Logo image
Back
Modeling Human Decision Making: From Complex Domains to Multi-Agent Scenarios
Dissertation   Open access

Modeling Human Decision Making: From Complex Domains to Multi-Agent Scenarios

Andrea Martin
University of West Florida Libraries
Doctor of Philosophy (PHD), University of West Florida
2023

Metrics

28 File views/ downloads
90 Record Views

Abstract

The study of human preferences has been a longstanding topic in economics, decision theory, and social choice theory. With recent advancements in artificial intelligence (AI), there is growing interest in understanding and predicting human preferences. This dissertation to investigate innovative methods for integrating behavioral knowledge into AI systems for single and multi-agents, with the aim of achieving various objectives, including the modeling of bounded-rational behaviors observed in humans. The first part of this dissertation discusses the challenges of modeling human decision-making over complex domains and proposes a combination of Multialternative Decision Field Theory (MDFT) and Fuzzy Constraint Satisfaction Problems (FCSPs) to represent an agent's preferences and cognitively model their decision-making process when presented with a sequence of choices. Three different procedures to model decision-making in complex domains are proposed and experimenatlly evaluated. The second part of this work extends the first by conducting a human study to analyze the behavioral effects of participants' decisions and how context can influence preferences. The chapter also examines the task of learning choice distributions of participants results using Recurrent Neural Networks (RNNs) and the combination of other data structures and AI models. Finally, this dissertation extends the state of the art on Stable Marriage Problems (SMPs) by incorporating simultaneously multi-attribute preferences with uncertainty and cognitive modeling of bounded-rationality. This work provides a psychologically grounded computational model of how humans may respond in the context of matching procedures. Here, a new type of matching problem, named the Behavioral Stable Marriage Problems (BSMPs), is defined, where agents represent their preferences via a psychology-based model of decision making. The notion of stability and fairness is redefined in this context, where choices are non-deterministic, and several related classes of problems are investigated. Several algorithms leveraging proposal-based approaches are proposed and experimentally evaluated, demonstrating the effectiveness of our algorithms in solving BSMPs and providing insights into the behavior of agents in multi-agent scenarios. Overall, this dissertation contributes to the field of AI by proposing novel ways of incorporating behavioral knowledge into AI models to improve cognitive modeling of human decision-making in complex domains and multi-agent scenarios. The work presented in this dissertation has important implications for understanding and simulating human behavior, which can be useful in various applications such as recommender systems, personalized medicine, and autonomous agents. By leveraging cognitive models and AI algorithms, this work provides a deeper understanding of how humans make decisions and how they can be modeled in AI systems, leading to more accurate and effective AI systems that can better predict and simulate human behavior in various contexts.
pdf
Modeling Human Decision Making2.41 MBDownloadView
Preprint Dissertation pdf Open Access

Details

Logo image