Abstract
As Artificial Intelligence (AI) I is increasingly becoming an inherent part of a large number of disciplines, including the social sciences, researchers are frequently relying on cautious language in their evaluation of its role and impact. This study focuses on how this uncertainty is expressed using hedging devices in recent research papers discussing the integration of AI into social science and education. Relying on Salager-Meyer’s taxonomy of hedges and Hurst & Johnston’s definitional approach to the framework of social impact, this study analyzes conclusion sections of ten research papers published in 2025 in top-tier Western journals. Using a qualitative, interpretive approach, this study discusses the types, frequency, and contextual nature of hedging devices, with special attention paid to their co-occurrence with social impact-expressing phrases. The findings show that hedging is a consistently used feature of this research discourse, which is most commonly expressed through the use of modal verbs, probability markers, compound hedges, and implicit shields. Together with social impact-related phrases, these hedges are used to frame AI in this field as a socially significant phenomenon that is still largely context-dependent. This study contributes to ongoing research on academic discourse because it shows how developing technologies help construct evaluative language in fields like social science and education. Furthermore, this study offers insights for academic writing instruction and other studies focusing on AI using discourse analysis frameworks.