Living with Risk: Folk Theories of AI Hallucinations Among Young Chinese Users

As generative artificial intelligence (GenAI) becomes increasingly integrated into everyday life, AI hallucinations have emerged as a significant challenge in human-AI interaction. Drawing on the perspective of algorithmic folk theories, this study combines textual analysis and in-depth interviews with both ordinary users and technical users to examine how they understand and respond to AI hallucinations. Users identify AI hallucinations by scrutinizing the AI’s reasoning process, making intuitive judgments, and conducting cross-validation, while their attitudes toward hallucinations vary. Users develop four types of algorithmic folk theories: the Human-AI Interaction Misalignment Theory, the Technical Limitations Theory, the Data Source Distortion Theory, and the Generative Rule Constraint Theory. They rise to four corresponding coping strategies: reducing communication errors, reconfiguring human-AI task allocation, intervening in data input while verifying outputs, and reducing AI’s processing load. This study further indicates how technological identity and sociotechnical contexts jointly shape algorithmic folk theories.