QUAIL '97 (Question of the Day)

Question:

What is the Knowledge Representation Hypothesis (in your own words)?

and for the ambitious/bored
What is knowledge?

Answers:
Everyone answered this one!  Lots of good answers; see esp
Eyal's answer to What is Knowledge.  And think of Patrick's
answer if they actually ask you it during the qual--you can start
by giving the arcane definition(3), and see if that gets you a 
laugh.  (I can just see it now, the new Turing Test)

Here's my attempt at the KRH:
The KRH says that any intelligent agent will have
a component that we, as outside observers, identify
as its knowledge and that this component will be
essential to the intelligent behavior of the agent.

An interesting point made in the paper is that this
hypothesis is what distinguishes KR research from
research in databases, data structures, etc.

What is Knowledge?
Power!  ha, just kidding.

I don't have a great answer...

I think a nice approach would be to identify certain
AI camps and their attitudes:
rationalist: knowledge defined functionally as what
             is necessary for intelligent behavior; more
             knowledge => better behavior (factoring in
             computation time, ie. limited rationality)
logicist: knowledge is a collection of facts and axioms...
expert-system fanatic: knowledge is power.

and for an archive of messages on the topic see:
http://www.swi.psy.uva.nl/mailing-lists/kaw/archives/0093.html


PATRICK'S ANSWER:

Um, the Knowledge Representation Hypothesis holds that in systems, composed
of structures, that act, we can see reflected in their actions propositions
which arise from the system structures, and that the propositions we
perceive are causal in the operation of the system. 


For extra credit:

knowl-edge \'na^:l-ij\ n
[ME knowlege, fr. knowlechen to acknowledge, irreg. fr. knowen]
(14c)
1 obs: COGNIZANCE
2a (1): the fact or condition of knowing something with familiarity gained
     through experience or association
2a (2): acquaintance with or understanding of a science, art, or technique
2b (1): the fact or condition of being aware of something
2b (2): the range of one's information or understanding 
2c: the circumstance or condition of apprehending truth or fact:
     COGNITION
2d: the fact or condition of having information or of being learned
     
3 archaic: SEXUAL INTERCOURSE
4a: the sum of what is known: the body of truth, information, and
     principles acquired by mankind
4b archaic: a branch of learning



AVI'S ANSWER:

The KR hyptothesis:  The workings of every intelligent machine will be
interpretable in terms of structural units that correspond to propositions
about the world.  


EYAL'S ANSWER:

> 
> What is the Knowledge Representation Hypothesis (in your own words)?

The KR hypothesis is that there will be (according to Ronny's notes -
instead of "there will be", we should put "there should be", or "it is 
worthwhile to have") , in any intelligent agent, some structures to which the
outside observer relates pieces of knowledge, and are interpretable to some
propositional representation, so that the outside observer can reason about the
meaning of these constructs.

> 
> and for the ambitious/bored
> What is knowledge?
> 

I am neither of the two, but it is the best question asked so far, so it
deserves some attention. Ronny's notes seemed to me to be a little confused,
or at least confused me a little. I will detail my perception of the subject.

When referring to knowledge I found about four (there are definitely more)
different positions.

1. Possessing Knowledge is the ability to do something (This is both Ronny's
	approach, and Ronen Brafman's approach).
Question: Does a program written in C carry the knowledge of its program?
	Its comments? Its data? 

2. The KR hypothesis, according to which, the knowledge is what the outside
	observer relates to that agent (This is McCarthy's approach, which is
	mentioned in his paper "Ascribing mental qualities to Machines" 1979
	(available both in his book, and on his web page, as do all his
	significant papers)).
Question: How does it help us analyze knowledge?

3. Possessing Knowledge is the ability to self introspect about its existence
	/truthfulness, by the agent, in a timely manner (timely=reasonable
	amount of time). (This is the position held by the Modal-representation
	of Knowledge community, and I guess many others. It was initiated
	(I think) in a book by Hintikka titled "Knowledge and Belief" 1962)

4. My opinion (somewhat contradictory to Ronny's approach) is that knowledge
	is split into two loosely coupled issues. The first is the abstract
	"piece of knowledge". The second is the	awareness/acquisition/possession 
	of the knowledge by the agent. I think all of the above do not do that
	distinction, or say that the important part is the possession.
	For the first part I will say that it is a "conceptualization" of
	something (e.g., the world). That is a piece of knowledge. It is still
	vague, but I did not have the chance to think about it more deeply.
	As far as the second one (possession), I guess any of the above might
	do. None of them is really compelling to me.


Long answer to a short question.

Eyal


URSZULA'S ANSWER:

> What is the Knowledge Representation Hypothesis (in your own words)?

KR hypothesis is an assertion that an intelligent system must consist 
of structural elements such that
(1) we (observers) can interpret them as propositions describing the 
knowledge of the system
(2) they cause the system's intelligent behavior.

Does it imply that the system with procedural knowledge are not intelligent? 

> What is knowledge?

I don't know. But I like the distinction between implicit and explicit
knowledge. Explicite - the basic facts the agent had been told and
believes in. Implicite - the deductive closure of the explicite knowledge.


PEDRITO'S ANSWER:

> What is the Knowledge Representation Hypothesis (in your own words)?

the k.r.h. claims that for processes to be considered intelligent, we as
external observers must be able to point to structures in the
representation and interpret them as representing certain knowledge the
process has. moreover, these structures must be more than mere comments;
they must actually take an active, causal role in producing the behavior
that leads us to ascribe the knowledge to the process. thus, the k.r.h.
might be viewed as one approach to bridging the gap between the knowledge
and symbol levels.

> and for the ambitious/bored
> What is knowledge?

a nice question to think about on a cold, rainy shabbat day while curled
up in a beanbag over a cup of hot chocolate. 

Back to the Question of the Day Page

Patrick Doyle November 1, 1996