ChatGPT for cultural heritage and the customization of generative AI: A talkthrough analysis of the Luigi Einaudi chatbot

Cultural heritage institutions have recently experimented with the customization of generative AI and specifically large language models (LLMs) to create chatbots that impersonate historical characters. While this application has the potential to enhance user engagement, challenges remain, especially considering the need to ensure authenticity and historical accuracy and LLMs’ tendency to hallucinate. Through the analysis of the case study of a chatbot developed to impersonate the historical figure of Luigi Einaudi, the first elected president of the Italian Republic, the article interrogates opportunities, problems and risks raised by the customization of generative AI, in a moment when decisions about applying or not applying LLMs are being considered by many cultural heritage institutions around the world. The article, moreover, introduces a variant of the walkthrough method aimed to study AI conversational agents, called ‘talkthrough’, which represents an important addition to the methodological toolbox that can be activated to study generative AI.