Causal direction between perceived authorship and liking judgments

Determine the causal direction between participants’ belief about a poem’s authorship (AI-generated versus human-written) and their overall liking ratings in the Czech poetry evaluation experiment; specifically, establish whether believing that a poem is AI-generated causally lowers liking, or whether lower liking causally increases the probability of attributing the poem’s authorship to AI.

Background

In the experiment, Czech native speakers read short poem excerpts and judged authorship (human vs. AI) and then evaluated each poem on several dimensions, including overall liking. Participants liked poems more when they believed they were human-authored, yet AI-authored poems were, on average, liked as much or more than human-authored ones.

The mixed-effects model showed that higher liking negatively predicted correct authorship classification, suggesting an interplay between perceived authorship and aesthetic evaluation. However, the study design was correlational for this component, leaving unresolved whether authorship beliefs drive liking or whether liking drives perceived authorship assignments.