Abstract
The objective of this work is to study the perceived complexity of 3D shapes from a human and a large generative model (i.e., ChatGPT) point of view. This work helps to better understand what makes 3D shapes, which are frequently used in spatial visualization tasks, perceived as complex. It also explores how well ChatGPT can capture the consensus of humans as to what makes shapes perceived as complex. Spatial visualization skills are correlated to success in many STEM fields. To enhance Virtual Reality applications aimed at developing spatial visualization skills, models capable of automatically generating shapes of varying complexities could be used to tailor tasks according to users’ skill levels. However, it is important to first understand how humans perceive the complexity of 3D shapes, and how this relates to their performance in spatial visualization tasks. The results of this work indicate that some visual features of 3D shapes, like symmetry, are correlated to their perceived complexity and the performance of individuals on spatial visualization tasks. More importantly, the results show that ChatGPT can generate shapes that are perceived as having different degrees of complexity by humans. The findings support the capabilities of large generative models, like ChatGPT, to capture aspects of human consensus, even in subjective matters such as the perceived complexity of 3D shapes. Hence, these models could potentially be used to automatically generate content, like for VR applications, which are tailored to an individual.