When AI begins ranking, recommending, and even speaking for us, "ethics" is no longer just a topic in philosophy class — it is embedded daily in the operating systems of products, organizations, and families. Today's four thinkers take four sharply different roots: Aristotle defines virtue as the habituated optimum he calls the mean; Mencius locates morality in the innate "four sprouts" of the human heart; Kant raises morality to the universalizable law of reason in his categorical imperative; Mozi pushes morality outward to a proto-utilitarian "impartial care" without preferential weighting. These four paradigms correspond, with uncanny precision, to four alignment strategies in AI — virtue ethics, intuitionism, deontology, consequentialism — and the conversation begun two and a half millennia ago continues in today's RLHF annotation rooms.
Aristotle
Western · Ancient Greek Philosophy
Nicomachean Ethics (c. 350 BCE)
Source / Core Claim
"Virtue is a state of character concerned with choice, lying in a mean — a mean between two vices, the one of excess and the other of deficiency — determined by reason and by that reason by which the person of practical wisdom would determine it."
Translation note: Courage stands between cowardice and recklessness; generosity between stinginess and waste; proper pride between self-abasement and arrogance. Virtue is not a universal formula but a contextual optimum (mesotes), calibrated case by case by practical wisdom (phronēsis).
Exposition
Aristotle's ethics is not a list of "what to do" but a project of "what kind of person to become." Virtue (aretē) is neither innate nor a rulebook; it is a stable character formed through repeated practice — "we become just by doing just acts." This sets him apart from Plato, who locates the Good in a transcendent Form: Aristotle brings the good back to concrete life, defining it as the excellent performance of what is distinctive to being human (his ergon argument). The mean is not bland compromise; it is the optimum located, in each situation, by phronēsis, and it shifts with object, timing, and intensity. The good person is not one who follows a formula, but one whose emotions and judgment have been trained, over time, to find the target naturally.
Cross-Disciplinary Resonance
Aristotle's "mean" in control theory is optimal control: minimizing a cost function between excess and deficiency. The exploration–exploitation trade-off in reinforcement learning is virtue formation in algorithmic dress — neither fully conservative nor fully adventurous, with phronēsis playing the role of the dynamically adjusting policy. Prefrontal modulation of the limbic system in neuroscience is almost the physiology of virtue: emotion is not suppressed but trained to match context. In complexity, the "mean" maps to the stable region of an attractor — stray too far and the system leaves its sustainable zone.
Contemporary Application
Everyday: A leader cannot be measured on a single axis of "decisive" or "cautious"; the question is whether, in each situation, the leader hits the mean — decisive in crisis, listening in normal times, lenient with newcomers, strict with principles. Virtue is context-sensitive capacity, not a fixed label.
BigCat: In investing, position sizing is an Aristotelian thesis on the mean: over-concentration is arrogance, over-diversification is fear, and the optimum shifts with volatility, information density, and life stage. Parenting too: "strict" vs. "lenient" is not either/or — in each concrete conflict, phronēsis must locate the right point. Today the child procrastinates on homework — empathy or a boundary? There is no formula, only judgment trained by repetition. Internalizing the mean as second nature is precisely the core capacity that cannot be outsourced in the AI era.
In Brief
Aristotle defines virtue not as following rules but as the habituated capacity to hit the contextual mean between excess and deficiency, guided by practical wisdom (phronēsis). Courage lies between cowardice and recklessness; generosity between stinginess and waste. Virtue is dynamic optimization, mirrored today in optimal control theory and the prefrontal modulation of emotion.
Question to Sit With
In your last major decision, what would "excess" have looked like, and "deficiency"? Did you hit the mean by formula, by emotion, or by judgment you had already trained?
Mencius 孟子
Eastern · Confucianism
Mencius, "Gongsun Chou I" (c. 4th century BCE)
Source / Core Claim
恻隐之心，仁之端也；羞恶之心，义之端也；辞让之心，礼之端也；是非之心，智之端也。人之有是四端也，犹其有四体也。
Translation: "The heart of compassion is the sprout of benevolence (ren); the heart of shame is the sprout of righteousness (yi); the heart of deference is the sprout of ritual propriety (li); the heart of right-and-wrong is the sprout of wisdom (zhi). That people possess these four sprouts is as natural as their having four limbs." — When you see a child about to fall into a well, an instinctive shudder of pity rises in anyone, not in order to win favor with the parents or reputation among neighbors — it is the moral seed already given in human nature.
Exposition
Mencius's "four sprouts" (si duan) is one of the boldest claims in the history of Chinese thought: morality is not a product of social conditioning but a "sprout" (duan, a beginning, a budding tip) already present in the human heart. It must be cultivated and extended, but its seed is innate. The position runs directly counter to Xunzi's "human nature is bad" and points in the opposite direction from Hobbesian social contract theory. The thought experiment of "the child about to fall into a well" is its core argument: strip away every external motive, and the uncalculated trembling that remains is conscience itself. From this Mencius infers that "anyone can become a Yao or Shun" — morality is not the privilege of a few sages but the default configuration of every human being, simply overlaid in most people by desire. Cultivation is therefore not injection from outside but "seeking the lost heart" (qiu qi fang xin) — recovering what was always already there.
Cross-Disciplinary Resonance
The four sprouts resonate strikingly with modern moral psychology. Jonathan Haidt's social intuitionism shows moral judgment arises in under 200 ms; rational argument is largely post hoc rationalization — the experimental version of celin zhi xin (the heart of compassion). Hamlin's infant morality experiments find that six-month-olds already prefer "helpers" over "hinderers," a near-empirical demonstration of innate moral inclination. Evolutionary psychology grounds this in kin selection and reciprocal altruism hard-wired into neural circuitry. The mirror neuron system means we activate corresponding regions when we see another suffer — the neural substrate of compassion. In AI alignment, pure rule-based ethics (RLHF reward models) remains unstable, hinting that the robustness of human morality draws partly on these innate "value anchors" — exactly the layer today's large models most lack.
Contemporary Application
Everyday: In organizational culture work, rehearsing value statements is less effective than designing situations that activate the four sprouts — letting employees encounter real customer suffering once awakens compassion more than ten compliance trainings.
BigCat: The most effective moral education in parenting is not explanation but refusing to block the child's innate sprouts. When the child grieves for an injured cat at the roadside, do not rush to "rationalize" or redirect — sit there in the feeling with them. You are protecting a seed that is already alive. Similarly, becoming an "AI-augmented individual" is not just stacking tool efficiency; it is keeping intact the inner system still capable of being moved by another's pain. That system is the meta-ethical source for deciding "what AI should be used for." When everyone has tools, only direction is scarce — and direction grows from a conscience that has not been worn smooth.
In Brief
Mencius argues that morality is innate: every human possesses "four sprouts" — compassion, shame, deference, and moral judgment — as natural as having four limbs. His thought experiment of a child about to fall into a well anticipates modern moral intuitionism and infant morality research, suggesting ethics is rooted in pre-reflective neural circuitry, not pure social construction.
Question to Sit With
Recall a recent moment when compassion was triggered in you. Did you act on it, or did you reason it away? Which response brought you closer to who you want to be?
Immanuel Kant
Western · German Classical Philosophy
Grundlegung zur Metaphysik der Sitten (Groundwork of the Metaphysics of Morals, 1785)
Source / Core Claim
"Act only according to that maxim by which you can at the same time will that it should become a universal law." (Formula of Universal Law)
"Act so that you treat humanity, whether in your own person or in another, always as an end and never merely as a means." (Formula of Humanity)
Translation note: The essence of morality is not the goodness of consequences but whether the motive can withstand the logical test of "what if everyone acted on this maxim?"
Exposition
Kant pulls ethics back from experience into reason. If morality is grounded in feeling, custom, or consequence, it remains contingent, always overturnable by exception. A genuine moral law must be necessary and universal, like a mathematical theorem, and can therefore only come from reason itself. The Categorical Imperative is this pure rational form — "categorical" (unconditional, not "if you want X") and "imperative" (an objective demand on the will). Two tests run together: (i) Can your maxim be universalized without self-contradiction? — lying cannot be universalized, because universal lying would destroy the very institution of promising; (ii) Are you treating the other as an end in themselves with free will, not merely as an instrument? This ethic is robust against corrosion — it refuses to sacrifice the individual "for the greater good"; its cost is occasional rigidity, as in the famous case of the murderer at the door.
Cross-Disciplinary Resonance
The universalization test in game-theoretic language is "what if every agent adopted this strategy" — the ethical inverse of Nash-equilibrium reasoning. Protocol design in computer science is almost entirely Kantian: a network protocol that only works if most nodes obey is not a legitimate protocol. In AI alignment, Anthropic's "Constitutional AI" descends directly from Kant — constraining model behavior by a set of universalizable principles rather than case-by-case annotation. "Treat the human as end, not merely means" is the heart of data ethics: when an algorithm reduces users to "attention units to be optimized," it has already violated the Formula of Humanity. Blockchain's "trustless design" echoes the same idea — rules are valid by their universal executability, not by counting on participants' goodwill.
Contemporary Application
Everyday: Considering a "gray-zone" act (an inflated résumé, hidden bad news, a small advantage taken at another's expense), run Kant's test: if everyone in this industry did this, would the industry still exist? If the answer is no, you are free-riding on a common structure you are simultaneously corroding.
BigCat: Leadership and parenting both face the Formula of Humanity. Treating employees as "KPI-producing resources," or children as "vessels for fulfilling your expectations," already violates Kant. A simple self-check: does the demand you make respect the other as a being with free will? In the AI era this matters more, not less — when tools amplify your impact on others, treating persons as ends rather than means is the line separating the "augmented individual" from the "amplified exploiter." Investing too: do you see the counterparty only as an adversary in a game, or as a real person inside a real institution? Long-term compounding often rewards the latter.
In Brief
Kant grounds morality in pure reason: an act is right only if its maxim can be universalized without contradiction, and if it treats persons as ends in themselves, never merely as means. This deontological framework is the philosophical ancestor of protocol design, Constitutional AI, and data ethics — robustness through formal universality rather than case-by-case consequence calculation.
Question to Sit With
Take a decision you made this week. If everyone in your company, family, and society acted on the same maxim, would the result be flourishing or collapse?
Mozi, chapters "Impartial Care" (Jian Ai) and "Against Offensive War" (Fei Gong), c. 5th century BCE
Source / Core Claim
视人之国若视其国，视人之家若视其家，视人之身若视其身。
兼相爱，交相利。
Translation: "Regard another's state as you regard your own, another's family as your own, another's person as your own." — "Care for each other impartially, benefit each other reciprocally." The chaos of the world arises from partial love: rulers do not care for other states, hence aggression; ministers do not care for other clans, hence disorder; people do not care for others, hence harm. Replace partial love (bie ai) with impartial care (jian ai) and the world is ordered. Care without gradation; benefit without private claim.
Exposition
Mozi is the "heretic" of Chinese intellectual history — he directly challenged the Confucian foundation of "graded affection." Confucianism holds that love for parents necessarily outweighs love for strangers, and makes this the root of ritual. Mozi argues, with an almost modern utilitarian eye, that gradated love is the very source of conflict. Jian ai (impartial care) does not require that you feel identical affection toward everyone; it requires that in your moral weighting, the welfare of others counts equally with your own — it is a stance, not a sentiment. Mohism developed a tight logic (the Mohist Canons), engineering (defensive siegecraft), and the "three standards" — any claim must withstand the tests of history, common experience, and utility. Together, jian ai, "no offensive war," "frugality," and "elevating worth" form China's earliest systematic proto-utilitarianism. Mozi anchors the system to "the will of Heaven" (tian zhi) to prevent impartial care from collapsing into relativism. Mohism was once a "prominent school" alongside Confucianism; it faded after the Qin and Han but its questions land more sharply today than ever.
Cross-Disciplinary Resonance
Algorithmically, impartial care corresponds to equal weighting in a utilitarian calculus — every person's welfare carrying the same coefficient in the objective function. This is the spirit of Bentham and Peter Singer's "expanding circle"; Effective Altruism is essentially contemporary Mohism. In game theory, "impartial care + reciprocal benefit" is the condition for stable cooperative equilibria: when participants internalize each other's payoff into their own utility (utility interdependence), the prisoner's dilemma dissolves. In AI safety, Stuart Russell's assistance games and inverse reinforcement learning — having AI optimize for human welfare as a whole rather than a single principal's instructions — sit very close to the Mohist project. From the complex-systems view, partial love creates locally optimal networks prone to involution and zero-sum traps; only impartial global weighting can escape them. Mozi posed the ethical foundation of multi-agent coordination twenty-five centuries ago.
Contemporary Application
Everyday: Cross-departmental conflict in organizations is often rooted in "partial love" — KPIs let each team optimize a local interest. Introducing "the other's gain is part of my own" dimensions (NPS cross-evaluations, cross-team OKR links) is Mohist ethics engineered for modern firms.
BigCat: In investing, Mozi reminds us that long-term excess return often comes from positive-sum positions, not from winning zero-sum games — investing in expanding the pie (infrastructure, AI platforms, education) tends to beat fighting over the existing stock. In parenting, cultivating the child's jian ai dimension — caring not only about classroom ranking but about community, planet, future generations — sets the ceiling of their long-run vision. An "augmented individual" who only amplifies partial love becomes a destructive node; only an impartial augmented individual amplifies the whole even while amplifying the self — perhaps the scarcest, least replaceable ethical posture in a tool-rich age.
In Brief
Mozi advocates "impartial care" (jian ai): the welfare of others must carry equal moral weight as one's own, treating every state, family, and person as one's own. Often considered an ancient proto-utilitarianism, it anticipates Effective Altruism, cooperative game theory, and multi-agent AI alignment — solving the tragedy of partial love by expanding the moral weight function.
Question to Sit With
If, in your next major decision (investment, hire, family arrangement), you raised the weight on "impact on people outside the system" from 0 to 0.3, would the decision change? In which direction — shorter-sighted or longer?
Four Ethical Paradigms: A Dialogue Across Two Millennia
Synthesis
Virtue ethics · Intuitionism · Deontology · Consequentialism
Exposition
These four thinkers cover, with elegant precision, the four main routes of ethics:
· Aristotle (virtue ethics): good action flows from trained character — answering "what kind of person to become."
· Mencius (intuitionism / innate-nature view): good action flows from preserved innate conscience — answering "what is already in the heart."
· Kant (deontology / rule ethics): good action flows from universalizable rational principle — answering "is the act itself permissible."
· Mozi (utilitarianism / consequentialism): good action flows from maximizing global welfare — answering "what are the consequences."
They are not mutually exclusive but different angles on the same ethical decision. A mature moral judgment usually activates all four at once: first run Kant to check the floor; then Mozi to estimate consequence distribution; then Mencius to verify that conscience is not being suppressed; finally Aristotle to locate the contextual mean. AI alignment research today faces exactly this problem of "ethical pluralist integration" — any single paradigm proves insufficient.
Question to Sit With
Next time you face a knotted moral dilemma, run all four lenses in turn. Which is the one you most often skip? Add it back — the decision will become more robust.