Abstract
Particularly in their first year of education, design students face challenges in understanding the assessment processes, as the answers are not simply right or wrong, but instead judged on a spectrum of better and worse. This confusion stems from the tacit nature of design knowledge, which is difficult for tutors to articulate. The open-ended nature of design problems, which allows for multiple correct solutions, augments this issue, leading to confusion among students about their performance. To address this subject, this paper proposes an AI pipeline to support both summative and formative assessment by utilising computer vision to enhance the transparency, objectivity, and efficiency of design assessments. First, it uses a Convolutional Neural Network (CNN) model for grading, using monochromatic photographs of design students’ work as visual data, which classifies work into five grade bands with 88% accuracy. Then, it implements a Large Language Model (LLM), GPT-4o, by leveraging recent developments to generate formative feedback aligned with the project brief. Findings indicate that the AI pipeline could clarify tacit criteria, objectify grading, and reduce tutor workload; however, it raises concerns about alignment, balanced usage, and the design’s contextual and multifaceted requirements, which may limit other potential use cases for design assessments. Despite its specific dataset, the workflow offers a scalable template for design-related programmes.
DOI
https://doi.org/10.21606/drslxd.2025.136
Citation
Alan, A.C.,and Gelmez, K.(2025) Can AI assess design students’ projects? Discussing the future of assessments with a proof of concept, in Clemente, V., Gomes, G., Reis, M., Félix, S., Ala, S., Jones, D. (eds.), Learn X Design 2025, 22-24 September 2025, Aveiro, Portugal. https://doi.org/10.21606/drslxd.2025.136
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License
Conference Track
Full Paper
Can AI assess design students’ projects? Discussing the future of assessments with a proof of concept
Particularly in their first year of education, design students face challenges in understanding the assessment processes, as the answers are not simply right or wrong, but instead judged on a spectrum of better and worse. This confusion stems from the tacit nature of design knowledge, which is difficult for tutors to articulate. The open-ended nature of design problems, which allows for multiple correct solutions, augments this issue, leading to confusion among students about their performance. To address this subject, this paper proposes an AI pipeline to support both summative and formative assessment by utilising computer vision to enhance the transparency, objectivity, and efficiency of design assessments. First, it uses a Convolutional Neural Network (CNN) model for grading, using monochromatic photographs of design students’ work as visual data, which classifies work into five grade bands with 88% accuracy. Then, it implements a Large Language Model (LLM), GPT-4o, by leveraging recent developments to generate formative feedback aligned with the project brief. Findings indicate that the AI pipeline could clarify tacit criteria, objectify grading, and reduce tutor workload; however, it raises concerns about alignment, balanced usage, and the design’s contextual and multifaceted requirements, which may limit other potential use cases for design assessments. Despite its specific dataset, the workflow offers a scalable template for design-related programmes.