نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشجوی دکتری موسسه آموزشی و پژوهشی امام خمینی قم، ایران
2 عضو هیئت علمی و مدیر دانشکده کامپیوتر دانشگاه علم و صنعت تهران، ایران
3 عضو هیئت علمی موسسه امام خمینی و مدیر گروه اخلاق موسسه امام خمینی
چکیده
کلیدواژهها
عنوان مقاله [English]
نویسندگان [English]
The emergence of large language models has created new possibilities for AI-powered Assistants for Self-evaluation, yet assessing abstract moral concepts remains challenging. This paper aims to identify and analyze the technical and conceptual challenges of measuring human morality with artificial intelligence and to propose theoretical strategies to address them. Using a qualitative, descriptive-analytical, and documentary approach, the study systematically extracts and classifies challenges through a thematic analysis of the specialist literature. Findings are presented within a four-part analytical framework: (1) conceptual-to-computational translation challenges, such as operationalizing intent and the absence of formal ethical ontologies; (2) data-driven challenges, including reliance on digital traces, a shortage of standardized datasets, and cultural bias; (3) algorithmic-logical challenges, such as model instability, the black-box problem, and sensitivity to phrasing; and (4) interactive and dynamic challenges, encompassing dysfunctional feedback loops and temporal attribution errors. The study concludes that overcoming these barriers requires moving beyond purely statistical approaches. Recommended strategies include developing multidimensional measurement profiles, designing hybrid architectures that integrate explainable AI, causal inference, and formal ontologies grounded in Islamic ethical principles, and shifting from passive observation to active, user-centered self-assessment.
کلیدواژهها [English]
)