Scaffolding Creativity or Homogenizing Outputs? GPT-based Virtual TA’s Impact on Student Ideation in Graduate Education

Abstract

The rapid integration of large language models in education presents both opportunities and challenges for creative ideation in graduate learning. This quasi-experimental study examines JeepyTA, a GPT-powered teaching assistant and discussion agent, and its impact on students’ educational applications of a commercial video game. Comparing 2023 (without JeepyTA) and 2024 (with JeepyTA), findings show that JeepyTA significantly increased the production of fully formed ideas, suggesting its role in scaffolding ideation. However, its impact on idea quality depended on prior domain knowledge, with familiarity moderating effectiveness. While JeepyTA boosted idea quantity, it did not simply generate superficial responses. Notably, JeepyTA-driven ideation led to a higher volume of contributions but with creative homogeneity, whereas student-driven ideation produced more unique, context-rich ideas. These findings highlight the importance of prior knowledge in maximizing the benefits of generative AI and offer insights into integrating AI tools to support, rather than substitute, independent creative thinking in graduate education.