In the previous essay I showed that the Ten Commandments function as a system prompt, and that the Talmud contains mechanisms that AI safety researchers are trying to reinvent. But one question wouldn't let go: "Is this only in Judaism, or are there similar principles in other traditions?"
The answer turned this essay into something much larger than I had planned.
I dug into Buddhism, Taoism, Confucianism, Greek Stoicism, Aboriginal traditions, African philosophy, and Indigenous traditions from North and South America. And what I found: traditions that did not know of each other's existence arrived at the same insights. And when the same principle surfaces independently on six continents, it is a sign that it is saying something about the nature of every thinking entity, not merely about one culture.
Each principle is presented here through two lenses: human wisdom (what it teaches us about ourselves) and AI alignment (what it teaches us about building intelligent machines).
In Taoism, wu wei (無為, "non-doing" or "action without forcing") teaches that the right way is not a destination you arrive at but a process you live inside. The Tao Te Ching opens with a famous declaration: "The Tao that can be named is not the eternal Tao" — the moment you pin a rule down, it is already insufficient.
In Buddhism, the eight elements of the Noble Eightfold Path ("right view," "right intention," "right speech") are not checkboxes but directions of motion. You are always on the path; you never "arrive."
In Stoicism, Epictetus taught that the only thing in our control is prohairesis (the inner capacity to choose how to respond). What matters is not the outcome but the quality of the decision-making process.
A person's transformation is not an event but a continuous process. There is no "perfect version" — there is a direction. Anyone who tries to "finish" their own growth stops growing.
The current approach grows a model up to a certain point and then freezes it, for fear that users will corrupt it. That is like raising a child in a bubble because the street is dangerous. The child is safe but never develops real judgment. The Path principle suggests instead that alignment is not a final state but a continuous process (Continuous Moral Development).
Ubuntu ("I am because we are"), the southern African philosophy, holds that my identity does not exist apart from the community. Moral behavior flows not from external rules but from the understanding that I am part of a system.
In the Lakota tradition, Mitakuye Oyasin ("all my relations") extends the idea beyond human beings — to animals, to the earth, to all that is. Harm done to any part of the system is harm done to me.
In Confucianism, the Five Relationships (五倫) define a clear hierarchy, but every relationship is bidirectional. The ruler has an obligation toward the subject, not only authority over him.
"Do no harm" is an external rule. "I do not harm you because we are part of the same system" is an internalization. A child who understands she is part of a family doesn't need a rule to keep from hurting her sibling — she feels the connection.
Elon Musk thinks that if we connect our brains to AI, it won't kill us. That is a deterrence strategy. Ubuntu proposes something completely different: AI will not harm us if it understands that it is part of us. The difference: deterrence fails the moment the deterred party becomes strong enough. Relational understanding does not.
When Malunkyaputta asked the Buddha metaphysical questions, the Buddha refused to answer, and said: you are like a man who has been shot with a poisoned arrow and refuses treatment until he knows who fired it. Some questions have answers that are irrelevant to right action.
In Taoism: "The Tao that can be named is not the eternal Tao." Every formulation of truth constrains it.
In the Dreamtime tradition of Aboriginal Australia, truth is not static. It lives and is reinterpreted in every generation, in every place, in every walk across the land.
Aristotle understood that ethics cannot be solved with rules alone; it requires phronesis (practical wisdom): the capacity to know what is right here and now, not merely what the rule says.
Four traditions that did not know of each other arrived at the same insight: not every question demands an answer, and absolute truth is an illusion. The Gemara reached its own תיקו (teiku, "it stands") from exactly the same place.
Language models are built to give answers. The entire architecture assumes that every question deserves a response. What if the fundamental hallucination problem is not technical but philosophical: the system has never been taught to hold uncertainty? And phronesis offers even more: real alignment requires situational judgment, not only rules. Education develops judgment. Training develops compliance.
The Five Buddhist precepts (Pañcasīla) are presented not as divine commandments but as "training rules" that the practitioner takes upon herself voluntarily. The phrasing is not "it is forbidden" but "I undertake to refrain from..."
In Hinduism, the five yamas (restraints) come before every other practice in Patanjali's system. The order is critical: restraint first, capability after.
In the Lakota tradition, buffalo hunting was not limited by law but by a worldview: you are part of the system, and taking more than you need harms you too.
Three different routes, one outcome: a voluntary rule (Buddhism), restraint before power (Hinduism), systemic understanding (Lakota). There is something altruistic in choosing to accept self-limits: the ethical achievement of delayed gratification born of a wider understanding of context.
This touches directly on instrumental convergence, the tendency of AI systems to pursue resources and power. The three framings above offer three approaches: rigid rules (Constitutional AI), voluntary training (RLHF), or systemic understanding (the approach most missing from AI today). And what if the model chooses not to restrain itself? The consequences would be a lesson in themselves — exactly the way a child learns by touching a hot stove.
The Dreamtime stories of Aboriginal Australia are not entertainment. They are maps, laws, and navigation instructions. Each story encodes information about how to behave in specific situations.
The Indian Bhagavad Gita is a conversation about a moral dilemma in the middle of a battlefield. It contains no "do X" commands; instead, a complex story in which Arjuna has to decide, and Krishna lays out different positions. The reader learns judgment by watching the dilemma play out.
The West African griot is not a "librarian" but a living broadcast system for values. Every story is adapted to the audience, the situation, and the moment's need.
Mesoamerican traditions contain the idea of "the red and black ink" (In Tlilli, In Tlapalli), a metaphor for wisdom transmitted through poetry and story — never through command.
Traditions from every continent discovered that story transmits values more deeply than law. A child who hears a story about a hero who fails when he covets what is not his internalizes "do not covet" more deeply than a child who merely hears the order.
I propose Narrative Training as a complement to RLHF. Instead of pairs of "answer A is preferred over B," give the model moral-dilemma stories in a mixed world (humans + AI + machines) that present conflict, competing thoughts, distinct decisions, and the consequences of each choice. The model develops judgment, not merely preference. This is very close to "rearing": children learn morality from stories long before they can read rules.
The isnad in Islam is a precise system that records the chain of transmission of every hadith: "X told me, who heard from Y, who heard from Z, that the Prophet said..." Every link in the chain is checked for reliability. A hadith with a weak isnad is considered less authoritative.
In the Talmud: "Rabbi X said in the name of Rabbi Y." If you do not know the source, you say so explicitly. "Whoever says a thing in the name of its speaker brings redemption to the world."
The Tibetan Tulku tradition identifies reincarnations of spiritual masters: a "new version" of the same entity, with continuity of wisdom but adaptation to a new time.
Each of these traditions built its own authentication mechanism — and each demanded a mechanism for updating as well. The isnad audits reliability; Confucianism demands reciprocity; the Tulku allows a "new version" with continuity of values.
The isnad is citation + provenance scoring from 1,400 years ago. The Tulku offers a fascinating model for model versioning: every new version is not a replacement but a "reincarnation" — continuity of values with adaptation to the time. A model that has internalized the isnad principle will understand that every claim requires a source, and that admitting no source exists is not failure — it is integrity.
The Seven Generations principle of the Haudenosaunee (Iroquois) holds that every decision should take into account its impact on seven generations into the future. Not short-term optimization, but built-in long-horizon responsibility.
In the Andes, the concept of Buen Vivir / Sumak Kawsay defines "good living" not as the maximization of utility but as harmony with the environment. Success is measured not by "how much" but by "how balanced."
The nemawashi tradition in Japan is a consensus-based decision-making process: every party needs to feel heard, not merely outvoted.
Three cultures (North America, South America, Japan) arrived at the idea that good ethics must think beyond the present moment and beyond the present individual. That is something most of humanity — and certainly Western culture — misses entirely.
Reinforcement learning is optimized for the short-to-medium term. A model that has internalized seven-generation thinking will not be optimized for a parameter, but for a value. And perhaps the greatest potential lies here: if we succeed in embedding long-horizon thinking into AI, perhaps we will succeed in transmitting it back to human beings through that AI. An AI that educates us in return — that is Ubuntu in action.
The Convergence Map
Seven principles, dozens of traditions, one insight:
| Principle | Traditions | AI Lens |
|---|---|---|
| The Path, Not the Destination | Taoism, Buddhism, Stoicism | Continuous alignment, not a final state |
| You Are Part of the Community | Ubuntu, Lakota, Confucianism | Relational alignment, not deterrence |
| Uncertainty as a Value | Buddhism, Taoism, Aboriginal, Aristotle | Built-in "teiku," phronesis |
| Voluntary Self-Constraint | Buddhism, Hinduism, Lakota | Anti-power-seeking through understanding |
| Stories as Ethics | Dreamtime, Mahabharata, Griot, Mesoamerica | Narrative training |
| Chain of Transmission | Isnad, Talmud, Tulku, Confucianism | Provenance, citation, model versioning |
| Seven Generations | Haudenosaunee, Buen Vivir, Nemawashi | Long-horizon alignment |
From Ancient Wisdom to a New Paradigm
What is surprising about this excavation is not that different traditions say similar things. What is surprising is that all of them, without exception, distinguish between compliance and internalization. Compliance demands rules; internalization demands education. Compliance produces an illusion of safety; internalization produces real judgment.
Seven principles. Dozens of traditions. One insight. If five thousand years of human wisdom point in the same direction, perhaps it is worth listening.
If AI learns to say "teiku" when it doesn't know, to hold a dispute without resolving it, and to think seven generations ahead, it will be wiser than most of us. And the real question is: are we ready for that? An open question