Means–ends analysis
In Simon’s discussion of the General Problem Solver (GPS), means–ends analysis connects differences between a current situation and a desired situation with actions relevant to reducing those differences. It requires representations of states, differences, and available actions, together with associations between differences and actions. (Chapter 5, printed pp. 121–122)1
Uses and boundaries
GPS illustrates a selective search process. Since a goal may require a sequence of actions and an attempted action may fail, the system must detect progress and try alternative paths. Difference reduction alone does not guarantee that a solution exists or that a particular action will succeed. (Chapter 5, printed p. 122)1
The relation to Action is procedural: means–ends analysis helps choose actions during search. It is not identical to a Policy, and does not require the belief-state formalism used in that entry. Nor is an organizational hierarchy of goals by itself a GPS procedure: a hierarchy does not specify how differences are detected or operators selected.
Operational example
For a file-editing agent, a desired state might be a document with every required heading present. Compare the current document against that state, select an edit for a missing heading, apply it, and inspect the result before continuing. Track unresolved differences and failed edits. This is an illustrative application; reducing a visible formatting difference does not establish that the document’s substantive content is correct.
Footnotes
-
Simon, Herbert A. The Sciences of the Artificial. Third edition. MIT Press, 1996. ISBN 9780262193740 (hardcover). Publisher page, paperback ISBN 9780262691918. Inspected Chapter 5, printed pp. 121–122 (PDF pp. 133–134). ↩ ↩2