2026-08-31 ワシントン大学（UW）
＜関連情報＞
- https://www.washington.edu/news/2026/08/31/august-research-highlights-nectar-robbing-anxious-attachment-styles-persnickety-plasma-more/
- https://dl.acm.org/doi/10.1145/3805689.3812418
添付ファイルのスタイルはChatGPTの利用状況に影響を与えるのか？ Do Attachment Styles Shape ChatGPT Usage?
Marx Wang, Jade Xiaoyi Li, Songling Ngo, Katie Davis, Alexis Hiniker
FAccT ’26: The 2026 ACM Conference on Fairness, Accountability, and Transparency
DOI:Published: 25 June 2026
DOI:https://doi.org/10.1145/3805689.3812418
Abstract
The widespread adoption of general-purpose conversational agents (GPCAs) has raised questions about the relationships that users may be developing with these systems. In this study, we ask whether users’ attachment styles predict how they interact with ChatGPT, a popular GPCA, and how they experience these interactions. We conducted a mixed-methods study, triangulating self-reported survey data (N = 168) with transcripts of users’ ChatGPT conversational history (N = 19, 930 messages). We find that attachment anxiety predicts emotional engagement with ChatGPT, trust in ChatGPT, and likelihood of adopting behavioral suggestions from ChatGPT, while attachment avoidance predicts reduced trust in ChatGPT and reduced likelihood of reporting that ChatGPT improves users’ self-efficacy. Furthermore, we find that attachment anxiety is associated with distinctive linguistic patterns in users’ ChatGPT conversations, including increases in affect words, self-referential pronouns, and future-focused thinking. These findings suggest novel privacy risks, such as the risk of implicitly disclosing attachment style through conversational patterns, even when conversing with a non-intimate GPCA for banal purposes like homework help and information-seeking. These findings also point to implications for designers and policy-makers, including the need to recognize anxiously attached individuals as a vulnerable user group and the potential future need to regulate psychological profiling of users.
