Customer Reviews

Building Large Knowledge-Based Systems: Representation and Inference in the Cyc Project

5 star:		(2)
4 star:		(0)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

This book, (written in 1989) captures the essence of writing a large software system which has a practical bent. The particular system developed here is the Cyc Knowledge Base. It outlines the criteria for inclusion of common sense knowledge in a computer system, with attention to the aspects which are necessary to transform raw facts into organized knowledge suitable for inference.

As it was written in the middle of the ten year Cyc project, it has little information about the eventually produced system, and more emphasis on the system as it was at that time. There is still a great deal of value in the book, however, as someone wanting to understand the current Cyc system can still get a feel for the project and the specialized mindset needed to contribute to it. The philosophical asides are essential for understanding the culture of creation within the Cyc project, with much of the culture continuing to the Cycorp organization.

The two authors are still highly involved in the artificial intelligence field, and highly respected. Dr. Lenat is currently head of Cycorp, and Dr. Guha has gone on to be involved in various companies and with the internet RDF work.

Within an admittedly focused niche, this book is extremely useful with very powerful ideas that are well presented. I would highly recommend this book to anyone interested in high tech fields with explosive growth potential.

"This book is a midterm report on the Cyc project, an attempt to construct a large knowledge base that codifies the "commonsense" knowledge that people use every day. Such a knowledge base is necessary if Natural Language Processing is going to progress, but you can only disambiguate a statement like "Squad Helps Dog Bite Victim" if you know something about dogs and squads.

As a midterm report, it omits many of the developments that are in Cyc today, such as microtheories, but I'm not aware of any equally synoptic documentation of Cyc today.

I read this book because I'm frustrated with the state of the art in RDF-based knowledge management tools. When "Anybody can say anything about any topic" you need a system that can ingest "facts" and decide if it wants to believe them or not. Cyc is the most advanced system to date in this area, and this book lays out many of the issues involved as well as solutions and pitfalls.

In my opinion, the next generation of Cyc-like systems are going to take a "memomic" approach: Cyc's top-down ontology looks complex and unfamilliar to people because it doesn't reflect the way we think. People build ontologies and taxonomies in their heads to address problems that are in front of them; today the task is to build an automated ontologist, not to build the ultimate ontology.