PhD Defense by Mikhail Jacob
Title: Improvisational Artificial Intelligence for Embodied Co-creativity
School of Interactive Computing
College of Computing
Georgia Institute of Technology
Date: Monday, July 29th, 2019.
Time: 2 – 4:30 p.m. (EDT)
Location: TSRB Auditorium (TSRB 118), 1st floor, Technology Square Research Building (TSRB)
Dr. Brian Magerko (Advisor) (School of Literature, Media, and Communication, Georgia Institute of Technology; School of Interactive Computing, Georgia Institute of Technology)
Dr. Ashok Goel (School of Interactive Computing, Georgia Institute of Technology)
Dr. Mark Riedl (School of Interactive Computing, Georgia Institute of Technology)
Dr. Anne Sullivan (School of Literature, Media, and Communication, Georgia Institute of Technology)
Dr. Mary Lou Maher (Department of Software and Information Systems, The University of North Carolina at Charlotte)
Improvisation is an important skill for co-creative agents to develop for success despite resource constraints, time pressure, open-ended problems, and ill-defined goals. An important subset of improvisation that has diverse applications is embodied narrative improvisation, i.e., physical improvisation of narratives with other agents using the various modalities of its body situated within a virtual or physical environment. Unconstrained human-computer embodied narrative improvisation is too challenging to undertake at present since it requires the incorporation of many cognitive faculties including narrative intelligence, social cognition, performance of linguistic/non-linguistic action, and commonsense reasoning.
This dissertation aims to explore the initial steps toward improvisational agents that can perform unconstrained embodied narrative improvisation with people. It focuses on improvisation within gestural and object-based problem domains, which are referred to collectively as movement improv domains. My research addresses a) the knowledge-authoring bottleneck and b) the improvisational action selection problem, which are both key challenges for creating improvisational agents in movement improv domains. My research seeks to validate the following thesis statement. “Embodied agents using a) interactive learning of embodied knowledge, b) formalizations of tacit knowledge, and c) creativity evaluation can address the knowledge-authoring bottleneck and improvisational action selection problem to perform movement improv with non-experts in improvisational domains having open-ended action spaces and ill-defined goal spaces, increasing user perceptions of enjoyment and agent creativity.”
The research described in this dissertation presents the following contributions. 1) real-time interactive learning systems for gestural and object interaction knowledge; 2) formal computational representations of tacit knowledge like the Viewpoints movement framework, physical attributes of objects, and improvisational response strategies to enhance the application of interactively learned knowledge; 3) computational models for evaluating the novelty, unexpectedness, and quality of perceived or generated actions; 4) models of creative arc selection and negotiation for improvisational action selection during movement improv; and 5) validated and publicly-disseminated interactive installations for studying human-computer movement improv with non-experts.