COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002		There are epistemological problems in a number of
C00006 ENDMK
C⊗;
	There are epistemological problems in a number of
areas.  In each case, the problem is to identify what kinds
of facts are likely to be available to a robot with the likely
opportunities to observe and what general facts are available - especially
what general facts are available that permit drawing conclusions.
In a few cases, we have some results.  The reader who is skeptical
of the value of considering epistemological problems separately
should ask himself if they can really best be solved in connection
with designing a program and whether success with the program
really co-incides with success in understanding the problem.

#. The relation between common sense reasoning and natural language.
One approach to common sense reasoning is to try to determine
what is considered a correct argument in natural language.  This
is based on the idea that what people know about the world and
use in reasoning is expressed in natural language.  Unfortunately,
this doesn't seem to be the case.  When we utter a declarative
sentence, we translate from our internal mental structure.  There
is no guarantee, and it seems unlikely, that the full content of
the mental structure is expressed in the sentence.  The hearer
of the sentence translates it into his own mental structure using
his existing beliefs to infer the meaning of the sentence.
There may be much in these internal mental structures that is
never expressed in language, and this unexpressed information
may be essential for drawing conclusions.  Therefore, it may be
that no adequate reasoning system can be devised that represents
all its information as natural language sentences of the kinds
that people actually utter.

	One approach to common sense reasoning would express the
facts in some natural language and draw conclusions according
to rules for correct reasoning in natural language.  I think
this would work to a limited extent, and it is certainly worthwhile
to seek rules of inference for natural language.  However, I
now believe that many facts used in human inference are never
expressed in natural language, and maybe they can't be so
expressed except as part of a scientific theory quite far
from the ordinary use of language.

	The reason is that natural language is used for communication
between intelligent beings with similar but not identical internal
mental structures.  When one person speaks, he translates his thought
from an internal form into the public language.  However, this translation
is never ccmplete.