Creative Learning Agents: Computational Co-Creativity and Relational Artifacts in Education

Creativity, as the ability to develop new and useful ideas and solutions, is considered an important learning outcome of 21st-century education. With the recent advancements in artificial intelligence (AI) research, many educators are interested in adopting AI applications in the arts and design education. However, its ability to easily generate texts and images poses a risk. While statistically novel, the computational outputs are inherently derivative and absent of the pedagogical opportunities to examine its creative process. Relying on the generative capabilities of AI indiscriminately can hamper the students’ opportunity to develop their creative capacity. In this thesis, I make contributions by developing a theoretical system that art and design educators can use to integrate AI into their pedagogies and a conceptual AI model that can be adopted across academic domains and industries. The thesis is based on three original mixed-method experimental studies. In Publication I, I designed an experimental model where AI was situated as a relational artifact in creative learning. The model captured the phenomenon of students perceiving relational artifacts when the AI was situated like their peers and found the presence of AI could lead to an increased likelihood of creative learning. In Publication II, I iterated on the model with the analytical methods using the cultural-historical activity theory. While the presence of AI led to increased incidents of associative action for creative learning, they were not as effective as having an engaging peer student because AI had an objective different from the students. In addition, the actions of storytelling and the teachers' direction are critical for creative learning. In Publication III, I refined the model in a virtual environment where AI learned directly from the students’ actions using reinforcement learning. I found the emergent relational dynamics in co-creative learning constantly in flux based on their perceived commonalities and the difficulty level of the task. By synthesizing the insights from the publications, I developed the Creative Learning Activity System, a theoretical system that arts and design educators can use to examine the impact of automatizing learning activities and relational dynamics with AI in their classrooms. For future research and development, I developed Creative Learning Agent, a conceptual AI model designed to increase the number of associative actions in creative learning. I discuss these contributions and their limitations from practical, empirical, methodological, and theoretical perspectives. The thesis contributes to discussions on the meaningful integration of AI in art and design curricula and the development of an AI system whose focus on creative capacity development can make an impact by being adopted across academic domains.