arXiv cs.CY· Jacqueline Rowe, Animesh Srivastava, Sai Teja Peddinti, Seliem El-Sayed, Nina Taft·· 10 天前AI 评分25
用社交媒体评论追踪不断演变的 AI 危害：LLM 辅助主题分析管线
Move Fast and Mend Things: Keeping Up with Evolving AI Harms Using Social Media Commentary
AI 导读
研究者提出一套 LLM 辅助主题分析管线，从大规模社交媒体数据中动态检测、分类并追踪新兴 AI 危害。基于 18 个月（2025 年 1 月至 2026 年 6 月）570 万条 Reddit 帖子摘要，团队构建并发布含 57.5 万条 AI 危害相关帖子的数据集，以及涵盖 12 个类别、47 个子节点的自下而上 AI 危害分类体系。
正文 · 原文
Abstract:The rapid deployment of AI systems has created socio-technical, psychological, and operational harms that can elude ex-ante threat modelling and ex-post incident tracking. We introduce an LLM-assisted thematic analysis pipeline to dynamically detect, categorise, and track emerging AI harms from large-scale social media data. Applying it to 5.7 million Reddit post summaries over 18 months (01/2025 to 06/2026), we curate and release a dataset of 575,000 AI harm-related posts and a bottom-up AI harm taxonomy of 12 categories and 47 subnodes. The taxonomy reliably covers established expert-defined risks while surfacing granular harms that top-down frameworks overlook, such as distinct forms of AI privacy violations. Temporal analysis surfaces evolving user-centric harms, such as agentic privacy and security breaches, premature AI adoption in the workplace, and grief from AI companion discontinuation. Our pipeline shortens harm-detection timelines and hereby complements efforts towards more participatory and responsive AI governance.
| Subjects: | Human-Computer Interaction (cs.HC); Computers and Society (cs.CY) |
| Cite as: | arXiv:2610.09082 [cs.HC] |
| (or arXiv:2610.09082v1 [cs.HC] for this version) | |
| https://doi.org/10.48550/arXiv.2610.09082 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Seliem El-Sayed [view email]
[v1]
Tue, 6 Oct 2026 20:24:10 UTC (226 KB)
来源：arXiv cs.CY · arxiv.org