COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "let.pub[let,jmc]" source
C00007 ENDMK
C⊗;
.require "let.pub[let,jmc]" source;
∂MEM Anne Gardner$$Your %2Notes Toward a Legal Reasoning Program%1$∞

	I like your %2Notes Toward a Legal Reasoning Program%1.
A program that could do the example on the first page in a reasonable
way would represent a considerable advance in artificial
intelligence.

	I agree with your solution to the problem as a decision tree
branching on the "significant" interpretations of the law and the
facts.

	I think you will have to use circumscription or some other
method of non-monotonic reasoning in a central way in the program.
The essence of legal rules is that they have exceptions that can't
be incorporated in the text.

	Perhaps you should do an axiomatic formalization before you attempt
a program.  It should be easier and will clarify the reasoning.  You
will have to decide on the most convenient way of handling non-monotonicity.
Apropos of axiomatization, you should read Bob Moore's forthcoming
M.I.T. Phd thesis.  I can show you a copy, and perhaps he will make
you one, though it's rather long.

	The biggest problem will come in putting in sufficient real world
knowledge.  I am not sure that Moore's axioms for knowledge and action are
usable, but they are certainly a step in the direction you need.

p3 - I don't agree with your reasons for rejecting MYCIN's probabilistic
reasoning.  While the judge cannot be frank if he has acted in the
way you describe, he may actually do so.  Incidentally, this raises
a big problem.  Suppose that the reasoning a lawyer or judge can put on paper
is governed by certain conventions, and does not parallel the reasoning
that goes on in his head.  Would you follow one of them or would your
program have both an internal reasoning process and what it "put on paper"?
The same comment applies to the separation of diagnosis and remedy.
Of course, there may be other reasons for rejecting the MYCIN model,
and my own opinion, and even hope, is that legal reasoning will require
new methods.

p3 - I agree with the limitations you have chosen.

p4 - I agree that you should choose a subdomain of law in which there
isn't major doctrinal controversy, and I suppose this contracts problem
is in such a subdomain.

p6 - It would very good if the program would state its assumptions, e.g.
the one about the Uniform Commercial Code.
.next page
p10 - Formulation or use of rules concerning burden of proof will involve
non-monotonic reasoning.  It seems to me that your proposal to use them
to solve the missionaries and cannibal problem merely displaces the
problem.

p11 - Maybe the program need not say "probably  the July 1 offer expired"
but rather "if the July 1 offer is considered to have expired, then ...".

p12 - The decision tree seems to be the right idea.