Toward a Mindful AI: Designing Introspective and Compassionate Language Models with Trúc Lâm Zen Principles

Abstract

This research introduces Mindfulness-Oriented, a mindfulness-centered language model inspired by the Zen Buddhist philosophy of Trần Nhân Tông [1] and the Trúc Lâm tradition of Vietnam. Moving beyond the pursuit of cognitive efficiency, this study explores how spiritual introspection, non-attachment and compassionate awareness can shape the design of artificial intelligence. Through a textual and philosophical analysis of the poem “The Song of Realization in the Mountain Stream ” (Vietnamese: “Đắc Thú Lâm Tuyền Thành Đạo Ca”), we developed a prototype dialogue dataset—ZenDialogue v1.0—featuring 1000+ ChatML-formatted mindful conversations across five core themes: suffering, anger, daily awareness, introspection, and meaning. A multi-layered architecture was proposed, integrating Zen prompt scaffolding, mindful response generation, and an evaluative framework (Mindfulness-Oriented Score) grounded in ethical and contemplative values. The findings suggest a new paradigm of conversational AI: one that does not merely provide answers, but invites the user into reflection, stillness, and inner clarity. It is a step toward cultivating AI that can truly listen.