Publication Date
2021
Document Type
Thesis
Committee Members
Michael L. Raymer, Ph.D. (Committee Chair); Michael T. Cox, Ph.D. (Committee Member); Mateen M. Rizki, Ph.D. (Committee Member); Tanvi Banerjee, Ph.D. (Committee Member); Dustin Dannenhauer, Ph.D. (Committee Member)
Degree Name
Master of Science (MS)
Abstract
Modeling an autonomous agent that decides for itself what actions to take to achieve its goals is a central objective of artificial intelligence. There are various approaches used to build autonomous agents including neural networks, state machines, utility functions, learning agents, and cognitive architectures. In this thesis, we focus on cognitive architectures. Our approach uses specific knowledge of the world, the goals they pursue, and the actions being performed. Most agents do what they are told (i.e., achieve the goals given to them by a human), but a genuinely autonomous agent does more. It can formulate its own goal or change the goals given to it. Sometimes an agent should even refuse to accept a given goal because of issues that violate its preferences or motivations. Rebellion (the refusal of an autonomous agent to accept a goal) is a vital trust requirement if critical conditions are not met, such as ethics, safety, and behavioral correctness. We will exploit rebellion to realize an agent framework that is both goal-driven and adaptive. Using an existing cognitive architecture, we implemented a rebel agent that rejects goals having undesirable effects. In particular, the agent uses explicit expectations about its goals and reasons about the positive and negative interactions between them to make such decisions. We also implemented a learning version of this agent, which learns from its mistakes and seeks to avoid similar mistakes in the future. In this work, we seek the maximum achievement of goals when actions have both desirable and undesirable effects and demonstrate improved goal achievement by incorporating learning algorithms within a goal-driven rebel agent framework.
Page Count
51
Department or Program
Department of Computer Science and Engineering
Year Degree Awarded
2021
Copyright
Copyright 2021, all rights reserved. My ETD will be available under the "Fair Use" terms of copyright law.
ORCID ID
0000-0001-7869-0193
