Exploring Art with modern AI LLM-Powered Guides for Personalized Museum Experiences
Translated title
Exploring Art with modern AI - LLM-Powered Guides for Personalized Museum Experiences
Author
Tue, Tobias
Term
4. Term
Publication year
2025
Submitted on
2025-01-17
Abstract
This thesis investigates whether a large language model (LLM)-powered conversational agent can function as a personalized virtual museum guide, and how visitors perceive interacting with such a guide in a museum context. To explore this, a prototype was built that presents four paintings in augmented reality via a head-mounted device and enables voice-based natural language dialogue with the guide. The prototype was evaluated in an exploratory field study at a museum with 22 visitors, who completed a questionnaire using a multimedia museum guide evaluation scale and took part in brief semi-structured interviews. Quantitatively, participants rated the prototype positively, and logs showed that most questions were fact-based (76.32%). Qualitative themes indicated that visitors valued personalized and selective information delivery, while also encountering challenges such as gaps in the information provided, difficulty formulating spontaneous questions, and discomfort speaking aloud in public. The study highlights the potential of LLM-powered virtual guides to enhance engagement and learning in museums, while surfacing design considerations and the need for further research. As an exploratory investigation, the findings are not definitive but offer a foundation for developing more accessible and user-centered solutions.
Denne afhandling undersøger, om en LLM-drevet samtaleagent kan fungere som en personlig virtuel museumsvejleder, og hvordan besøgende oplever at interagere med en sådan guide i en museums-kontekst. Til formålet blev der udviklet en prototype, hvor fire malerier præsenteres i AR (udvidet virkelighed) via en hovedmonteret enhed, og brugerne kan føre stemmebaserede dialoger med guiden i naturligt sprog. Prototypen blev evalueret gennem et eksplorativt feltstudie i et museum med 22 besøgende, som efter brug udfyldte et spørgeskema baseret på en skala til evaluering af multimedieguider samt deltog i en kort semistruktureret interview. Kvantitativt vurderede deltagerne prototypen positivt, og logdata viste, at størstedelen af de stillede spørgsmål var faktabaserede (76,32 %). De kvalitative fund pegede på, at deltagerne værdsatte den personlige og selektive informationsformidling, men også oplevede udfordringer såsom informationshuller, vanskeligheder ved spontant at formulere spørgsmål og ubehag ved at tale højt i det offentlige rum. Studiet peger dermed på potentialet for LLM-drevne virtuelle guider til at styrke engagement og læring i museer, men understreger samtidig designmæssige hensyn og behov for videre forskning. Resultaterne er eksplorative og skal ses som et fundament for fremtidig udvikling af mere tilgængelige og brugercentrerede løsninger.
[This apstract has been generated with the help of AI directly from the project full text]