Story-enabled Hypothetical Reasoning

Story understanding is a central competence that illuminates all other aspects of human intelligence. In this work, I demonstrate how our story understanding ability sheds light on our ability to think in terms of hypothetical situations. Using the Genesis story understanding system as a substrate, I develop a story-enabled hypothetical reasoning system that models several high-level human abilities, including judging actions in terms of moral alternatives, contextualizing stories by considering what could have otherwise happened, and deliberating about personality to decide what characters will do next. In developing this system, I built many new computational mechanisms and representations, including a program for answering whatif questions, a side-by-side story comparator, rules for making presumptive inferences, heuristics for evaluating personality fit, and a problem-solving approach for evaluating moral character. Together, they take Genesis’s story understanding capabilities to another level and advance our understanding of human intelligence. Thesis Supervisor: Patrick H. Winston Title: Ford Professor of Artificial Intelligence and Computer Science