations are felt most acutely,
and where growth would most naturally occur if cultivated.
The "interior" is composed of those regions of our human
functionality that are safely left to their own devices
for now -- their default operation is not so faulty that
it needs to be taken off automatic. Still, this language
needs to be guarded against confusion with another sense
of words like these, especially in regard to the common
distinction of "central" versus "peripheral" components
of computers and cognitive systems. The eyes amount to
a human peripheral that are constantly being extended,
from glass and radio telescopes to computer tomography
and virtual reality, but this is a sense of the evolving
periphery that is peripheral to my own sense of boundaries
in this discussion. I am, in contrast, concerned with the
limitations that are imposed on our learning and reasoning
agilities by the present confines of our mainly human frame.
Jon Awbrey
o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o