Current AI has developed methods for working with data and producing fascinating results, but it lags behind in explicit representation and working with contradictory information, both of which are necessary for story understanding. Humans, for example, can efficiently reason about the multiple different perspectives of the people around them, even if the information held by each is not consistent. Moreover, most AIs cannot maintain contexts that inherit from one another and accurately describe a dynamic system of characters and their beliefs. Scone is a smart active memory system developed to be adept at exactly this. Scone has a virtual copy mechanism that allows it to efficiently make a clone of a world-model that differs only in a few pieces of knowledge. This allows Scone to represent different belief-states or time-states in a story. We explore what Scone is able to do through the classic fairytale of Little Red Riding Hood, illustrating its use in some examples. We explore how Scone can model characters' belief states and intentions and answer queries about them, allowing Scone to reason about and understand a story much like a human. Understanding stories, even fairytales, is an important problem for the future of AI, and Scone is a powerful tool for doing so. In addition to its contribution to AI, we hope that this work will help us gain interesting insights into human cognition through the use of Scone and its representation of mental states.

28 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