COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	histor[w86,jmc]		Mental histories
C00005 ENDMK
C⊗;
histor[w86,jmc]		Mental histories

Deductive reasoning has a certain ahistorical character not possessed
by reasoning involving any form of nonmonotonic inference.
If we take the set of beliefs to be closed under deduction, it
can be represented by any set of sentences from which the
complete set of beliefs can be deduced.  On the other hand a
system that does nonmonotonic inference must keep track of the
history of its inferences, and subsequent nonmonotonic inferences
actually depend on this history and not merely on the set of
sentences currently believed.  This first became apparent in
connection with Vladimir Lifschitz's discovery of the defects
in my simple abnormality theory for the blocks world.  The
trouble could be fixed by suitably giving priority to minimizing
the abnormality of the initial situation and earlier situations
in general, but that would give the wrong result if the system
were inferring facts about the past from facts about the present.
It seems the priority has to be given to minimizing the abnormality
of the facts known at the beginning of the reasoning - or so it
currently seems.

Mental reorganization

	The afore-mentioned priority to the first known facts is
sometimes changed in mental reorganizations associated with
discoveries.  One comes to certain tentative beliefs by long
chains of hypothetical nonmonotonic reasoning, and suddenly
these hypotheses become more solid than the reasoning that led
to them, i.e. the mental space is reorganized and some of the
new facts become the axiomatic basis of one's understanding
of the phenomenon in question.  Machines will have to do this
eventually, but I fear not soon.