What is a Knowledge Representation? COMP34512 Sebastian Brandt - - PowerPoint PPT Presentation

What is a Knowledge Representation? COMP34512

Sebastian Brandt brandt@cs.manchester.ac.uk

(slides by Bijan Parsia bparsia@cs.man.ac.uk)

Tuesday, 25 February 2014

Where are we?

2

Tuesday, 25 February 2014

Where are we?

• We’ve been fumbling toward one

– Knowledge Acquisition

• Including Coursework

– Some formalization – Some thoughts about formalisms

2

Tuesday, 25 February 2014

Where are we?

• We’ve been fumbling toward one

– Knowledge Acquisition

• Including Coursework

– Some formalization – Some thoughts about formalisms

• But what is it that we’re producing?

– For an answer, we turn to a famous paper

• “What is a Knowledge Representation?”
• Randall Davis, Howard Shrobe, and Peter Szolovits (1993)
• http://bit.ly/whatIsA

2

Tuesday, 25 February 2014

A question!

3

Tuesday, 25 February 2014

A question!

• What is DSS’s definition of a KR?

– It’s not clear that they have one – It’s not clear that there is one! – Characterisation rather than definition

• I.e., a framework for thinking about and analysing KRs

3

Tuesday, 25 February 2014

A question!

• What is DSS’s definition of a KR?

– It’s not clear that they have one – It’s not clear that there is one! – Characterisation rather than definition

• I.e., a framework for thinking about and analysing KRs
• They do so by means of a set of “roles”

3

Tuesday, 25 February 2014

Five Roles

4

Tuesday, 25 February 2014

Five Roles

• 1. Surrogate

– That is, a representation

4

Tuesday, 25 February 2014

Five Roles

• 1. Surrogate

– That is, a representation

• 2. Expression of ontological commitment

– of the world

4

Tuesday, 25 February 2014

Five Roles

• 1. Surrogate

– That is, a representation

• 2. Expression of ontological commitment

– of the world

• 3. Theory of intelligent reasoning

– and our knowledge of it

4

Tuesday, 25 February 2014

Five Roles

• 1. Surrogate

– That is, a representation

• 2. Expression of ontological commitment

– of the world

• 3. Theory of intelligent reasoning

– and our knowledge of it

• 4. Medium of efficient computation

– that is accessible to programs

4

Tuesday, 25 February 2014

Five Roles

• 1. Surrogate

– That is, a representation

• 2. Expression of ontological commitment

– of the world

• 3. Theory of intelligent reasoning

– and our knowledge of it

• 4. Medium of efficient computation

– that is accessible to programs

• 5. Medium of human expression

– and usable

4

Tuesday, 25 February 2014

Five Roles

• 1. Surrogate

– That is, a representation

• 2. Expression of ontological commitment

– of the world

• 3. Theory of intelligent reasoning

– and our knowledge of it

• 4. Medium of efficient computation

– that is accessible to programs

• 5. Medium of human expression

– and usable

A KR is a representation of the world and our knowledge

• f it that is accessible to programs and usable

4

Tuesday, 25 February 2014

A Surrogate

5

Tuesday, 25 February 2014

A Surrogate

• Surrogate?

5

Tuesday, 25 February 2014

A Surrogate

• Surrogate?
• Representations are abstractions

– No representation captures everything – It wouldn't be useful it if did!

5

Tuesday, 25 February 2014

A Surrogate

• Surrogate?
• Representations are abstractions

– No representation captures everything – It wouldn't be useful it if did!

• The “goodness” of a representation is context sensitive

and interest relative

– It depends on the applications – We might trade off

• Accuracy for intelligibility
• Detail for performance

– Other properties might be interest and context independent – But we always exclude something

• And we may even (deliberately) misrepresent!

5

Tuesday, 25 February 2014

A Surrogate

• Surrogate?
• Representations are abstractions

– No representation captures everything – It wouldn't be useful it if did!

• The “goodness” of a representation is context sensitive

and interest relative

– It depends on the applications – We might trade off

• Accuracy for intelligibility
• Detail for performance

– Other properties might be interest and context independent – But we always exclude something

• And we may even (deliberately) misrepresent!
• What does the representation actually represent?

5

Tuesday, 25 February 2014

Ontological Commitment

6

Tuesday, 25 February 2014

Ontological Commitment

• Imperfection of reps entails choice of what to represent

– And how to – Even if we captured everything, organisation matters!

6

Tuesday, 25 February 2014

Ontological Commitment

• Imperfection of reps entails choice of what to represent

– And how to – Even if we captured everything, organisation matters!

• Representation is relational

– See surrogacy – Formalisms (typically) constrain, not determine the relations

• Model theory (as you shall see)

– Early choices constrain future choices

• But not entirely!

6

Tuesday, 25 February 2014

Ontological Commitment

• Imperfection of reps entails choice of what to represent

– And how to – Even if we captured everything, organisation matters!

• Representation is relational

– See surrogacy – Formalisms (typically) constrain, not determine the relations

• Model theory (as you shall see)

– Early choices constrain future choices

• But not entirely!
• Formal objects vs. “real” objects

– Cat SubClassOf: Animal – G1 SubClassOf: G2 – What’s the difference?

6

Tuesday, 25 February 2014

Taxonomic position

7

Tuesday, 25 February 2014

Taxonomic position

• Not just hard, but (perhaps) impossible

– No one hierarchy

• “Discussion of the hierarchies frequently will elicit comments from the

domain expert about the hierarchy structure. Not infrequently in biomedicine, there is no canonical determination of a concept’s correct tree position. For example, meningococcal meningitis may be classified correctly as both a disease of the central nervous systems and a bacterial disease.” —Modeling a description logic vocabulary for cancer research

– Hierarchies aren’t neutral!

• “In meta-utopia, the lab-coated guardians of epistemology sit down and

rationally map out a hierarchy of ideas...This presumes that there is a "correct" way of categorizing ideas, and that reasonable people, given enough time and incentive, can agree on the proper means for building a hierarchy. Nothing could be farther from the truth. Any hierarchy of ideas necessarily implies the importance of some axes over others.” —

Metacrap: Putting the torch to seven straw-men of the meta-utopia 7

Tuesday, 25 February 2014

Lab Coat View

Mary Van Rensselaer Buell (1893-1969) http://www.flickr.com/photos/smithsonian/3322785642/

http://www.well.com/~doctorow/metacrap.htm#2.5 8

Tuesday, 25 February 2014

Lab Coat View

Mary Van Rensselaer Buell (1893-1969) http://www.flickr.com/photos/smithsonian/3322785642/

http://www.well.com/~doctorow/metacrap.htm#2.5 8

Tuesday, 25 February 2014

Duelling Manufacturers

9

Tuesday, 25 February 2014

Duelling Manufacturers

9

Tuesday, 25 February 2014

Duelling Manufacturers

vs

9

Tuesday, 25 February 2014

Two Views

10

Tuesday, 25 February 2014

Two Views

10

Tuesday, 25 February 2014

Two Views

10

Tuesday, 25 February 2014

Two Views

10

Tuesday, 25 February 2014

Two Views

10

Tuesday, 25 February 2014

Two Views

(Navigational, not generalisation)

10

Tuesday, 25 February 2014

“A KR is not a Data Structure”

• Critical point!

– ER diagrams* -- 4 levels of views of data:

• 1. "Information concerning [E] and [R] which exist in our minds."
• 2. "Information structure -- organization of information in which [E] and [R]

are represented by data. "

• 3. "Access-path-independent data structure -- the data structures which

are not involved with search schemes, indexing schemes, etc."

• 4. "Access-path-dependent data structure."

– “the network model, as currently implemented, is mainly

concerned with level 4; the relational model is mainly concerned with levels 3 and 2; the entity set model is mainly concerned with levels 1 and 2.”

– The KR model is mainly concerned with level 1

– The “cognitive view” – Foreshadow role 5 “medium of human expression”

* Chen, The Entity-Relationship Model-Toward a Unified View of Data

11

Tuesday, 25 February 2014

12

Tuesday, 25 February 2014

12

Tuesday, 25 February 2014

Data Structures?

• Data structures implement (or realise) representations

– And there is a lot of choice – Many details of the structures play no representational role – That is, those details don’t commit

13

Tuesday, 25 February 2014

Data Structures?

• Data structures implement (or realise) representations

– And there is a lot of choice – Many details of the structures play no representational role – That is, those details don’t commit

• Choices at every level are critical

– But in different ways – (Foreshadow role 4 “medium of efficient computation”)

13

Tuesday, 25 February 2014

Intelligent Reasoning

• Representation and reasoning are correlative

– Tendentious, but reasonable – Representation without manipulation (or manipulability)

• isn't a representation
• If you can't use it to acquire information, in what sense does it

represent?

14

Tuesday, 25 February 2014

Intelligent Reasoning

• Representation and reasoning are correlative

– Tendentious, but reasonable – Representation without manipulation (or manipulability)

• isn't a representation
• If you can't use it to acquire information, in what sense does it

represent?

• Components

– Fundamental conception

• Deduction, human behavior, etc. etc.

– Sanctioned inferences – Recommended inferences

14

Tuesday, 25 February 2014

Intelligent Reasoning

• Representation and reasoning are correlative

– Tendentious, but reasonable – Representation without manipulation (or manipulability)

• isn't a representation
• If you can't use it to acquire information, in what sense does it

represent?

• Components

– Fundamental conception

• Deduction, human behavior, etc. etc.

– Sanctioned inferences – Recommended inferences

• There are many many many formalisms

– With different fundamental conceptions – Model theory (we shall see) is pretty unifying

14

Tuesday, 25 February 2014

Strong Cognitive Adequacy

15

Tuesday, 25 February 2014

Strong Cognitive Adequacy

• Should the conception be a model of human reasoning?

– Usually associated with the psychological or neuroscience or commonsense reasoning traditions – KRs can help us understand human intelligence – Flip side, since we are the most successful cognitive agents we know, aping us might be successful

15

Tuesday, 25 February 2014

Efficient Computation

• Programs have to work with representations

– The representation management system is a component – Programmers will compensate for inefficient components – (Strong Cognitive Adequacy to Efficient Computation fallacy)

16

Tuesday, 25 February 2014

Efficient Computation

• Programs have to work with representations

– The representation management system is a component – Programmers will compensate for inefficient components – (Strong Cognitive Adequacy to Efficient Computation fallacy)

• Most interesting systems work at “large” scale

– Consider databases

16

Tuesday, 25 February 2014

Efficient Computation

• Programs have to work with representations

– The representation management system is a component – Programmers will compensate for inefficient components – (Strong Cognitive Adequacy to Efficient Computation fallacy)

• Most interesting systems work at “large” scale

– Consider databases

• Representations get complex quickly

– People need prosthetics to work well with them

16

Tuesday, 25 February 2014

Human Expression

• Humans interact with representations

– Effectively – At least of certain kinds

17

Tuesday, 25 February 2014

Human Expression

• Humans interact with representations

– Effectively – At least of certain kinds

• People must work with KRs

– Generating them – Using them to build systems – Using them when using the system

17

Tuesday, 25 February 2014

Human Expression

• Humans interact with representations

– Effectively – At least of certain kinds

• People must work with KRs

– Generating them – Using them to build systems – Using them when using the system

• Weak Cognitive Adequacy

– KRs must be usable

17

Tuesday, 25 February 2014

Five Roles

• 1. Surrogate

– That is, a representation

• 2. Expression of ontological commitment

– of the world

• 3. Theory of intelligent reasoning

– and our knowledge of it

• 4. Medium of efficient computation

– that is accessible to programs

• 5. Medium of human expression

– and usable

Representing Computing Interacting

18

Tuesday, 25 February 2014