UC Santa Cruz

The novel computational abilities offered by language models and generative AI suggestthe potential for new creative practices. But so far, they are being used to replace, not enhance, human creativity. Creative workers of all kinds are being threatened and displaced even as AI output is increasingly viewed with contempt. But what about that initial promise? Can we instead build creative experiences with language models that respect human creative labor and agency? In this thesis, I explain the design of three experimental systems that demonstrate alternative pathways for language models to provide creative experiences. In the first, language models provide dynamic reference material for a game master during a tabletop roleplaying game. In the second, language models power an experimental writing tool that doesn’t generate words for the author but rather provokes reflection. In the third, language drives a novel game prototype where manipulating AI agents becomes a creative surface. Taken together, these projects point toward uses of language models that are positive and constructive rather than harmful.