A Study on Enhancing the Reliability of Design Evaluation Using ChatGPT-Based AI : A Comparative Analysis of Table Lamp Cases

This study investigates the effectiveness of ChatGPT as an AI-based design critique tool by examining its ability to replicate human evaluative patterns. Traditional design critique depends on expert judgment to assess aspects such as functionality, context, and creativity. As AI technology evolves, it is essential to evaluate whether such systems can perform both quantitative and qualitative assessments in design. The research employed a two-stage experiment using 20 table lamp designs. In the first stage, 10 lamps were evaluated by 32 human experts and an untrained version of ChatGPT using a 5-point Likert scale across five criteria: form, manufacturability, usability, appropriateness, and emotional appeal. Based on the discrepancies identified, ChatGPT was subsequently trained using human feedback. In the second stage, a new set of 10 lamps was assessed again by both trained ChatGPT and human experts. Post-training results indicate that ChatGPT’s error margin decreased significantly in quantitative assessments and showed partial improvement in qualitative areas. This suggests that ChatGPT can internalize certain contextual and emotional aspects when guided by expert data. The findings highlight the potential of ChatGPT as a supportive tool in design evaluation, particularly for time-efficient and scalable critique processes. Further development with more diverse design datasets is recommended to enhance its reliability and depth in qualitative analysis.