"In Search of Common Sense"

10/28/05: 10:30 AM, webcast
11th Floor Large Conference Room
Host: Tim Chklovski, schedule

Abstract: It is generally accepted that the intelligent machines of the future will require large amounts of general ("common sense") knowledge to function. However, constructing and exploiting large common-sense knowledge bases remains a formidable challenge in AI. In my work at Boeing and before, I've been involved in several attempts to make a dent in this challenge, including: assembling knowledge bases from modular components; using Controlled Language tools for rapid, manual knowledge acquisition; and automatically extracting knowledge from text corpora. In this presentation, I'll trace some of the journeys, dead-ends, and rays of hope I've had along this path, speculate on why automated common-sense is so challenging, and suggest some ways forward. I'll conclude by commenting on what might be called "the myth of common sense knowledge" -- namely, that knowledge acquisition really is the bottleneck -- and suggest that perhaps a radical rethinking of reasoning methods is a more pressing concern.

About Peter Clark: Peter Clark is an Associate Technical Fellow in Boeing's Mathematics and Computing Technology Organization in Seattle, WA, where he leads research in the areas of knowledge-based systems, machine reasoning, and controlled language processing. Prior to that he worked at the University of Texas at Austin, the Canadian National Research Council, and the Turing Institute in the UK. He received his PhD in 1991 and MSc in 1985.

Last updated: Mon Jun 19 17:44:06 2006