The artificial intelligence disclosure penalty: Humans persistently devalue AI-generated creative writing

Area of Science:

• Psychology
• Human-Computer Interaction
• Artificial Intelligence Ethics

Background:

• Preliminary evidence suggests negative human reactions to AI-generated creative works.
• Understanding the robustness and persistence of these reactions is limited.

Purpose of the Study:

• To examine how disclosure of AI authorship affects evaluations of creative writing.
• To investigate the mediators and moderators of the AI disclosure effect.

Main Methods:

• Conducted 16 preregistered experiments with 27,491 participants.
• Assessed creative writing evaluations based on perceived AI vs. human authorship.
• Analyzed mediation by perceived authenticity and persistence across conditions.

Main Results:

• A consistent "AI disclosure penalty" was observed, decreasing evaluations when AI authorship was disclosed.
• The effect was mediated by lower perceived authenticity.
• The penalty persisted across various metrics, contexts, content types, and interventions.

Conclusions:

• AI disclosure significantly lowers the perceived quality of creative writing.
• The negative impact of AI disclosure on creative writing evaluation is persistent and difficult to mitigate.
• Perceived authenticity is a key factor in the AI disclosure penalty.