Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!europa.chnt.gtegsc.com!news.sprintlink.net!demon!uknet!newsfeed.ed.ac.uk!edcogsci!asch
From: asch@cogsci.ed.ac.uk (Andreas Schoter)
Subject: Re: why Ginsberg grouses
Message-ID: <DArysL.H0w@cogsci.ed.ac.uk>
Organization: Centre for Cognitive Science, Edinburgh, UK
References: <3s9gq5$6f4@pyrix.cs.uoregon.edu> <gat-210695114055@milo.jpl.nasa.gov>
Date: Mon, 26 Jun 1995 10:03:59 GMT
Lines: 102

gat@robotics.jpl.nasa.gov (Erann Gat) writes:

>In article <3s9gq5$6f4@pyrix.cs.uoregon.edu>, ginsberg@t.uoregon.EDU
>(Matthew L. Ginsberg) wrote:

>> Comments?

>Your philosophical points about the scientific process are right on, but:

>> 1. Reasoning can be reasonably captured using first-order logic and
>> some kind of bookkeeping mechanism.  My bookkeeping mechanism of
>> choice is called a "bilattice."

>Perhaps you'd better define "reasoning."  The claim that FOL is an adequate
>model of most human mental processes seems to me to be untenable.  Poeple
>regularly draw conclusions that are invalid under FOL (Example: "I do not
>understand how something as complex as a human could have evolved,
>therefore
>God exists.")  FOL is monotonic and consistent; human reasoning is not.

>I know this is an old and tired debate, but perhaps it would be worthwhile
>to revisit it in comp.ai.  It might be instructive to some readers.

>E.

Other people in this thread have mentioned the need for a paradigm
shift - I think, perhaps, this shift is already happening;
specifically the move from the primacy of `truth' as the foundational
notion in logic to the primacy of `information'.  Situation semantics
comes to mind as a detailed articulation of this, but it's also
present in Belnap's work and Veltman's.  To my mind the constructive
logics also, implicitly, embody this.  I've found bilattices also
allow one to embody this idea.

With respect to AI, bilattices do offer an interesting implementation
tool.  You should check out Fitting's short paper [2] where he applies
bilattices (specifically the minimal four-point bilattice) to logic
programming and comes up with an interesting system (also [3] is of
interest).  A variation on Fitting's system underlies part of my
implementation: I've used Ginsberg's [4] bilattices combined with
Belnap's computaional epistemic agent [1] to build a reasonable
efficient model-based reasoner.  The formal details of my system
(evidential bilattice logic) can be found in [5] which is ftpable from
ftp.cogsci.ed.ac.uk in pub/CCS-RPs/ccs-rp-64.ps.gz

Andreas
<asch@cogsci.ed.ac.uk>


[1] @InCollection(Belnap77,
	author =	"Belnap, N.D.",
	editor =	"J.M. Dunn and G. Epstein",
	title =		"A Useful Four-Valued Logic",
	booktitle =	"Modern Uses of Multiple-Valued Logic",
	volume =	2,
	series =	"Episteme",
	publisher =	"D. Reidel Publishing Company",
	pages =		"8--37",
	year =		"1977"
	)

[2] @Inproceedings{Fitting89:negation,
	author = 	"Fitting, M.C.",
	title = 	"Negation as refutation",
	booktitle = 	"Proceedings of the Fourth Annual Symposium on
			 Logic in Computer Science", 
	year = 		1989,
	editor = 	"Rohit Parikh",
	pages = 	"63-70",
	organization = "IEEE"
	}

[3] @Article{Fitting91:logprog,
        author =        "Fitting, M.C.",
        title =         "Bilattices and the Semantics of Logic
		  Programming",
        journal =       "Journal of Logic Programming",
        volume =        11,
        pages =         "91--116",
        year =          1991
        }

[4] @Article(Ginsberg88,
	author =	"Ginsberg, M.L.",
	title =		"Multivalued Logics: A Uniform Approach to
			 Reasoning in Artificial Intelligence",
	journal =	"Computational Intelligence",
	volume =	4,
	pages =		"265--316",
	year = 		1988
	)

[5] @TechReport(Schoter:EBL,
	author =	"{Sch\"oter}, A.",
	title =		"Evidential Bilattice Logic and Lexical
			 Inference", 
	institution =	"University of Edinburgh, Centre for Cognitive
			 Science",
	type = 		"Research Paper",
	number = 	"EUCCS/RP-64",
	year =		1994
	)
