Adaptive Virtual Reality Museum: A Closed-Loop Framewor for Engagement-Aware Cultural Heritage

Personalization improves engagement without disruption: The adaptive system increased reading engagement and exploration time by 2-3x compared to static presentations, while maintaining high usability scores (84.3), suggesting that AI-tailored content depth can enhance learning experiences when adjusted in real-time based on visitor behavior.

Implicit sensing raises transparency concerns: The system infers visitor engagement through continuous monitoring of gaze, head movement, and locomotion without explicit user consent or awareness, highlighting a key tension in responsible AI—the trade-off between personalization benefits and visitor privacy/autonomy in cultural spaces.

Rule-based classifiers offer interpretability advantages: The researchers used transparent rule-based engagement inference rather than opaque machine learning models, demonstrating that explainable AI approaches can enable real-time adaptation while maintaining clarity about how systems make decisions about users.

Generative AI in heritage contexts needs careful validation: The system uses Large Language Models to dynamically generate explanations, but the authors acknowledge that "engagement validation" and the appropriateness of AI-generated content for cultural heritage remain open questions requiring larger studies and stakeholder input.

Technical feasibility doesn't equal readiness for deployment: Despite achieving sub-millisecond latency on consumer hardware, the authors present this as preliminary work (N=16) and explicitly call for larger-scale investigation, cautioning against premature implementation of AI systems in cultural institutions without sufficient evidence and community engagement.