Date of Thesis

Spring 2019

Thesis Type

Honors Thesis

Degree Type

Bachelor of Science in Computer Science and Engineering

Major

Computer Science & Engineering

Minor, Emphasis, or Concentration

Mathematics

First Advisor

Christopher L. Dancy

Second Advisor

L. Felipe Perrone

Keywords

ACT-R, cognitive architecture, Project Malmo, exploration exploitation trade-off, Adaptive Gain Theory, reinforcement learning

Abstract

Cognitive architectures such as ACT-R provide a system for simulating the mind and human behavior. On their own they model decision making of an isolated agent. However, applying a cognitive architecture to a complex environment yields more interesting results about how people make decisions in more realistic scenarios. Furthermore, cognitive architectures enable researchers to study human behavior in dangerous tasks which cannot be tested because they would harm participants. Nonetheless, these architectures aren’t commonly applied to such environments as they don’t come with one. It is left to the researcher to develop a task environment for their model. The difficulty in creating one prevents cognitive architectures from being utilized in more advanced studies. This project aims to address that issue by building a bridge between ACT-R and Project Malmo, an artificial general intelligence test suite. The bridge facilitates easy integration of new missions by allowing researchers to specify how to create the world and update it without worrying about the overhead of Malmo. Furthermore, this study analyses how well ACT-R’s utility learning system will adapt in a complex environment. The Adaptive Gain Theory was implemented to improve how the system adapts by using task engagement, derived from measures of utility, to dynamically modify noise. The system was tested using a modified Symbolic Maze task. Tests revealed the parameters of the Adaptive Gain mechanism need to be refined to have a greater impact on model performance. Nonetheless, the bridge provides an interface for ACT-R to be used to study decision making in a complex environment. Improving the bridge will enable more advanced experiments to be conducted whilst improving the Adaptive Gain Theory implementation will move us one step closer to understanding everyday intelligent behavior.

Share

COinS