Representations of commonsense knowledge: Ernest Davis

✍ Scribed by Daniel S. Weld

• Publisher: Elsevier Science
• Year: 1993
• Tongue: English
• Weight: 443 KB
• Volume: 61
• Category: Article
• ISSN: 0004-3702
• DOI: 10.1016/0004-3702(93)90098-v

✦ Synopsis

Introduction

I highly recommend Davis' Representations of Commonsense Knowledge; it is a truly advanced synthesis of over twenty years of research in a broad area of artificial intelligence. As the first truly graduate-level textbook for core AI, Davis' book fills a huge gap.

By "graduate-level" I mean that the book requires a firm grasp of logic, demands considerable mathematical sophistication, and is not appropriate for students who have not completed a serious undergraduate course on AI; the book is extremely dense. By "core AI" I mean that the text is not a general book on artificial intelligence; for example, there is no discussion of expert systems, robotics, or natural language. In fact, some "core" topics are missing as well: search and learning are absent, and there is almost no discussion of reasoning, inference, or automated deduction. As the title suggests, the scope is commonsense knowledge of an intelligent being. Davis has integrated and formalized a large fraction of the knowledge required to manipulate the physical world and interact with other agents.

Since the treatment is deep and the subject area large, it is unsurprising that the book has a bias--it emphasizes representation rather than reasoning. In a manner reminiscent of Hayes' "Naive physics manifesto" [5], Davis' style is to insightfully encode the axioms necessary for competence in a domain, but