Creativity Assessment and Machine Creativity Using Generative AI in a Sandbox Physics Game (Stage 2, 8:17 AM)

• In Event: e-Lightning Ed-Talk Session 1 - Stage 2

Abstract

This study examines GPT 4o’s capability in assessing and generating creative game levels in the educational sandbox game Physics Playground, which emphasizes Newtonian physics. The research pursued two goals: (1) evaluating GPT 4o’s ability to score the creativity of user-designed levels and (2) testing its potential to generate new, creative levels. Using prompt engineering and structured creativity rubrics, the model’s assessments were compared to human ratings, achieving strong alignment (r = .81) when guided by a detailed rubric. Level generation was explored through direct prompts and a vector image–to–JSON approach, with the latter producing more aesthetically and structurally complex levels.

Authors

• Seyedahmad Rahimi, University of Florida

• Salah Esmaeiligoujar, University of Florida

• Hongming Li, University of Florida