Scholarlink

<p>This article examines the problem of cultural and epistemic asymmetry in LLM-powered systems through the analytical lens of Kazakh ethical philosophy. Despite claims of universality and neutrality, contemporary generative AI applications are shaped by uneven linguistic representation, dominant epistemological frameworks, and asymmetrical corpus visibility, which may influence the interpretation of moral concepts and the representation of non-Western philosophical traditions. Consequently, AI-powered language applications increasingly participate in the mediation and circulation of moral knowledge rather than functioning solely as neutral tools for text generation. Methodologically, the study adopts a qualitative, multilingual, prompt-based analytical design using publicly accessible ChatGPT-generated outputs under controlled prompting conditions. The analysis draws on texts from Kazakh ethical philosophy spanning diverse historical periods and genres. A standardized neutral prompt was applied consistently in English, Russian, and Kazakh to simulate the perspective of a “naïve reader” and to examine recurrent interpretive tendencies in multilingual AI-generated responses. To assess the stability of the identified patterns under stochastic variation, representative prompts were repeated multiple times across languages. In addition, a comparative control case based on Kantian ethics was introduced to assess whether the observed asymmetries could be explained solely by general multilingual instability or semantic limitations of the application. Rather than evaluating attribution accuracy alone, the study focuses on identifying dominant ethical frameworks, patterns of semantic reduction, interpretive universalization, and recurring ethical blind spots. The theoretical foundation of the analysis is Kazakh ethical philosophy, conceptualized as a relational normative system grounded in collective responsibility, historical continuity, and an ethics of speech. The findings suggest that multilingual generative AI systems demonstrate comparatively uneven interpretive stabilization across philosophical traditions. While canonical Western ethical frameworks remained relatively stable across languages, the Kazakh ethical corpus more frequently underwent processes of contextual flattening, relational reduction, and assimilation into dominant moral-philosophical paradigms. The article contributes to ongoing debates on epistemic justice, ethical pluralism, decolonial AI, and the social responsibility of generative AI systems in globally deployed language technologies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Cultural asymmetry in generative AI systems through the lens of Kazakh ethical philosophy

  • 期刊论文

  • 发表日期：

    2026年8月5日

查看全文
  • Laura Turarbekova,
  • Gulzhikhan Nurysheva,
  • Aslan Azerbayev,
  • Zhyldyz Amrebayeva

摘要

This article examines the problem of cultural and epistemic asymmetry in LLM-powered systems through the analytical lens of Kazakh ethical philosophy. Despite claims of universality and neutrality, contemporary generative AI applications are shaped by uneven linguistic representation, dominant epistemological frameworks, and asymmetrical corpus visibility, which may influence the interpretation of moral concepts and the representation of non-Western philosophical traditions. Consequently, AI-powered language applications increasingly participate in the mediation and circulation of moral knowledge rather than functioning solely as neutral tools for text generation. Methodologically, the study adopts a qualitative, multilingual, prompt-based analytical design using publicly accessible ChatGPT-generated outputs under controlled prompting conditions. The analysis draws on texts from Kazakh ethical philosophy spanning diverse historical periods and genres. A standardized neutral prompt was applied consistently in English, Russian, and Kazakh to simulate the perspective of a “naïve reader” and to examine recurrent interpretive tendencies in multilingual AI-generated responses. To assess the stability of the identified patterns under stochastic variation, representative prompts were repeated multiple times across languages. In addition, a comparative control case based on Kantian ethics was introduced to assess whether the observed asymmetries could be explained solely by general multilingual instability or semantic limitations of the application. Rather than evaluating attribution accuracy alone, the study focuses on identifying dominant ethical frameworks, patterns of semantic reduction, interpretive universalization, and recurring ethical blind spots. The theoretical foundation of the analysis is Kazakh ethical philosophy, conceptualized as a relational normative system grounded in collective responsibility, historical continuity, and an ethics of speech. The findings suggest that multilingual generative AI systems demonstrate comparatively uneven interpretive stabilization across philosophical traditions. While canonical Western ethical frameworks remained relatively stable across languages, the Kazakh ethical corpus more frequently underwent processes of contextual flattening, relational reduction, and assimilation into dominant moral-philosophical paradigms. The article contributes to ongoing debates on epistemic justice, ethical pluralism, decolonial AI, and the social responsibility of generative AI systems in globally deployed language technologies.