Ware, Stephen G

Stephen G Ware Lab Introduction

The Narrative Intelligence Lab in the Computer Science Department at the University of Kentucky is a highly interdisciplinary research group that investigates how computers can use narrative to interact more naturally with people. We combine research in Artificial Intelligence, Narrative Theory, and Cognitive Science to create computational models of narrative. These models allow machines to understand, generate, adapt, and tell stories.

Our work has numerous applications in education, training, therapy, and entertainment. We place a high value on the rigorous empirical evaluation of our work, which often involves human subjects interacting with virtual worlds. We have a close relationship with computer games and game-like technologies which allow us to embody the algorithms that create interactive stories.

Solution Definitions in Narrative Planning

A traditional interactive narrative (for a video game, training simulation, educational application, etc.) is authored as a branching path of possibilities; non-player characters (NPCs) behave in predetermined ways. Research in narrative artificial intelligence (AI) aims to generate a wider space of possible stories than manually-authored content while still guiding the player through a coherent narrative and maintaining lifelike NPC behavior.

In particular, narrative planning builds on classical planning to let a central AI agent direct a group of NPCs. The output of a narrative planning algorithm is a sequence of NPC actions that meets a desired set of constraints. The choice of particular constraints is critical in determining whether desirable narratives will be generated and undesirable narratives avoided, as well as the computational complexity of finding a satisfying sequence of actions to begin with.

In this project, we enumerate and compare the sets of action sequences that arise for different sets of constraints for particular problems; we compare them based on how they satisfy certain narrative criteria, and we investigate how they affect the best choice of planning algorithm.

Personnel:

Stephen G. Ware, PI
Cory Siler (Ph.D. student), Added 06/28/2021

Computational Methods:

Graph-search-based AI planning (implementations are publicly available on the lab webpage)

Software:

Java

UK and non-UK Collaborators:

Grants:

Publications:

2020

Siler, Cory, and Stephen Ware. "A good story is one in a million: solution density in narrative generation problems." In Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, vol. 16, no. 1, pp. 123-129. 2020