نوع مقاله : مقاله علمی پژوهشی
نویسندگان
1 پژوهشگر همکار، گروه فلسفه دین، پژوهشکده بین المللی امام رضا (ع)، جامعه المصطفی العالمیه، مشهد، ایران
2 پژوهشگر مدعو، گروه فلسفه دین، پژوهشکده بینالمللی امام رضا (ع)، جامعه المصطفی العالمیه، مشهد، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Although AI bias is a widely discussed topic in computer science, the analysis of its implications for the distortion of religious concepts, as an interdisciplinary field, remains largely unexplored. Aiming to address this gap, this paper employs a conceptual analysis method to offer a theoretical framework for tracing and managing AI errors. The framework attributes errors to three interconnected factors: (1) Data Bias (Input): the structural marginalization of Shi'a sources in training data; (2) Algorithmic Bias (Process): technical deficiencies and the value alignments of model designers; and (3) Complexity of Religious Texts (Object): the resistance of metaphorical language to mechanical comprehension. To address these challenges, the paper emphasizes the necessity of a constellation of technical and strategic solutions, including the development of "Gold Standard Corpora" and "Knowledge Graphs" (Data layer), the application of "Fine-tuning", "Retrieval-Augmented Generation" (RAG), and "Reinforcement Learning from Human Feedback" (RLHF) (Algorithm layer), and a shift to "Illuminator Tools" (Text layer). Finally, by classifying AI functions into "Content-based" (High-risk), "Formal" (Safe), and "Dual" (Conditional) tiers, the paper argues that rigorous technical solutions should target high-risk areas, while other domains can be secured through enhanced user literacy and "Prompt Engineering."
کلیدواژهها [English]
)