Institute for Human and Machine Cognition
The idea of putting 'common sense' knowledge into a program was one of the earliest goals of AI. Progress so far has been rather disappointing, however, especially when compared with progress in other AI topics, such as machine vision or expert systems. Problems that were first identified 25 years ago still have not been adequately addressed, and the complexity of test examples has not significantly increased during a period when computer speeds and memory capacity have increased by many orders of magnitude. This talk reviews some of the ground assumptions of traditional 'logical common sense' and suggests how these caused the field to run into a number of blind alleys, and gives an overview of an alternative approach which may provide a way to make more useful progress.