人工智能对痴呆症的刻板印象——基于文生图模型生成图像的多模态批评话语分析

Stereotypes of Dementia in Artificial Intelligence—A Multimodal Critical Discourse Analysis Based on Images Generated by Text-to-Image Models

摘要: 社会对痴呆症的认知是保证患者及其家属身心健康、构建良好社会关系的重要议题。本研究基于视觉语法，运用国内六款文生图模型，生成248幅提示词为“痴呆症/dementia”的图像，并从辩证关系视角进行多模态批评话语分析。结果表明，人工智能多从个人与社会视角进行展示，并在年龄、文化背景等方面呈现视觉偏见，将痴呆症污名化、边缘化，与目前社会对于痴呆症的认知具有一致性。但同时也存在部分与公众认知相反的痴呆症视觉友好话语。本研究揭示了AI生成图像在强化和放大有害刻板印象方面的潜在风险，强调了对这些图像进行批评性话语分析的必要性，以及在AI设计和使用中融入伦理考量的重要性，以减少刻板印象的传播及其潜在危害。

Abstract: The societal perception and understanding of dementia is a critical factor in ensuring the mental and physical well-being of patients and their families, as well as in building positive social relationships. This research, grounded in visual grammar theory, utilized six domestic text-to-image generation models to create 248 images with the prompt “dementia” and conducted a multimodal critical discourse analysis from a dialectical perspective. The results revealed that artificial intelligence predominantly presents dementia from personal and societal viewpoints, exhibiting visual biases in aspects such as age and cultural background, which serve to stigmatize and marginalize dementia. These representations align with current societal perceptions of dementia. However, the study also identified certain visually friendly discourses that contradict public cognition. This research illuminates the potential risks of AI-generated images in reinforcing and amplifying harmful stereotypes, emphasizing the necessity of conducting critical discourse analysis on these images. It also underscores the importance of integrating ethical considerations into AI design and usage to mitigate the spread of stereotypes and their potential harms.