We live in a world that is getting closer and closer to the dream of broad, human-like AI. However, most resources are dedicated to data-based AI, machine learning, and text processing. These methods have given us great results in certain simple tasks but are not sufficient to achieving true language understanding of stories, which must combine the text being processed with background knowledge. They work very well, for example, in parsing absolute truths and facts but fall short when representing the ideas of deception and differing mental states. We believe in that building common-sense reasoning in computers is crucial to achieving a true and realistic representation of real world scenarios. Here, we present an experiment to show the potential and promise that knowledge-based AI holds, by showing that reasoning and understanding tasks are possible using Scone, a common-sense reasoning engine. We show that Scone can understand and accurately answer queries about a Sherlock Holmes story that involves multiple mental states, truths, and a whole lot of deception. We are focusing on representing and reasoning about the objects, characters, beliefs, and events in the story, and not on syntactic processing of the raw text input, which is handled by a sister project. This representation is something we believe text processing and information extraction cannot do. Hence we argue for the need for a system that combines the power of knowledge-based and data-based AI.

33 pages

Thesis Committee:
Scott Fahlman (Chair)
Daniel Fried

Srinivasan Seshan, Head, Computer Science Department
Martial Hebert, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by reports@cs.cmu.edu