Fostering human–AI collaboration in robotic dance creation through large language models

This paper explores the potential of generative artificial intelligence (GenAI) in the artistic domain of robotic dance, by focusing on Human-AI co-creation. Specifically, we evaluate the capabilities of three state-of-the-art Large Language Models (LLMs) in generating robotic dance choreographies in collaboration with human artists. We design and test three prompting techniques tailored for robotic dance creation, in which we progressively introduce human expertise of choreographers in the form of examples and natural language feedback through an iterative process. In this way, we analyzed the dynamics of human-AI co-creation across different LLMs on varying of the parameters of the models and of different design of the prompts. The experimental analysis is conducted through a quantitative and qualitative evaluation, with the help of a human audience composed by 210 participants, and using a state-of-the-art evaluation scheme. The results revealed that each model shows different strengths and limitations. Moreover, when enriched by human intervention, the output generated from LLMs generally improve its artistic impact. These results highlight the potential of different LLM architectures in the artistic domain and in creative tasks. Ultimately, starting from the artistic domain of dance, this work contributes a modular methodology to enhance the intersection of human creativity and AI-driven creations, to develop frameworks for human-AI collaboration.