INHIBITION: ANOTHER METHOD OF NON-MONOTONIC REASONING


	Circumscription may be suitable for computer reasoning
and may help understand non-monotonic reasoning in general, but
it seems somewhat implausible as the mechanism used in human
non-monotonic reasoning.  Circumscription
schemas with their predicate and function variables are too
hard for non-logicians to understand to have much psychological
plausibility.  For artificial intelligence, psychological
plausibility isn't everything, so we'll continue to develop
circumscription,
but here is a quite different mode of non-monotonic reasoning;
at least it seems quite different.

	We will call it  inhibition.  Consider the bird example.
Using circumscription we write

	∀x.bird x ∧ ¬prevfly x ⊃ canfly x

and

	∀x.ostrich x ⊃ prevfly x

and then circumscribe  prevfly  in whatever facts we
choose to take into account.

	In the inhibition formalism we write

	∀x.law1: bird x ⊃ canfly x

and

	∀x.ostrich x ⊃ inhibit law1 x

This formalism is frankly metalogical.
We remark that it isn't certain whether we should write
the formula as above or should write

	∀x.law1(x): bird x ⊃ canfly x

or perhaps

	law1: ∀x.bird x ⊃ canfly x.

The latter has the advantage that the formula labelled
is entirely in first order logic.  However, in order to
do what is required we need to be able to write

	inhibit law1(x)

and not merely

	inhibit law1.

	Yet another possibility is

∀x.[law1(x):bird x ⊃ canfly x] ∧ [ostrich x ⊃ inhibit law1(x)]

John McCarthy, Stanford University