COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	intell[f83,jmc]		Defining intelligence
C00007 ENDMK
C⊗;
intell[f83,jmc]		Defining intelligence

	Defining intelligence is not very fruitful from the point of
view of AI or cognitive science, because AI requires identification
of a whole collection of intellectual mechanisms and their separate
analysis.  Examples of these intellectual mechanisms include
tree search, pattern matching (which in turn involves the study of what
classes of patterns occur in the world and can be recognized), learning
from experience, communication, logical inference and non-monotonic
inference.  All present programs include only a few of the above
mechanisms, and there are undoubtedly others waiting to be discovered.

	With regard to human intelligence, several decades of efforts
aimed at identifying separate abilities were unsuccessful.  Almost
all of the variance could be explained by a single  g  factor.  I
think  g  stood for "general".  I don't know whether their efforts
to identify separate factors corresponded in any way to the intellectual
mechanisms that are studied in AI.  If not renewed efforts in that
direction might be worthwhile assuming there are psychologists not
intimidated by the propaganda against trying to measure human
intelligence.

	Nevertheless, we can try to say what intelligence is apart
from specific manifestations.  Whether the following will be regarded
as an attempt at a definition is for readers to decide.  Those who
regard it as an attempt then have to decide how successful it is.

	In the first place, our attempt will involve undefined
terms.  The relation between the performance of programs and
experiments with people and the undefined terms of the theory
is likely to be quite complex.  The fact that we have undefined
terms rather than operational definitions is just one consequence
of the complexity.

	The reason for the complexity is that we live in a complex
world and what is accessible to experiment is not necessarily
the most elementary phenomena.  Thus physics doesn't attempt an
operational definition of charmed quark.  There is no law of nature
to the effect that entities that evolve intelligence in a world and a desire
to understand the world will necessarily have direct access to
the fundamental components of that world.

	Therefore, our concept of intelligence will be mutually
recursive with a collection of other concepts and only the
constellation as a whole will have connections with behavior.
We will indicate this phenomenon by putting the names for
concepts being co-described in quotes.  Words that are not
quoted are to be taken in the sense of ordinary language.

	1. An intelligent entity is "good at" "understanding"
what it "observes" or "is told" about the situation in which
it has to act.  This "understanding" may be demonstrated by
successful action but isn't identical with it.  "Understanding"
involves making appropriate inference from what it "believes"
about its situation.

	2. An intelligent entity is "good at" deciding what to
in a situation assuming it "has" "goals".