Back to search

University of Washington

Understanding Problem Solving and Collaboration in Open-Ended Environments

Abstract

dc:description.abstract

Countless human pursuits depend upon creative problem solving, especially in complex, open-ended domains. As technological support for doing this kind of work in online digital environments grows, an opportunity exists to create a new generation of intelligent problem- solving systems. These environments have the possibility of actively guiding and facilitating individual and collaborative problem solving toward the most productive outcomes. They could scaffold effective solving strategies for novices, intervene in the solving process to suggest areas of focus, or take the form of layers of machine intelligence that schedule individual and group work and dynamically adapt environmental parameters to increase solution quality. Few of these innovations will be possible, however, without a deep understanding of the problem-solving process in the domain of interest. Such an understanding would need to address the full space of strategies solvers employ, how they fit together and change over time, and how they contribute to both success and failure. In this dissertation, I investigate individual and collaborative problem-solving behavior in open-ended environments to address the question what makes groups and individuals successful problem solvers? First, I present an general framework for automatically extracting patterns of problem- solving behavior from user actions. This framework addresses constructing multivariate time series from logging data on user actions, recursively clustering these time series to produce fine-grained patterns of behavior, and selecting the model with the most understandable set of patterns. I evaluate this framework on three domains: the scientific-discovery games Foldit and Mozak and the real-time strategy game Starcraft II. These evaluations demonstrate the generality of this technique as well as how the extracted patterns illuminate high-performing behavior. Second, I develop a visualization-based analysis to identify patterns in Foldit users’ problem-solving structure. I describe the design of a domain-specific visualization of the problem-solving process in Foldit that balances preserving the complexity of the data and producing a tractable representation of behavior. I use these visualizations to identify patterns relating to exploration, optimization, and the use of automated tools. Analysis of how these patterns differ between high- and lower-performing users indicates that successful problem solvers explore more broadly and more frequently avoid local minima. Finally, I address a suite of questions concerning collaboration in Foldit. I investigate how social systems in Foldit impact individuals, finding evidence that collaboration has a positive effect on both participation and performance. Next, I explore factors associated with group performance, and find that measures of collective and individual skill have a strongest correlation with group performance. Lastly, I present an ontology of team structures in Foldit and show that the amount of collaborative refinement is far more predictive of team performance than parallel exploration or team size.

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Bauer, Aaron William
Advisor dc:contributor.advisor
  • Popovic, Zoran

Subjects

dc:subject × 5

Rights

dc:rights
Statement dc:rights
  • none
Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
http://hdl.handle.net/1773/44764

Chain of custody

source
Harvested from
University of Washington
Base URL
digital.lib.washington.edu/server/oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Bauer, Aaron William. Understanding Problem Solving and Collaboration in Open-Ended Environments. 2019. http://hdl.handle.net/1773/44764