Effective use of AI-generated Historical Reconstructions in representing Multivocality in Heritage (cum laude)

Wimalasuriya, Ravini. "Effective use of AI-generated Historical Reconstructions in representing Multivocality in Heritage". MA thesis, Vrije Universiteit Amsterdam, 2025

This research thesis explores the intersection of Cultural heritage and Generative Artificial Intelligence (Gen-Al), focusing on the potential of Al-generated historical image reconstructions to facilitate multivocality in heritage interpretation. While recognising that heritage is not merely the factual historical past, but a process of appropriation shaped by power dynamics and socio-cultural contexts, the study positions itself within critical heritage studies. Accordingly, the study underscores the importance of representing multiple, often underrepresented perspectives, not blindly but critically, to address any historical silences and foster inclusivity. Accordingly, this research evaluates three prominent AI image generation models using three heritage test cases with multiple perspectives. A total of 13 user prompts were transformed into 39 images, which were then linguistically analysed and followed by a systematic visual assessment by heritage experts, using predefined criteria. The study reveals both the strengths and limitations of current generative AI models in producing historically accurate and culturally/contextually sensitive (image) reconstructions. Based upon the findings, a 'Conceptual Framework cum Guideline for User Prompt Optimisation' is proposed by this study to guide users in crafting effective prompts to obtain better AI-generated historical visualisations. Furthermore, it emphasises the necessity of cross­verification of AI outputs with credible historical sources and ethical considerations in generating images. As practical implications, the study suggests utilizing this simulation for visitor engagement, development of interpretive strategies and research. This thesis concludes by emphasizing that generative AI should not be regarded as a definitive source of historical truth, but rather as a supplementary tool for facilitating multivocality and enhancing public engagement with cultural heritage. Therefore, it is crucial that people use this technology in a meaningful, responsible, and critical manner to represent our human past in an inclusive and dynamic way.

Full Version