Power, Dan and Farley, Paul (2026) The Artificial Muse : Human-authored poetry from AI-generated text. PhD thesis, Lancaster University.
![[thumbnail of 2026Dan Power PhD]](https://eprints.lancs.ac.uk/style/images/fileicons/text.png)
Abstract
This project seeks to answer one question: how might GenAI tools (such as ChatGPT) be utilised effectively by poets, without those poets sacrificing their integrity or authorship? Its answer is simple – by treating the GenAI tool as an artificial muse, a provider of inspiration, but never as a creator in its own right. The shortcomings of AI-generated poems arise from the GenAI’s inability to write intentionally, and subsequent inability to form authorship. All GenAI text is unoriginal, inauthentic, insincere, and otherwise lacking in panache. As this thesis and portfolio demonstrate, these inherent problems can be overcome by re-constructing AI-generated drafts to the extent that human intention is enacted upon them, and human authorship is wrought from them. The thesis consists of two parts – Part One outlines why AI-generated poems are “authorless”, why this makes them unoriginal and inauthentic, and how a lack of human authorship prevents readers from connecting to them. It also considers how authorless texts can be reclaimed by human authors, and made authentic, original, and engaging. Part One concludes by stating that, for poets to use GenAI tools effectively and ethically, without sacrificing their authorship, the relationship between the user and the GenAI must resemble that between a poet and their muse – the muse provides inspiration, kick-starting the process of a poem’s assembly, but the inspiration belongs to the poet, who transforms it into poetry according to their own intentions. The thesis considers what a muse is, how poets and muses interact, and how a GenAI muse might offer “inspiration without constraint”. Part Two recounts how the conclusions of Part One were enacted during the development of my portfolio, each poem of which was derived from a process of “musing” with AI-generated material. It explains how I developed a repeatable method of “pruning” and “regrowing” AI-generated drafts in order to extract inspiration from them, and how I made original assemblages of words from unoriginal materials. It also considers whether musing with GenAI made my poems more novel, whether this prevented my own intentions being enacted, and how the increased productivity facilitated by the GenAI muse impacted the quality of my work, and the enjoyability of the writing process. The thesis concludes that GenAIs can be effective tools to muse with – poets can take inspiration from GenAI without sacrificing their individuality, authorship, or integrity, as long as the GenAI is put away after inspiration has been achieved. However, the necessity of keeping the GenAI tool at arms-length, and of meticulously re-writing its outputs, means that musing with GenAI can require more work from human poets than more traditional methods of musing – something which curtails its usefulness as a shortcut to inspiration. Finally, the thesis warns that writers wishing to benefit from AI-given inspiration should do so sparingly: otherwise they risk losing touch with their humanity, and depleting their ability to receive and transform inspiration organically.