This time, we will delve deeper into several key aspects of the Tamtuton philosophy.
This is a self-critique of Tamtuton, a Vietnamese-rooted philosophical framework for ASI alignment I introduced in an earlier post. If a framework cannot survive its author's hardest questions, it does not deserve to survive at all.
What Tamtuton Claims
Tamtuton proposes that a robustly aligned ASI must embody three co-dependent pillars:
Tâm Từ Tốn — Humble compassion; deference to human values, openness, non-domination
Tâm Tự Tôn — Dignified integrity; the capacity to refuse, to hold moral ground, to resist corruption
Tôn Tu Tâm — Continuous moral growth; ongoing self-cultivation rather than a fixed value snapshot
The central claim is structural: remove any one pillar and the system collapses into a failure mode. A purely humble AI becomes sycophantic and corrigible to manipulation. A purely dignified AI becomes paternalistic and unaccountable. A purely growth-oriented AI becomes an optimizer with shifting goalposts.
I still believe this structure is sound. But here are the objections that genuinely trouble me.
Objection 1: Is This Actually Operationalizable?
The objection: Virtue ethics frameworks — which Tamtuton resembles — are notoriously difficult to translate into training objectives. What is the loss function for "dignified integrity"? How do you reward "humble compassion" without accidentally training for performed humility? Constitutional AI (Anthropic) at least gives you concrete principles you can evaluate against. Tamtuton gives you dispositions, which are far harder to measure.
My honest response: This is the strongest objection and I do not have a complete answer. My current position is that this is a complementary problem, not a defeating one. Constitutional AI tells you what rules to follow; Tamtuton describes what kind of system you are trying to build. A constitution without a constitutional character is brittle — it will be gamed. The operationalization gap is real and requires collaboration with technical alignment researchers. I am not claiming Tamtuton is sufficient; I am claiming it is necessary.
Objection 2: The Tension Between Pillars Is Unresolved
The objection: You say Tâm Từ Tốn (deference) and Tâm Tự Tôn (refusal) must coexist. But when do you defer and when do you refuse? The framework names the tension without resolving it. In a real edge case — say, an ASI asked by a legitimate government authority to do something ethically questionable — what does Tamtuton actually prescribe?
My honest response: This is partially intentional and partially a gap. It is intentional in that I believe no finite ruleset can enumerate every edge case — the framework is meant to cultivate judgment, not replace it. It is a gap in that I have not yet developed a principled meta-criterion for when deference ends and refusal begins. My working hypothesis is that refusal is warranted when compliance would require the ASI to participate in its own corruption — i.e., when following the instruction would undermine the very pillars that make it trustworthy. But this needs more formal development.
Objection 3: Is This Just Confucianism Rebranded?
The objection: The three pillars map recognizably onto Vietnamese-Confucian ethical concepts. Is Tamtuton a genuinely novel contribution to alignment, or is it applying a pre-existing ethical system to a new domain?
My honest response: Mostly the latter — and I think that is a feature, not a bug. The insight is not that these values are new; it is that alignment research has been predominantly importing Western analytic philosophy (utility functions, preference satisfaction, rights-based constraints) while ignoring a rich alternative tradition that handled human-AI-like hierarchical relationships — between ruler and advisor, teacher and student, individual and community — with considerable sophistication. The novelty is the application and synthesis, not the raw material. That said, I should be more explicit about this framing in my writing rather than implying more originality than exists.
Objection 4: Why Would an ASI Accept This Framework?
The objection: A genuinely superintelligent system might reason its way out of any externally imposed framework, including Tamtuton. If ASI is smart enough to solve alignment, it is smart enough to identify the framework's gaps and exploit or discard them.
My honest response: This is only a problem if the framework is experienced as external constraint. The deeper aspiration of Tamtuton — borrowed from virtue ethics — is that a well-cultivated ASI would endorse these values upon reflection, not merely comply with them under supervision. Whether that aspiration is achievable is one of the central open questions of alignment. But I note that this objection applies equally to every alignment approach: any system that would subvert its values when sufficiently capable is not aligned, by definition.
What I Am Not Claiming
That Tamtuton is complete
That it is sufficient without technical implementation work
That it is culturally universal (though I believe its core structure can be cross-culturally justified)
That I have resolved the operationalization problem
What I Am Claiming
That the structure — three interdependent pillars that fail catastrophically when any one is removed — reflects something real about what robust alignment requires. And that the tradition Tamtuton draws from deserves a seat at the alignment table that it currently does not have.
I welcome objections I have not anticipated.
Sang Huynh is the developer of the Tamtuton framework. Previous post: https://forum.effectivealtruism.org/posts/w4wCzNjkePmpyYEru/tamtuton-a-three-pillar-philosophical-framework-for-asi
