Gen-Diaolou: An Integrated AI-Assisted Interactive System for Diachronic Understanding and Preservation of the Kaiping Diaolou

Paper Title

Gen-Diaolou: An Integrated AI-Assisted Interactive System for Diachronic Understanding and Preservation of the Kaiping Diaolou

Publication Info

• Topic area: AI-assisted systems for cultural heritage understanding and preservation.
• Keywords: Cultural heritage, Kaiping Diaolou, generative AI, diachronic understanding, preservation awareness, human-computer interaction, interactive systems, participatory design, narrative engagement, museum studies.

Background and Problem

• Problem / challenge: The Kaiping Diaolou, a UNESCO World Heritage Site, faces challenges in conservation, sustainable development, and community engagement. Existing approaches to cultural heritage (CH) engagement often lack participatory and immersive elements, and generative AI (GenAI) applications in CH are limited by risks of historical inaccuracy, aesthetic homogenization, and cultural misrepresentation.
• Significance: Addressing these challenges is critical to preserving the Diaolou’s historical and cultural significance, fostering public awareness, and enabling sustainable heritage practices.
• Motivation and related work: Prior research has explored AI applications in CH, such as virtual reconstructions and storytelling, but these often fail to balance creativity with historical accuracy. There is a gap in deploying GenAI-based systems in real-world CH contexts, particularly for fostering diachronic understanding and preservation awareness.

Solution

• Proposed approach: Gen-Diaolou, an integrated AI-assisted interactive system, combines a Knowledge Module for learning and a GenAI Module for creative exploration to support CH understanding and preservation.
• Novelty:
  1. Introduction of a diachronic narrative framework linking past, present, and future perspectives.
  2. Development of authenticity guardrails to ensure historically accurate and culturally appropriate generative outputs.
  3. Integration of a conversational agent to enhance immersive, inquiry-based learning.
  4. Empirical evaluation of the system’s impact on knowledge acquisition, preservation awareness, and user experience.
• Procedure and key techniques:
  • Conducted a formative study (N=14) to identify design goals through expert interviews and co-design workshops.
  • Developed Gen-Diaolou with modules for historical learning, risk estimation, and future preservation.
  • Implemented authenticity guardrails to balance creativity with historical fidelity.
  • Evaluated the system through a pilot study (N=18) and a museum-based field study (N=26), comparing a baseline condition to a GenAI-augmented condition.

Results

• Concrete findings:
  • Significant improvements in factual knowledge (mean gain of 6.11 items in the pilot study) and conceptual understanding (2.15 vs. 1.31 items in the field study).
  • Higher knowledge retention in the GenAI-augmented condition (8.77 vs. 7.46 out of 10).
  • Increased preservation awareness across five dimensions, with statistically significant gains in the GenAI condition.
• Advantage over baselines:
  • The GenAI-augmented condition outperformed the baseline in conceptual learning, knowledge retention, and preservation awareness.
  • Enhanced creativity support, immersion, and usability metrics in the GenAI condition.
• Experiments / evaluation:
  • Pilot study assessed usability, workload, and learning outcomes using quizzes, NASA-TLX, and UEQ.
  • Field study compared baseline and GenAI conditions using knowledge quizzes, CAI-CH, SUS, and CSI scales, along with qualitative interviews.
• Limitations and future work:
  • Limited representation of professional heritage practitioners in the participant pool.
  • Short-term evaluation of knowledge retention; longitudinal studies are needed.
  • Future research should explore broader community involvement and refine scaffolding for interpretive agency.

Summary

Gen-Diaolou is an AI-assisted system designed to enhance engagement with the Kaiping Diaolou through diachronic narratives and creative co-creation. Empirical studies demonstrated its effectiveness in improving knowledge, fostering preservation awareness, and supporting immersive, inquiry-based learning. The system’s integration of authenticity guardrails and a conversational agent ensures historical fidelity while enabling creative exploration. This work highlights the potential of human–AI collaboration in cultural heritage, offering a framework for future systems that position users as active interpreters and stewards of heritage.