Key points are not available for this paper at this time.
Advances in generative artificial intelligence (AI) are reshaping cultural production, yet questions remain about whether machine-authored texts can authentically represent marginalized lives. This study examines the capacity of OpenAI’s GPT-4o to represent the lived experiences of transfeminine individuals using a three-phase Hybrid Poetics Framework (HPF). Ten transfeminine participants were recruited through purposive maximum variation sampling. Semi-structured interviews and a participatory focus group (FGD-1) informed the creation of both human-generated poems (HGPs) and AI-generated poems (GPT-4o poems). In the reception phase, participants completed an authorship discrimination task and poetic quality ratings, followed by a structured poetry-reflection focus group (FGD-2). Results show that participants identified authorship above chance and rated HGPs higher for Emotional Quality, Atmosphere, and Structural Quality, while AI poems were praised for polish but perceived as emotionally distant. By contrast, creativity showed overlap across conditions. Findings foreground authenticity, affective fidelity, and ethical risks in AI-mediated representation. We propose bounded legibility and community co-governance as safeguards and introduce the Hybrid Poetics Framework (HPF) as a methodological approach for operationalizing participatory evaluation in Human Computer Interaction (HCI).
Dev et al. (Wed,) studied this question.