From Tool to Partner: Exploring Common Patterns of Human-AI Collaboration Among PhD Students in Social Sciences

This study contributes new insights into how PhD students in the social sciences use AI tools in research, focusing on how they select and collaborate with AI to support and enhance their academic workflows. The authors conducted a survey among PhD students in eight countries (Italy, Norway, Belgium, the United Kingdom, Spain, Netherlands, France, and Estonia), combining quantitative descriptive analysis with qualitative critical discourse analysis. The findings reveal that most respondents began their PhDs after 2022, coinciding with the public availability of generative AI tools. ChatGPT emerged as the most frequently used, often alongside DeepL, Elicit, and SciSpace, with use cases tailored to individual research tasks. Although most participants used the free versions, some opted for paid models. The article identifies four typologies of AI agency—teaching agent, subordinate assistant, cognitive partner, and substitutive partner—reflecting different patterns of human-AI collaboration. AI is primarily used for academic writing, with background research and literature reviews as secondary uses. While originality was rated low, translation and writing support were seen as most beneficial. These findings underscore the versatility of AI and the evolving dynamics of co-produced knowledge in academia.