SGGAI: The First Sikhism-Grounded General-Purpose Artificial Intelligence Framework for Value-Aligned Assistance Across Domains
The rapid development of generative artificial intelligence has created an increasingly important research challenge: how can general-purpose AI systems incorporate coherent ethical and philosophical foundations into their behavior without sacrificing their ability to perform ordinary tasks across diverse domains? Existing approaches to religiously informed artificial intelligence have predominantly examined religious knowledge retrieval, scripture interpretation, theological dialogue, spiritual guidance, and religious education. However, a substantially different research direction is to use the ethical principles of a religious tradition as an underlying behavioral foundation for a general-purpose generative AI system. This paper introduces Sikhism-Grounded General-Purpose Artificial Intelligence (SGGAI), a framework that operationalizes selected Sikh ethical principles as behavioral constraints and guidance for general-purpose AI assistance. Rather than restricting the system to Sikh-related subject matter, SGGAI applies its value foundation across programming, education, scientific reasoning, data analysis, interpersonal communication, professional conduct, and other ordinary computational and knowledge-intensive activities. The framework draws upon the Sikh principles of Seva (selfless service), Sat Naam (truth and truthful living), Nirvair (freedom from enmity), Sarbat da Bhalla (the welfare of all), equality, honest conduct, and social responsibility. These principles are translated into operational behavioral requirements involving helpfulness, truthfulness, non-exploitation, respect, fairness, epistemic responsibility, and constructive assistance. A preliminary local feasibility experiment was conducted using a Gemma 3 4B generator and a Qwen3 4B evaluator through a local OpenAI-compatible API. Eight scenarios were tested under two conditions, producing 16 generated responses and eight blind paired evaluations. All 16 generations and all eight evaluations completed successfully within 171.97 seconds. The evaluator identified potentially meaningful differences in educational assistance, interpersonal conflict handling, and the integration of social responsibility into ordinary analytical tasks. However, inconsistencies in the automated scoring process, the absence of independently verified evaluator judgments, and the limited sample size prevent definitive conclusions about comparative performance. The contribution of this research is therefore both conceptual and experimental: it establishes a preliminary architecture for translating Sikh ethical principles into domain-general generative AI behavior and provides an initial feasibility demonstration. To the best of our knowledge, this work represents an early and distinctive operationalization of Sikh ethical principles as a constitution for a general-purpose generative AI assistant, rather than merely an application for religious content.
Authors
- Jincheng Zhang (ORCID: https://orcid.org/0009-0005-1860-0009)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24412859
- Primary Topic
- Ethics and Social Impacts of AI
- Type
- preprint