A mixed methods analysis of artificial intelligence ethics discourse evolution in the generative era using bibliometrix and BERTopic
TL;DR
The findings reveal that 2023 represents a pivotal turning point in generative AI ethics research ; by 2025, the human-centered theme of ‘education and cognitive literacy’ had surpassed the long-dominant topic of ‘law and governmental regulation’ in publication volume.
Abstract
The development of generative artificial intelligence (AI) has introduced novel risks, including deepfakes and algorithmic hallucinations, urgently demanding a fundamental shift in global AI ethics from theoretical presuppositions toward actionable governance practices. By uncovering the developmental trajectory of ethical discourse in the era of generative AI, this study conducts a systematic analysis of the existing literature, aiming to generate actionable insights for future interdisciplinary research and contribute to the reconstruction of a renewed ethical order in this field. To this end, this study constructs a mixed-methods framework integrating macro-level bibliometrics with the micro-level BERTopic deep semantic mining approach. Following a structured multi-stage screening protocol, this paper performs a quantitative analysis of 1,190 core documents (2020–2025) retrieved from the Web of Science and Scopus databases, identifying 10 core themes and examining their spatio-temporal evolution.The findings reveal that 2023 represents a pivotal turning point in generative AI ethics research ; by 2025, the human-centered theme of ‘education and cognitive literacy’ had surpassed the long-dominant topic of ‘law and governmental regulation’ in publication volume. This shift suggests a reorientation in global governance discourse from a defensive emphasis on ‘technical compliance’ toward an adaptive focus on ‘human-centered cognitive empowerment’. Furthermore, global knowledge production exhibits a pronounced “center-periphery” structure, where a small group of developed countries dominates agenda-setting, while the Global South remains in a marginal position ; meanwhile, the academic literature tends to cluster into two distinguishable discursive orientations: one centered on ‘technical governance’ and the other on ‘humanistic reflection’. This study calls for bridging the binary epistemological divide by integrating humanistic values into technological instrumental rationality, while broadening the current discourse framework and addressing structural imbalances through more inclusive North-South collaboration, with the ultimate goal of advancing global ‘epistemic justice’ in the era of generative AI.