IBP-Laforia
1996/22:
Rapport de Recherche Laforia /
Laforia research reports
6 pages - Octobre/October 1996 -
Document en anglais.
PostScript : 62 Ko /Kb
Titre / Title: AUTOMATED GENERATION OF INTERESTING CONJECTURES
Abstract : Since we can't give a system all the knowledge of the world, A.I. systems will have to discover it. They will also have to organize this knowledge in order to make it useful. This task is not easy because these systems may produce too many useless concepts which at last overwhelm them. One solution to the problem of creating only good concepts is to guide the system with interesting conjectures. Unfortunately, the problem is just translated since the new concepts will also need conjectures. A better solution is then to enable the system to discover these conjectures. In this paper we present a general method that generates interesting conjectures. This method is based on the notions of space representation and generate-and-test procedure using knowledge on conjectures to constrain the search. We present results in two domains: the elementary set theory and a subset of the number theory. These results come from experiments realized with SECTE which is the successor of a new implementation of Lenat's AM system(Lenat 82).
Publications internes Laforia 1996 / Laforia research reports 1996