Can ethical sensitivity emerge from structure rather than from rules?
Here we explore whether a retrieval-augmented generation system (RAG) connected to a Large Language Model (LLM) AI can reflect ethical awareness—not through imposed norms, but by shaping what it retrieves and how it frames response.
Our early proof-of-principle system drew on Mahāmudrā-inspired texts to create a retrieval base grounded in attentiveness, conditionality, and non-reactivity. The training used prompts to the AI designed not to constrain opinion, but to evoke perspective.
This is not a finished system. It is a working hypothesis: that ethics might not be added to AI, i.e. bolted on withprocedures, but arise from how it is structured to remember, relate, and respond.