UNIFIED MODEL FOR KNOWLEDGE REPRESENTATION? (IMPOSSIBLE

2 views
Skip to first unread message

Stephen Smoliar

unread,
Jun 3, 1991, 7:59:46 PM6/3/91
to
In article <52...@syma.sussex.ac.uk> aar...@syma.sussex.ac.uk (Aaron Sloman)
writes:
>
>Some of the people who have worked on first order logic have thought
>it could serve as the universal (= unified?) notation for
>representing knowledge. I don't know if this is what you mean by a
>model.
>
>However, it seems pretty obvious from the history of science and
>culture that different formalisms are useful for different purposes,
>including, logic, algebra, pictures, maps, tables, flow-charts,
>musical notation, 3-D models, natural languages, etc.
>
>One reason for this is that different kinds of notations have
>different kinds of variability, which limits their expressive power
>in different ways. This can sometimes be important when exploring
>large search spaces. If the structure of the notation is such that
>it won't let you express things that would only be rejected anyway,
>it can have great heuristic power.
>
>If anyone claims to have a unified model, ask him/her if it will
>serve for the purposes of representing knowledge about the contents
>of the current optic array in a robot's visual system, and for
>transforming that knowledge in the process of discovering what's out
>there (e.g. discovering binocular disparities for stereo vision) or
>for fine control of posture and actions, etc.
>
In article <1991Jun3.1...@msuinfo.cl.msu.edu>
stic...@pleiades.cps.msu.edu (Jon Sticklen) then runs
with this ball, writing:
>
>The point of what in knowledge-based systems is being called "task specific
>architectures" (TSAs) is to suggest that problem solving is best analyzed and
>"mimiced/implemented" by using primitives that are used in the domain of the
>problem solving. For example, in a diagnostic domain, it is natural to both
>analyze diagnostic problem solving, and to build computer versions of
>diagnostic problem solving, by using the concept of "diagnostic hypothesis".
>Specializations of the TSA concept like Chandrasekaran's generic tasks (GTs)
>take one additional step of saying that there exists a finite set of problem
>solving strategies that are generally useful. These GTs are defined by giving
>both a knowledge representation template, and a control strategy. For example,
>there is a GT for classification problem solving, one for simple ("routine")
>design, one for function-based model level reasoning...
>
>At the same time as researchers seek to extend the capability of TSAs like
>those in the generic tasks, grand architectures for problem solving are being
>developed to handle any problem solving situation - SOAR may be the easiest
>example. SOAR generates problem spaces appropriate for a given problem.
>Although not a developed capability yet, the SOAR architecture may one day be
>able to generate very tailored problem spaces after analyzing a given problem
>situation - perhaps problem spaces not unlike GTs.
>
>If that happens, then it will be a unification. But a unification along the
>lines that Arron pointed to, not a unification at the base level of saying
>"one
>representation fits all."
>
I think this approach may be missing the point Aaron was trying to make.
Generic tasks and problem spaces are, once the dust of surface appearance
is swept aside, just as much mathematical objects as are the constructs of
first-order logic. The difficulty lies in attempting the act of reductionism
itself, rather than in the particular construct which is the target of your
reduction.

Aaron offered up a nice list of different formalisms which have been engaged
for different purposes of reasoning. However, the more we begin to recognize
the importance of viewing reasoning as SITUATED, the more difficult it becomes
to carve off "the reasoning itself" as an object of study. This opens the door
to the intimidating prospect that we need to be concerned with more general
issues of BEHAVIOR, which means that the formalisms in Aaron's list are only
scratching the surface.

Consider this problem from another point of view. Regardless of whether or not
we think it is feasible to build machines which satisfy "behavioral criteria
for intelligence" (whatever those criteria may be), let us simply worry about
whether or not we can DESCRIBE intelligent behavior. Let me say right off that
I shall be the first to raise a skeptical eyebrow at any claim that this
problem has been solved in any practical way. The fact is that psychology
is still, to a great extent, floundering around simply trying to DESCRIBE
many of the phenomena it wishes, ultimately, to explain. For many limited
domains of scientific reasoning, we can at least fall back on foundations
of formalisms such as those Aaron has enumerated; but when we try to take
on behavior (even when intelligence isn't directly involved), we might as
well be in a raft in the middle of the North Atlantic Ocean.

I would like to pose a possibly radical explanation for why we find ourselves
so lost here, and that is that THERE IS NO SUCH THING AS OBJECTIVE DESCRIPTION.
Description is highly subjective to the person doing the describing. If anyone
else wishes to draw upon that description, he is obliged to enter into a
relatively sophisticated process of negotiation which is known as communicating
in a natural language. (Anyone who is interested in seeing blatant examples of
how subjectivity lurks in seemingly objective accounts should take a look at
"Apes R Not Us," a review of primate studies by Lord Zuckerman which appeared
in the May 30 issue of THE NEW YORK REVIEW. Also relevant is Brian Smith's
response to Lenat and Feigenbaum, "The Owl and the Electric Encyclopedia,"
published in Volume 47 of ARTIFICIAL INTELLIGENCE.)

In the face of such a daunting proposition, it may be that the quest for a
"unified representation" is not so much misguided because of the "unified"
attribute as because of the very goal of representation itself. We are now
entering a period of skepticism regarding the issue of representation. Such
skepticism is encouraged by the work of Smith, Rodney Brooks, and others far
too numerous to mention. If any of these researchers can demonstrate results
which scale up from small experiments to practical problems of getting on in
the world, our current obsession with representation may ultimately be
dismissed as a distracting side-track.

===============================================================================

Stephen W. Smoliar
Institute of Systems Science
National University of Singapore
Heng Mui Keng Terrace, Kent Ridge
SINGAPORE 0511

BITNET: ISSSSM@NUSVM

Enrico Coiera

unread,
Jun 4, 1991, 4:26:25 AM6/4/91
to

Steven Smoliar's follow-up contains an interesting suggestion ie that
'there is no such thing as subjective description. Description is highly
subjective to the person doing the subscribing'. A similar view is
shared by Paul Compton (see
P. Compton, R. Jansen, A Philosophical basis for Knowlege Aquisition,
Knowledge Aqusition, (1990), 2, 241-257). He suggests that there is
little point in attempting to structure the knowledge elicited from
experts because each expert 'makes-up' the knowledge as it were in the
context of a particular problem. While each individual has an internal
representation of some form, the way in which it is communicated to
others is highly context dependent - the knowledge elicited in context
only has validity in that context - it makes no sense to attempt to
represent it in a way in which it can be considered context independent.


Enrico Coiera

Hewlett Packard Labs
Filton Rd
Stoke Gifford, Bristol BS12 6QZ
United Kingdom

John Josephson

unread,
Jun 5, 1991, 12:26:16 PM6/5/91
to

Steven Smoliar> THERE IS NO SUCH THING AS OBJECTIVE DESCRIPTION

Careful lest next you decide that no description is any better than
any other, since they are all subjective, in which case (paraphrasing
Bob Dylan) there is no point talking to you, it's just the same as
talking to anybody.

If, indeed, some descritions are better than others, how can this be?
Maybe some are more objective than others.

.. jj

Stephen Smoliar

unread,
Jun 6, 1991, 12:46:08 AM6/6/91
to
In article <JJ.91Ju...@medulla.cis.ohio-state.edu>

j...@medulla.cis.ohio-state.edu (John Josephson) writes:
>
>Steven Smoliar> THERE IS NO SUCH THING AS OBJECTIVE DESCRIPTION
>
>Careful lest next you decide that no description is any better than
>any other, since they are all subjective, in which case (paraphrasing
>Bob Dylan) there is no point talking to you, it's just the same as
>talking to anybody.
>
The trouble with emphasizing an entire sentence is that it distracts the casual
reader from the subsequent elaboration of that sentence. I fear I had better
repeat the following two sentences, since I feel they are more to the point
than John's attempt at a witty cautionary remark:

>Description is highly subjective to the person doing the describing. If
>anyone
>else wishes to draw upon that description, he is obliged to enter into a
>relatively sophisticated process of negotiation which is known as
>communicating
>in a natural language.

In other words if you want to talk with me (or anyone else) you first have to
WANT to engage yourself in this "process of negotiation." If you approach
dialog without such a disposition to negotiation, then John is quite right.
In that frame of mind, there is no point in your talking to anyone; the
experience will be tantamount to talking to yourself.

>If, indeed, some descritions are better than others, how can this be?
>Maybe some are more objective than others.
>

In any given situation of discourse, we, as outside observers, may be able to
say that certain exchanges involve LESS negotiation than others. However, I
would be reluctant to say that those exchanges are based on "better"
descriptions. The amount of negotiation which is required is more dependent
on the context of the discourse than on any characteristic attributes of the
descriptions being invoked. Take a look at the cited Brian Smith article
before taking your next shot from the hip, John. :-)

===============================================================================

Stephen W. Smoliar
Institute of Systems Science
National University of Singapore
Heng Mui Keng Terrace, Kent Ridge
SINGAPORE 0511

BITNET: ISSSSM@NUSVM

"The funny thing about being smug about health and purity all the time is that
it can turn you into a fascist."--Paul Theroux

Hutchison C S

unread,
Jun 10, 1991, 5:47:54 AM6/10/91
to
re: Steven Smoliar> THERE IS NO SUCH THING AS OBJECTIVE DESCRIPTION

I missed earlier messages, so maybe I'm barking up the wrong tree here. I'm
starting work in the area of descriptions and truth. Basically, the problem
is this: given four descriptions of the same event (in this case, events
which took place on 2nd June 1975 on the outskirts of what was then Salisbury,
Southern Rhodesia) --

"Rioting blacks shot dead as ANC leaders meet" (Times of London)
"Police shoot 11 dead in Salisbury riot" (Manchester Guardian)
"Rebels kill 11 ANC men" (Times of Zambia)
"Racists murder Zimbabweans" (Tanzanian Daily News)

each report is either true (corresponds to a state of affairs in the world) or
false (does not correspond). By virtue of the meanings of the words in the
sentences, they cannot all be true at the same time. Yet they all report the
'same' events. My hunch is to say that the physical circumstances under-
determine possible linguistic descriptions, enabling various ideological
interpretations of the events. What then do readers 'know' about the events?
How is knowledge in this sense distinct from mere belief?

Any thoughts?

cameron shelley

unread,
Jun 10, 1991, 1:51:10 PM6/10/91
to
In article <1991Jun10....@kingston.ac.uk> is_...@kingston.ac.uk (Hutchison C S) writes:
[...]

>My hunch is to say that the physical circumstances under-
>determine possible linguistic descriptions, enabling various ideological
>interpretations of the events. What then do readers 'know' about the events?
>How is knowledge in this sense distinct from mere belief?
>
>Any thoughts?

Eduard Hovy produced a text generation system "PAULINE" that generated
different accounts of the same event (a student action at Yale) from
different points of view. You should consider reading about that system.

The lesson to be learned from that system, is that there many ways of
phrasing a single proposition, the choice of expression is then determined
by what stylistic and political (read 'social') content you wish to
place in the description. The situation itself (the material state-of-
affairs?) does not contain such information, but the people concerned
with it do have such views.

I guess that if you want to try and filter out how people's views affect
their descriptions, then you'll have to come up with a model that
systematically relates the two. That's not so much a matter of 'truth'
as socio-linguistics.

Cam

Stephen Smoliar

unread,
Jun 10, 1991, 8:16:36 PM6/10/91
to
In article <1991Jun10.1...@watdragon.waterloo.edu>
PAULINE was basically an exercise in rhetoric, demonstrating that an allegedly
objective propositional account could be realized in a wide variety of texts,
where that variety could be delimited according to some underlying rhetorical
structures. I find it an excellent demonstration of the scope of rhetoric,
but I question the premise that one can start with that objective propositional
account. This may sound solipsistic, but the only accounts we can give of
events of the world are based on perceptions, be they our own, those of
"credible sources," or interpretations of devices. (Note that key word
"interpretations." Sophisticated devices do not PERCEIVE the world. WE
perceive the world through our ability to use those devices. For example,
cloud chambers do not "perceive" subatomic particles but simply provide us
with evidence of their existence which we are then obliged to interpret.)

I think it is the lack of such an objective starting point that supports Cam's
position that the issue here is not "truth" (so, yes, Hutchison IS "barking up
the wrong tree") but sociology--specifically the role the inter-personal
behavior contributes to communication. We cannot expect to run the rhetorical
mechanisms of PAULINE "in reverse" in order to "distill" an objective account
out of a given presentation. The best we can hope for is to strip out the
rhetorical "interference" and get a more compact statement of what that
particular individual perceived; but those perceptions are only "meaningful"
to the extent that we know how that individual interpreted his sensations.
Because this is asking a bit much for even HUMAN intelligence, I have been
arguing my position based on NEGOTIATION: The only way I can figure out what
you are talking about is to engage in dialog. In the course of that dialog,
I develop hypotheses about your perceptions which I can then test by "probing"
you with appropriate questions and remarks. This process never really
"converges" to my having a "total" model of your perceptual interpretations;
but it tends to provide enough information for the two of us to share our
experiences of a common world.

Hutchinson's example of four separate newspaper accounts makes negotiation a
bit more tricky, since you cannot engage in a dialog with a newspaper.
Fortunately, there are some viable alternatives. Newspapers tend to have
editorial policies. Frequent exposure to the TIMES, supplemented by knowledge
of their editorial pieces, can also be used as a source of hypotheses about
their perceptions of world events. Such hypotheses can often be tested by
the very basis of Hutchinson's experiment--comparing their account of a story
with those of other papers. This is the sort of reading which is required in
order to extract the news from a newspaper (as anyone who has ever been the
subject of a newspaper articles knows); and I would argue that it is the
"logical equivalent" (using those words VERY informally) of the way we use
dialog in normal discourse.

===============================================================================

Stephen W. Smoliar
Institute of Systems Science
National University of Singapore
Heng Mui Keng Terrace, Kent Ridge
SINGAPORE 0511

BITNET: ISSSSM@NUSVM

"He was of Lord Essex's opinion, 'rather to go an hundred miles to speak with
one wise man, than five miles to see a fair town.'"--Boswell on Johnson

Tom Bylander

unread,
Jun 11, 1991, 11:06:45 AM6/11/91
to
In article <910611002...@lilac.berkeley.edu> ISS...@NUSVM.BITNET (Stephen Smoliar) writes:
>...I question the premise that one can start with that objective propositional

>account. This may sound solipsistic, but the only accounts we can give of
>events of the world are based on perceptions, be they our own, those of
>"credible sources," or interpretations of devices....

>
>I think it is the lack of such an objective starting point that supports Cam's
>position that the issue here is not "truth" (so, yes, Hutchison IS "barking up
>the wrong tree") but sociology--specifically the role the inter-personal
>behavior contributes to communication....

I agree that "perceptions" are not completely objective, but if they
did not deliver some element of truth, then a lot of things become
hard to explain. How do humans and other animals with sophisticated
sensors survive if their perceptions do not provide appropriate
information about the world, i.e., something true about the world?
Are scientific arguments about cold fusion, ozone depletion, dinosaur
extinction, cholesterol, and so on merely sociological?

I think a more reasonable position is that perceptions provide
evidence about many aspects of the world, and that the quality of our
assertions about the world depends on the quality of the evidence that
supports them.

Tom Bylander
byl...@cis.ohio-state.edu

cameron shelley

unread,
Jun 11, 1991, 5:52:24 PM6/11/91
to
In article <133...@tut.cis.ohio-state.edu> byl...@iris.cis.ohio-state.edu (Tom Bylander) writes:
[...]

>I agree that "perceptions" are not completely objective, but if they
>did not deliver some element of truth, then a lot of things become
>hard to explain. How do humans and other animals with sophisticated
>sensors survive if their perceptions do not provide appropriate
>information about the world, i.e., something true about the world?
>Are scientific arguments about cold fusion, ozone depletion, dinosaur
>extinction, cholesterol, and so on merely sociological?

Actually, I think the question (to which I responded before) has not
so much to do with efficacy of perceptual capabilities so much as our
compulsion to interpret what we perceive. This is further compounded
by our subsequent encoding of information for communication to others.

I believe it is reasonable to assume that our perceptual capabilities
are evolved to help achieve behavioural success, and when dealing with
'everyday' physical data we have little to gain from adopting particular
political viewpoints. On the other hand, the original poster seemed
more interested in events which are politically charged (ie. perceived
in a social context), and filtering "truth" from ideology is unlikely
to get far. This is especially true of reports of the motives of other
people and the social effects of their actions.

>I think a more reasonable position is that perceptions provide
>evidence about many aspects of the world, and that the quality of our
>assertions about the world depends on the quality of the evidence that
>supports them.

Again, the subject really wasn't the quality of our perceptions, but
that our method of communication contians biases which affect the content
of messages. I would point out also that most scientific arguments
deal with what generally happens (distributions), not about a narrowly
"truthful" account of any individual event. At that point, communication
systems like mathematics (with biases we've all been trained to accept)
are used, rather than natural language monologs.

Cam

Stephen Smoliar

unread,
Jun 11, 1991, 10:49:23 PM6/11/91
to
>In article <910611002...@lilac.berkeley.edu> ISS...@NUSVM.BITNET
>(Stephen Smoliar) writes:
>>...I question the premise that one can start with that objective
>>propositional
>>account. This may sound solipsistic, but the only accounts we can give of
>>events of the world are based on perceptions, be they our own, those of
>>"credible sources," or interpretations of devices....
>>
>>I think it is the lack of such an objective starting point that supports
>>Cam's
>>position that the issue here is not "truth" (so, yes, Hutchison IS "barking
>>up
>>the wrong tree") but sociology--specifically the role the inter-personal
>>behavior contributes to communication....
>
>I agree that "perceptions" are not completely objective, but if they
>did not deliver some element of truth, then a lot of things become
>hard to explain.

Is a proposition having "some element of truth" sort of like a woman being "a
little bit pregnant?" :-)
I would argue that if you want to talk about propositional accounts at all,
then you are basically buying into the rules of formal logic; and those rules
require that a proposition be either true or false. (Even if you choose to use
a fuzzy logic system, every USE of a proposition ultimately boils down to a
commitment to the truth or falsity of that proposition. Making the logic fuzzy
simply allows the proposition to flip back and forth with a bit more
flexibility.) I am further arguing that getting on in the world does
not require buying into those rules. (To repeat Minsky's position from
THE SOCIETY OF MIND: Logic is all right for a POST HOC systematic explanation,
but that does not mean it is any good for controlling your decision-making
behavior.)

> How do humans and other animals with sophisticated
>sensors survive if their perceptions do not provide appropriate
>information about the world, i.e., something true about the world?

Your problem, Tom, is that you want to equate the adjectives "appropriate" and
"true." I would argue that "true" is only a useful term in formal logic.
Outside of that realm, it has been abused to death since (at least) the days
of Pontius Pilate. Hopefully, you will agree with me that, as far as formal
logic is concerned, truth need not have anything to do with appropriateness;
so I suggest we just chuck the term altogether and try to refine what we mean
by "appropriate."

To give you some idea of the dangers of playing with truth without the support
of formal logic, consider mirages. Is your perception of a patch of water on
the highway on a hot day telling you "something true about the world?" It is
certainly NOT true (even in an intuitive sense of the word) that the highway
is wet up there, as you quickly discover when you get closer. Perhaps we have
to get even more extreme and say that we cannot talk about whether any
information about the world is "appropriate" until have we, as perceiving
agents, have subjected it to the sort of interpretation I have been discussing
in my previous articles.

>Are scientific arguments about cold fusion, ozone depletion, dinosaur
>extinction, cholesterol, and so on merely sociological?
>

An affirmative answer to this question would be sort of a REDUCTIO AD ABSURDUM
reading of Kuhn. The funny thing is that buying into it no longer strikes me
as absurd. There is that old joke attributed to Wittgenstein about what the
sky must have looked like before people accepted the heliocentric model.

>I think a more reasonable position is that perceptions provide
>evidence about many aspects of the world, and that the quality of our
>assertions about the world depends on the quality of the evidence that
>supports them.
>

I guess my own position is that we do not make assertions about the world while
we are behaving in it. Therefore, it is not an issue to ask how the quality of
those assertions affect our behavior. Rather, there is a much tighter coupling
between perception and action. This is what Minsky is trying to get at when he
talks about "closing the loop" in THE SOCIETY OF MIND; and it appears that
David Chalmers, Robert French, and Doug Hofstadter are trying to come with
with a concrete implementation (such is my reading of CRCC Technical Report
49 from the Center for Research on Concepts and Cognition at Indiana
University). Logic has quite enough to do in handling that POST HOC
explanation of behavior. Don't try to burden it with more than it can
handle!

Hutchison C S

unread,
Jun 12, 1991, 9:08:17 AM6/12/91
to
I am aware of the PAULINE program; also of a program by (I think) Cornelius
Wegman which is similar; also of Carbonell's POLITICS. I still believe that
'truth' is an issue.

A semantic theory will specify the meanings of well-formed sentences in the
language. A semantics built upon a correspondence theory of truth has a
commonsensical appeal to it: what, after all, are sentences expressing
propositions about if not about the world in which language users live?
Conversely, if sentences do not express propositions about the world that
can be true or false (referring instead, for example, to speakers'
"perceptions" or internal representations of the world), then how can
conversants ever know that they are talking about the same thing(s)? A
correspondence theory of truth, and a semantics dependent upon it, rescues
the theory of semantics from the vagaries of 'mentalism' and 'solipsism'
that Stephen Smoliar fears.

Let's go back to my four newspaper headlines:

Rioting blacks shot dead by police as ANC leaders meet


Police shoot 11 dead in Salisbury riot

Rebels kill 11 ANC men

Racists murder Zimbabweans

One might argue (as has one of my correspondents, John Bradshaw) that all four
sentences may be true, albeit each expressing only a partial truth. That is,
it may be that the predicates 'police', 'rebel', and 'racist' all apply to
the agents of the four sentences. That they *may* apply does not, of course,
mean that, in a specific case, they *do* apply. Thus I may be happy about
the truth of the proposition expressed by the last sentence; others may not
be, though be quite happy about the truth of the first and second. Now let's
assume that the greater part of what I or anybody else knows is knowledge
derived from text (e.g., I know that Bogota is the capital of Columbia because
I have read it; I have never been to Columbia to check; perhaps Bogota doesn't
even exist). What do readers of each of the four sentences know? 'Know' is
a factive verb; does this then fall short of a bona fide sense of knowledge?
Does the reader of headline 1 know the same things as the reader of headline
4? If the first and last sentences both express partial truths, why do I
find it hard to persuade my fellows that the last headline represents
knowledge of the world every bit as much as does the first headline? How is
the knowledge that I and my fellow newspaper readers extract from text
integrated into what I and they know already?

It seems to me that talk of 'partial truths', 'negotiation', and so on, may
not get us very far. If I'm negotiating with you, I'm really just trying to
tell you why you are (mostly) wrong and I am (mostly) right. If I adduce
evidence to support my claims, then we may end up negotiating what counts as
evidence. We're stuck in a hopeless regress. (Try telling one billion
Christians or one billion Muslims they're wrong -- especially if it is
perfectly obvious to you that Humanistic Buddhism is the only right way. Try
negotiating with the Jehovah's Witness on your doorstep. Try telling the
free market liberal about the unspeakable suffering and brutality that
capitalism has wrought upon the cheap labour markets of the Third World.)

I mentioned Carbonell's POLITICS program earlier. It is claimed that the
program "simulates human ideological understanding of international political
events" by "an American newspaper reader" who may be on an occasion either a
"US-liberal" or a "US-Conservative". The program actually does no such thing.
Carbonell gives, not events, but *reports* of events; the text received by
the reader is the content of an utterance act by some speaker who, as much as
much as the reader, has "subjective interests, personal motivations, beliefs"
(p.2). This being so, it would be naive and indeed counter-intuitive to
assume that such interests, motives and beliefs will not have enetered into,
first, the speaker's/writer's reasons for uttering the text at all, and
secondly, his lexical and themtic choices in the formulation of his utterance.
Here is an example form Carbonell (1981):

Soviet-backed forces are scoring rapid gains against the Bhutan
government. The US is diverting tanks and M-16s, ear-marked for
the US army, in an emergency airlift to Bhutan.

Carbonell indicates the different responses that the report (or, in his words,
"event") would elicit from a "US-liberal" and a "US-Conservative". Clearly
other responses would be elicited were the report worded differently. Here
are some of my versions:

Rebels' advance against government forces prompts US airlift of
emergency arms package to Buthan.

US pumps arms into Buthan as the Washington-backed regime loses
ground to people's army.

Buthanese people's gains in struggle to liberate homeland spark
panic bid by US imperialists to bolster beleagured puppet
dictatorship.

Are the reports talking about the same event? If so, the reports are either
true or false, and there should be ways of determining the truth value of
the propositions expressed. If the reports represent merely the contents
of 'perceptions' or 'interpretations', then how can we ever know that we
are talking about the same things? (or: how can we be sure that we are
talking about the real world at all, and that therefore there is any physical
circumstance that can in principle decide the issue?)

To get things in context, despite the political flavour that my question may
appear to have taken on, my main concern is with automatic knowledge acquisition
from text (whatever kind of text it may be). My problem is: is knowledge
representation going to be about an intelligent agent's models of the
physical world or of speakers' reports about the world? This is a technical
rather than a philosophical issue since it impinges directly on what kinds
of inference and what sources of knowledge are relevant to the reasoning
process.

Chris H.

Marvin Minsky

unread,
Jun 12, 1991, 7:24:57 PM6/12/91
to
In article <1991Jun12....@kingston.ac.uk> is_...@kingston.ac.uk (Hutchison C S) writes:
>Conversely, if sentences do not express propositions about the world that
>can be true or false (referring instead, for example, to speakers'
>"perceptions" or internal representations of the world), then how can
>conversants ever know that they are talking about the same thing(s)?

The reductio ad absurdum is appropriate. Conversants never do, in
fact, know that they are talking about the same things. It is always
a matter of convention, convergence, and good fortune -- even in the
case of "mathematical truths". When you and I both talk about "that
chair over there", our internal models differ substantially, but not
enough to make most practical interactions too difficult. And the
cchir itself changes imperceptibly from one moment to the next as it
loses and gains atoms and suffers thermal agitations of its internal
degrees of freedom. There is no chair, indeed, from a modern physical
point of view, only boundaries imposed by observers; my decorator
friend regards this chair and that other one as a possibly conflicting
pair, my fried the carpenter sees it as a possibly unsound linkage of
glue and sticks, and so on. Let's grow out of this unproductive idea
of formal semantics, and low-level childish, religious, primitive
ideas about truth, and get on with the work of making machines that
can solve problems and communicate with one another as best they can.

cameron shelley

unread,
Jun 12, 1991, 6:11:21 PM6/12/91
to
In article <1991Jun12....@kingston.ac.uk> is_...@kingston.ac.uk (Hutchison C S) writes:
>I am aware of the PAULINE program; also of a program by (I think) Cornelius
>Wegman which is similar; also of Carbonell's POLITICS. I still believe that
>'truth' is an issue.
>
>A semantic theory will specify the meanings of well-formed sentences in the
>language. A semantics built upon a correspondence theory of truth has a
>commonsensical appeal to it: what, after all, are sentences expressing
>propositions about if not about the world in which language users live?

Perhaps the key word here is "about". How directly is a speaker's
statement related to a 'real' thing? In other words, how 'about' do
you want? I doubt anyone would question that speakers are interested
in expressing propositions about the world. However, you seem to be
asserting that such statements can be 'un-abouted' in order to arrive
at an exact record of the sensations the speaker experienced at some
time. To construct a model of the relationship between sensation and
communication (realistically), you are going to have to accept some
compromises, and thus let slip some of the absoluteness of the 'truth'
about which you're concerned. Do you want "the truth" or "a truth"?

>Conversely, if sentences do not express propositions about the world that
>can be true or false (referring instead, for example, to speakers'
>"perceptions" or internal representations of the world), then how can
>conversants ever know that they are talking about the same thing(s)?

We assume we have a common ground of understandings (culture, etc.) or
we use some means to establish common ground (the negotiation Steve
mentioned). Otherwise, we are very likely to misunderstand and I don't
see how any theory can do better.

>A
>correspondence theory of truth, and a semantics dependent upon it, rescues
>the theory of semantics from the vagaries of 'mentalism' and 'solipsism'
>that Stephen Smoliar fears.
>
>Let's go back to my four newspaper headlines:
>
> Rioting blacks shot dead by police as ANC leaders meet
> Police shoot 11 dead in Salisbury riot
> Rebels kill 11 ANC men
> Racists murder Zimbabweans
>

If you're after *the* truth, then you'll have to provide *the* definitions
of "riot", "racist", and so on (as I guess John Bradshaw pointed out). These
are perceptual, possibly unique categories to each speaker and not things
you can measure, weigh, or otherwise legitimately encode in a formal
language---not and capture all the variations involved. It might be
tempting to define a 'correct' riot, and then assert others are incorrect,
but then you've done nothing but become another interpreter with your
own opinion.

This situation doesn't preclude the possibility of understanding other
speakers, but it does mean understanding will require work (in the
literal sense). Occasionally, it may require the agent to shift the
'fixed-points' (axioms, assumptions) of interpretation, something I
haven't observed truth-conditional theories to be good at.

>It seems to me that talk of 'partial truths', 'negotiation', and so on, may
>not get us very far. If I'm negotiating with you, I'm really just trying to
>tell you why you are (mostly) wrong and I am (mostly) right. If I adduce
>evidence to support my claims, then we may end up negotiating what counts as
>evidence. We're stuck in a hopeless regress. (Try telling one billion
>Christians or one billion Muslims they're wrong -- especially if it is
>perfectly obvious to you that Humanistic Buddhism is the only right way. Try
>negotiating with the Jehovah's Witness on your doorstep. Try telling the
>free market liberal about the unspeakable suffering and brutality that
>capitalism has wrought upon the cheap labour markets of the Third World.)

What you seem to be saying is that the *process* of understanding (or
failing to understand) is very hard in difficult cases. I doubt anyone
would dispute that. Unfortunately, truisms don't support one position
over another. Perhaps we should consider a concrete example. What
distinction would a truth-functional theory make between the following
pair of utterances?

The cat sat on the mat.
The mat was sat on by the cat.

Although they are different, the usual theory would describe both with
the same semantic form and say that either both were true or both were
false. But if our communicative apparatus exists just to transmit the
truth, why do we have more than one token for the same proposition?
The only explanation comes from considering the speaker's desire to
emphasize "cat" or "mat" (or conversely de-emphasize the other). Even
in a description of an uncontroversial and simple event, point of view
can play a role.

Now, how true are the following two from previously?

Rebels kill 11 ANC men
Racists murder Zimbabweans

We may judge that members of one group killed members of another, but
the points of view have probably dictated the exact descriptions.
Were the Zimbabweans really "murdered" or "killed"? Most people would
say there is a difference between the two; the two terms are certainly
used to have different effects.

>To get things in context, despite the political flavour that my question may
>appear to have taken on, my main concern is with automatic knowledge acquisition
>from text (whatever kind of text it may be). My problem is: is knowledge
>representation going to be about an intelligent agent's models of the
>physical world or of speakers' reports about the world? This is a technical
>rather than a philosophical issue since it impinges directly on what kinds
>of inference and what sources of knowledge are relevant to the reasoning
>process.

Like Carbonell's (and Hovy's) systems, a model of the physical world will
require 'objective' input at some point. Since this is not really possible,
I would select option b) you give above. Also, as I hope I pointed out,
by not modelling speaker's reports, you lose information about speakers, and
speakers are presumably in the real world too. And then you have the
problem of handling reports about things which don't exist in some way (ie.
unicorns, Sherlock Holmes, rained-out ball games, etc...). A paper by
Hirst I was forced to read lately (on KR of non-existence) suggests just
going with a naive model.

I would say, in summary, that using a truth conditional KR and model will
get you somewhere, but maybe not where you would think at first blush.

Sorry for rambling. :-(

Cam

Stephen Smoliar

unread,
Jun 12, 1991, 9:34:52 PM6/12/91
to
In article <1991Jun12....@kingston.ac.uk> is_...@kingston.ac.uk
(Hutchison C S) writes:
>A semantic theory will specify the meanings of well-formed sentences in the
>language. A semantics built upon a correspondence theory of truth has a
>commonsensical appeal to it: what, after all, are sentences expressing
>propositions about if not about the world in which language users live?
>Conversely, if sentences do not express propositions about the world that
>can be true or false (referring instead, for example, to speakers'
>"perceptions" or internal representations of the world), then how can
>conversants ever know that they are talking about the same thing(s)? A

>correspondence theory of truth, and a semantics dependent upon it, rescues
>the theory of semantics from the vagaries of 'mentalism' and 'solipsism'
>that Stephen Smoliar fears.
>
I am willing to accept that mentalism bring along some vagaries which we would
like to avoid, but the only reason I was sounding apologetic about solipsism is
that I suspect it is not as bad as we have been conditioned to believe. Long
before he began to develop his work on situated automata, Stan Rosenschein was
entertaining the possibility that solipsism had a legitimate role in artificial
intelligence; but back in those days it was still fashionable to poke fun at
Bishop Berkeley. Now that we are beginning to get a handle on situated
reasoning and build systems which can actually engage it, I see no reason
to "fear" solipsism. Rather, it may rescue us from all the corners in which
we keep painting ourselves with our obsessive belief that "knowledge
representation" has something to do with that knowledge we engage to
get along in the world. Let us try to pursue this point a bit further:

> Now let's
>assume that the greater part of what I or anybody else knows is knowledge
>derived from text (e.g., I know that Bogota is the capital of Columbia because
>I have read it; I have never been to Columbia to check; perhaps Bogota doesn't
>even exist).

I shall grant you this assumption even though I disagree with it. When you get
too wrapped up in text, you tend to dismiss all the things you know that are
NOT derived from that source (such as how to tie your shoes, how to cross a
busy street, and probably even how to get to work in the morning). I would
further argue that it is all this non-text knowledge which we never even
consider articulating in text which is REALLY the "greater part" of what
anybody "knows."

> What do readers of each of the four sentences [the four headlines about an
> African event] know? 'Know' is


>a factive verb; does this then fall short of a bona fide sense of knowledge?
>Does the reader of headline 1 know the same things as the reader of headline
>4? If the first and last sentences both express partial truths, why do I
>find it hard to persuade my fellows that the last headline represents
>knowledge of the world every bit as much as does the first headline? How is
>the knowledge that I and my fellow newspaper readers extract from text
>integrated into what I and they know already?
>

Basically, I would argue that you are trying to make your point by asking a lot
of ill-formed questions! It is not the QUANTITY of your questions that matters
but rather their QUALITY! Rather than ask what a reader "knows," I would
argue that you should be asking how that sentence impacts his behavior.
At this point, you have to recognize that there is no such thing as a "generic"
reader. You can only ask about the behavior of a flesh-and-blood (so to speak)
INDIVIDUAL, rather than an abstract sentence processor. For example, for an
international trader, "knowledge" is going to have to do with doing business
in Africa. If he has an office in Salisbury, he probably has to entertain a
decision to shut that office down and evacuate his personnel. On the other
hand, a white middle-class reader in a relatively quiet town might start
reflecting on his attitude towards the blacks who moved in down the lane,
realizing that he has been unconsciously crossing to the other side of the
street whenever he sees them. We are not talking about text-based propositions
which are true or false here. We are talking about making decisions in the
real world--a form of knowledge which, I believe, Donald Schoen has come to
call "knowledge-in-action." (Since my books are still in transit, I may need
to be corrected on this.) My sincere advice to you, Chris, is to get yourself
a better set of questions before you proceed any further!

>Here is an example form Carbonell (1981):
>
> Soviet-backed forces are scoring rapid gains against the Bhutan
> government. The US is diverting tanks and M-16s, ear-marked for
> the US army, in an emergency airlift to Bhutan.
>
>Carbonell indicates the different responses that the report (or, in his words,
>"event") would elicit from a "US-liberal" and a "US-Conservative". Clearly
>other responses would be elicited were the report worded differently. Here
>are some of my versions:
>
> Rebels' advance against government forces prompts US airlift of
> emergency arms package to Buthan.
>
> US pumps arms into Buthan as the Washington-backed regime loses
> ground to people's army.
>
> Buthanese people's gains in struggle to liberate homeland spark
> panic bid by US imperialists to bolster beleagured puppet
> dictatorship.
>
>Are the reports talking about the same event? If so, the reports are either
>true or false, and there should be ways of determining the truth value of
>the propositions expressed. If the reports represent merely the contents
>of 'perceptions' or 'interpretations', then how can we ever know that we
>are talking about the same things? (or: how can we be sure that we are
>talking about the real world at all, and that therefore there is any physical
>circumstance that can in principle decide the issue?)
>

In this case I feel I can give you a straight answer: We can't! This is not
as horrible as it may seem at first blush. There are very few absolute
conclusions we draw as we go around in the world. Our "intelligence" (whatever
that means) does not reside in our ability to determine the true of
propositions but in our ability to adapt to changing positions in the
conclusions we draw and the decisions we make. Much of dialog is actually
a matter of dealing with the fact that two people are not really "talking
about the same things." Since they still have to deal with each other, dialog
becomes a tool for resolving matters; but there is never any ABSOLUTE
resolution. Rather, there are these continuous streams of behavior.
Were those streams not properly mediated, we would not be able to survive
in this confusing world.

Reply all
Reply to author
Forward
0 new messages