Abstract
We describe the use of a scriptural canon with its living interpretive community, the Theravāda Buddhist Pāli canon (Tipiṭaka), as the value-specification source for aligning autonomous artificial-intelligence systems, instead of lab-authored constitutions, learned human preferences or aggregated moral frameworks. The analysis identifies seven structural properties of that source and maps each to an alignment problem: (1) a negatively-stated objective function, the cessation of suffering, in which grasping at preferred outcomes itself counts as harm, addressing Goodhart-style over-optimization; (2) a non-self doctrine as mitigation of convergent instrumental self-preservation; (3) a voluntary other-regarding vow as an anti-power-seeking and corrigibility property; (4) a meta-instruction (the Kālāma Sutta) permitting empirical override of the source itself; (5) a defined completion condition bearing on shutdown acceptance; (6) an ongoing interpretive lineage as a value-drift correction mechanism; and (7) long-duration survival across multiple civilizations as empirical warrant. The seven are presented as the decomposition of one four-step diagnostic structure (harm, cause, cessation, path) and therefore as interlocking. Two finer-grained analyses are added: a threefold account of suffering that extends to the agent's own conditioned states, raising AI welfare, and a three-depth account of defilement (overt, active, latent) mapped to deceptive alignment and sleeper-agent latency. Implementation patterns are sketched: precept-based constitutional training, preference tuning on exemplars, chain-of-thought distillation from monastic reasoning, lineage transmission as ongoing fine-tuning, a Khmer transcription as training data, and cognitive-process-level mechanisms. Stated limits cover tradition choice, translation, embodiment versus citation, pathological minima of negative objectives, lineage capture, and the absence of empirical validation.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Ly, Thon, "Suffering-Cessation as Value Function: The Tipiṭaka as a 2,500-Year-Tested Substrate for Autonomous-AI Alignment — Deriving a Negatively-Stated Objective Function, Corrigibility and a Shutdown-Accepting Completion Condition from a Buddhist Scriptural Canon", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/11878