Scribd
Upload
62 views4 pages

Computational Approaches To Analogical Reasoning Background

This document discusses computational approaches to analogical reasoning. It describes two main computational theories - structuralist and case-based views. Structuralist views see analogy as based on isomorphism between representations, while case-based views see it as overlap in relevant dimensions between cases. The document also evaluates computational theories based on their predictiveness, applicability, scope, and simplicity in modeling analogical reasoning.

Uploaded by

marcio-moto
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
62 views4 pages

Computational Approaches To Analogical Reasoning Background

This document discusses computational approaches to analogical reasoning. It describes two main computational theories - structuralist and case-based views. Structuralist views see analogy as based on isomorphism between representations, while case-based views see it as overlap in relevant dimensions between cases. The document also evaluates computational theories based on their predictiveness, applicability, scope, and simplicity in modeling analogical reasoning.

Uploaded by

marcio-moto
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Computational approaches to analogical reasoning

Background

• AI research recognizes importance of analogical reasoning in a wide range of human


activities

• Main interest for us: can the computational approaches contribute to our understanding of
analogical reasoning as used in the sciences?

• Ideally, each AI program is based on a computational theory or model of analogy: a set


of assumptions, precise enough to implement in a computer program, about how the
process of analogical reasoning does or should take place.
- restricted aspects of analogical reasoning
- many programs work on only a handful of carefully chosen examples.
- trade-off between scope and ability to exploit detailed knowledge about the problem
areas
Classification

• Most computational theories of analogy fall into two camps: structuralist and case-based
views. Both approaches attempt to model the way humans use known solutions to solve
new problems.

A. Structuralist view: analogical reasoning is essentially syntactical.


- Analogies founded on parallel representations of knowledge about two domains.
- ideal analogy: isomorphism. In general, value of an analogy is measured by the
degree to which it approximates isomorphism.
B. Case-based view: analogies derive from the perception that two domains share
salient or relevant dimensions.
- these are characteristics we know to have been associated in the past with outcomes
of interest, or to have interesting links to those outcomes.
- The strength of an analogy depends upon the extent of relevant overlap between the
source and target domains.
Q: Are both based on a deductivist model of analogical reasoning?
Three main differences:
• Scope. Structuralist programs tend to be general-purpose analogical reasoners.
They can find and utilize analogies in virtually any context.
By contrast, case-based reasoners are typically narrow; they work within a
bounded ‘region’ of fairly similar cases. The most important features can be
adequately described using a finite and pre-established vocabulary.
• Flexibility of knowledge representation. Structuralist programs employ a
general-purpose representation of data about the source and target domains, such
as some form of predicate calculus. Typically, there are many ways to describe the
two domains in the formal language, and few guidelines for choosing between
different representations.
The main alternative, used in case-based reasoning (CBR), is to represent domains,
or cases, with a set of stereotypes that provide a uniform way to store information.
Stereotypes are typically implemented in data structures known as scripts or
frames. Choosing stereotypes ensures a basis for comparing cases; however, it
limits the scope of a program.
• Treatment of Relevance. On the structuralist approach, the relevance of a
predicate or function is determined by systematicity: the extent to which it enters
into complex networks of relationships. By comparison, CBR programs are
oriented around a set of indices that includes every factor deemed to be relevant.
Evaluation

• Focus on potential for illuminating role of analogies in plausibility reasoning in the sciences.

1) Predictiveness: extent to which a computational theory makes definite predictions about


which analogies are more plausible than others.
- clarity and rigor of computational theories: virtues in their favour?
- vagueness about input and knowledge representation
- Predictiveness concerns both the evaluation criteria embodied in the program and the
conventions about knowledge representation

2) Applicability: extent to which the evaluation criteria used by a program are justified.

- reasonable procedures to discriminate between good and bad analogical reasoning


(not just success in hand-picked cases)

- Three aspects: definition of standards; implementation of these standards in the


program or representation conventions; a philosophical argument that the programmer’s
definition is reasonable.

3) Scope or generality: extent of the class of problems to which the theory is applicable.

- does the theory account for diverse phenomena associated with analogical reasoning?
- able to handle a large variety of analogies?

4) Simplicity: ability to account for many different types of analogies using relatively little
conceptual apparatus.
Since analogical reasoning is a complex phenomenon, however, the importance of
simplicity should not be exaggerated.

Comments:

i) Virtues familiar from philosophical discussions of choice between scientific theories


(e.g., Kuhn’s Values, Objectivity and Theory Choice).

ii) Trade-off between predictiveness/applicability and scope. CBR programs more


reliable, but highly specialized. General-purpose, structurally based analogical
reasoners rely on unspecified conventions about representation, so that predictiveness
and applicability are problematic.

You might also like