Examining trends in AI ethics across countries and institutions via quantitative discourse analysis

Article / Journal

Author(s) / editor(s):
Oshri Bar gil

Year: 2025

Bar-Gil, O. Examining trends in AI ethics across countries and institutions via quantitative discourse analysis. AI & Soc (2025). https://doi.org/10.1007/s00146-025-02673-4

Keywords: Quantitative discourse analysis; AI ethics; Code of ethics; Responsible innovation; AI governance
Language(s): English

Abstract:
This study examines how institutional contexts influence the AI ethics landscape through quantified qualitative discourse analysis. We analyzed ten foundational AI ethics documents from academic, industry, military/defense, and national sectors (2018–2021) to investigate whether purportedly universal ethical principles maintain consistent meanings across contexts. The methodology integrated computational frequency analysis of purposive sample, targeting influential texts functioning as obligatory passage points in AI ethics discourse. We identified 14 ethical principles through systematic word list development, analyzed 2351 coded segments across documents, and mapped semantic co-occurrence patterns. The analysis revealed that universal principles undergo systematic recontextualization through institutional appropriation. Privacy transforms from rights-based frameworks in EU documents to security-balance approaches in US military contexts to collective security conceptualizations in Israeli frameworks. Military frameworks uniquely emphasize governability and traceability, while these principles remain absent in academic and industry documents. Industry texts prioritize technical operationalization over distributive justice concerns, with equitability completely absent despite appearing in 15.8% of academic codes. These findings challenge assumptions about universal AI ethics, demonstrating that institutional logics constitute rather than merely influence ethical discourse. Effective AI governance requires acknowledging this contextual heterogeneity through sector-specific guidelines that recognize interpretive variations while maintaining core principles. The research contributes methodologically by demonstrating how quantified qualitative analysis reveals systematic patterns invisible to purely qualitative approaches, and theoretically by establishing institutional positioning as constitutive of ethical meaning in AI governance discourse.

https://link.springer.com/article/10.1007/s00146-025-02673-4

Post created by: Oshri Bar gil