Text comprehension moderates cognitive biases toward AI authorship in literary content

The growing potential of artificial intelligence (AI) in generating artistic products has stimulated an increasing multidisciplinary debate. Beyond discussions on the nature of AI creativity itself, numerous studies report that people are negatively biased towards AI-labelled artistic contents. Here we asked whether this bias is modulated by beholders’ approach to the artifact, by exploring whether and how the level of comprehension—an index of participant’s engagement—could influence this bias in literary context. We asked participants to rate the social prestige and pleasantness of three texts from a reckon writer that were presented as Human-, AI- and Human + AI-made. We discovered that the comprehension level emerges as a key, bidirectional modulator of evaluative judgments. Indeed, we observed that only participants that resulted in a low level of text comprehension showed a negative bias against AI-labelled texts compared to Human- and Human + AI-labelled ones, while those that demonstrated high comprehension showed a positive bias towards AI. Our results suggest that bias against AI is not intrinsic to the content but is driven by preconceptions that can be mitigated or exacerbated by the level of understanding achieved.