BCS SGES Evening Lectures

Department of Computer Science
Birkbeck College
University of London
Malet Street
London WC1E

Further details of location


Wednesday 27th March 2002 (6 pm)

Gerry Wolff (University of Wales)

UNIFYING AI

ABSTRACT

Since 1987, I have been developing a conceptual framework and computer models intended to integrate concepts in cognitive science, artificial intelligence and other aspects of information processing in natural and artificial systems. This talk will present an overview of the framework and its range of applications, focussing mainly on the way it may be applied in AI and computing. Examples will be illustrated with output from the computer models.

The talk will not merely repeat my article about these ideas in the Autumn number of Expert Update. Different examples will be presented and there will be opportunities for questions and discussion.

The conceptual framework -- "information compression by multiple alignment, unification and search" (ICMAUS) -- originated from the long-established idea that 'economy', 'parsimony' or 'simplicity' has an important role in human perception and cognition. More immediately, the framework is founded on principles of Minimum Length Encoding pioneered by Solomonoff, Wallace and Boulton, Rissanen and others. The framework provides an interpretation for established models of computing such as the Universal Turing Machine and the Post Canonical System. It also provides an interpretation for a range of concepts in mathematics and logic.

The system may be used to parse sentences syntactically and produce them. Discontinuous dependencies in syntax can be marked in a simple and direct manner. The system has been designed to facilitate the integration of syntax and semantics but this area has not yet received close attention. The system is able to recognise patterns despite errors of omission, commission or substitution. It also supports the recognition of patterns at multiple levels of abstraction in a hierarchy of classes, with inheritance of attributes from higher levels to lower. The system may function as a knowledge base with facilities for best-match information retrieval and indirection in information retrieval.

Various kinds of probabilistic reasoning may be modelled in the system including probabilistic 'deduction' and abduction, chains of reasoning (and other forms of composite reasoning), nonmonotonic reasoning, 'explaining away', and solving geometric analogy problems. Given appropriate input, the most recent computer model can achieve unsupervised learning of simple grammars. The system has potential for knowledge discovery and as a self-organising knowledge base. More work is needed on these aspects of the framework.


Up until 1983 GERRY WOLFF was a lecturer in psychology at the University of Dundee working on computer models of language learning. Following a one-year research fellowship with IBM in Winchester, he worked for four years as a software engineer with Praxis Systems in Bath. Since the late 1980s, as a lecturer and now research fellow in the School of Informatics, University of Wales at Bangor, he has been developing the 'ICMAUS' framework of ideas which is the subject of this talk.


The Evening Lectures are free to both members and non-members of SGES.

For further information contact: Dr. Chris Christodoulou, Department of Computer Science, Birkbeck College chris@dcs.bbk.ac.uk.

[Evening Lectures Home Page]