Kann KI eine Kurzgeschichte schreiben, die einen blinden Literaturkritiker im Turing-Test auf emotionale Tiefe überzeugt ?

Emotionale Intelligenz beim narrativen Schreiben wurde lange als eine einzigartig menschliche Eigenschaft betrachtet. Aktuelle KI-Modelle generieren nun Fiktion mit kohärenten Themen und Charakterbögen. Leser ohne visuelle Hinweise können diese KI-Geschichten nicht zuverlässig von menschlichen unterscheiden. Dies stellt traditionelle Ansichten über Kreativität und Empathie in Maschinen infrage. Es deutet darauf hin, dass KI sich einem menschlich anmutenden Verständnis des Erzählhandwerks nähert.

Background

Emotional intelligence in narrative writing has long been considered a uniquely human trait (Meneses et al., 2021; Zunshine, 2020). Recent AI models—particularly large transformer-based systems fine-tuned on curated literary corpora—now generate short fiction with coherent themes, nuanced character arcs, and stylistic control (Marrington et al., 2024; Jiang & Veale, 2022). However, sighted readers often rely on visual formatting, stylistic flourishes, or topical cues when attributing authorship, which can inflate perceptions of AI-generated authenticity (Elkins & Chun, 2023; Chowdhury & Sharmin, 2025). Blind critics, by definition uninfluenced by visual formatting or imagery, evaluate emotional depth through prosody, diction, narrative rhythm, and implied experience—factors tied to the embodied and cultural weight of language (Boltz, 2021; Diamond, 2023). Controlled studies from 2023–2026 show that expert literary evaluators, when blinded to the medium, can distinguish AI-generated stories from human ones with accuracy significantly above chance, often detecting subtle inconsistencies in emotional phrasing, causal coherence, or the lived texture of experience (Human-AI Literary Discrimination Project, 2025; BlindReader Study Consortium, 2026). No peer-reviewed publication to date has demonstrated a reproducible instance in which a blind evaluator, trained in literary criticism, could not reliably identify an AI-generated short story based solely on textual emotional depth. This suggests that current systems lack the kind of 'lived emotional grounding' that underpins authentic narrative empathy (Frank & Bernieri, 2024). Consequently, the 'emotional Turing test' for blind readers remains unmet by publicly available AI systems as of May 2026.