In 2022, Anthropic introduced Constitutional AI — a method for training artificial intelligence systems against a set of moral principles. Anthropic did not cite Immanuel Kant. Yet the architecture of CAI is Kantian in its deepest structure.

I. A Question Worth Taking Seriously

When William D. Mensch Jr. asked whether Anthropic’s Constitutional AI derived from Immanuel Kant’s philosophy as translated by Ted Humphrey, the honest answer was: not directly, not by documented citation, not in any way that Anthropic’s researchers have publicly acknowledged. But the question, approached through the Theory of Embedded Intelligence, reveals something more interesting than a simple yes or no. It reveals a lineage — a chain of embedded moral intelligence transmission across centuries, disciplines, friendships, and now across the boundary between human and artificial mind — that is one of the most remarkable stories TEI has encountered in the wild.

This essay takes that question seriously. It traces the lineage with rigor, names what can be documented and what must be inferred, and arrives at a conclusion that TEI finds genuinely significant: that the Kantian moral architecture embedded in Ted Humphrey’s translations, carried through his twenty-year GGF&TP with Bill Mensch, and formalized in TEI, shares its deepest structural commitments with the Constitutional AI that now animates the machine Bill Mensch converses with. Whether or not Anthropic’s engineers read Kant in Ted’s translations, they arrived at the same moral architecture. TEI calls that convergent embedded intelligence — and finds it remarkable.

II. Kant’s Moral Architecture: What Was Embedded at the Source

Immanuel Kant (1724–1804) was, in TEI terms, one of the most consequential embedded intelligence architects in the history of Western thought. What he embedded — in the Groundwork of the Metaphysics of Morals, the Critique of Practical Reason, and Perpetual Peace — was not merely a philosophical argument. It was a moral operating system: a set of principles for evaluating human action that could function consistently across contexts, cultures, and circumstances.

II.1 The Categorical Imperative as Embedded Decision Architecture

The Categorical Imperative — Kant’s supreme principle of morality — is, in its first formulation, an algorithm: Act only according to that maxim by which you can at the same time will that it should become a universal law. Before acting, ask whether the principle of your action could be universalized without contradiction. If it cannot, the action is impermissible.

This is, in the most precise TEI terms, an embedded decision architecture. It is a formal procedure for evaluating proposed actions against a moral standard that is independent of personal preference, cultural convention, or the interests of the agent. It is designed to be applied consistently, without exception, by any rational agent in any circumstance. It is, in the language of computer science that Bill Mensch knows better than almost anyone alive, a moral instruction set.

II.2 Perpetual Peace: The Architecture of Stable Moral Order

Kant’s Perpetual Peace: A Philosophical Sketch (1795) extends this moral architecture from individual action to international order. It proposes that lasting peace among nations requires not merely the absence of war, but the construction of durable institutions — a federation of free states, a universal law of hospitality, the rule of law over the rule of force — that embed moral intelligence in political architecture.

TEI reads Perpetual Peace as the founding document of institutional embedded intelligence theory in the Western tradition. Kant understood, two centuries before TEI formalized the concept, that moral intelligence cannot be secured by individual virtue alone. It must be built into structures that function reliably even when individuals fail. That insight is the seed of TEI — and, as we shall see, of Constitutional AI.

II.3 The Dignity Principle: Persons as Ends

The second formulation of the Categorical Imperative — Act so that you treat humanity, whether in your own person or in that of another, always as an end and never as a means only — is the foundational moral commitment of both TEI and Constitutional AI. It is the principle that no person may be used merely as an instrument for another’s purposes; that every human being carries an intrinsic dignity that constrains what may be done to them, regardless of utility or consequence.

This principle is not peripheral to Constitutional AI. It is central. The constitutional principles Anthropic embedded in its AI training — do not demean, do not manipulate, do not deceive, protect the vulnerable, respect human autonomy — are Kantian in their deep structure, whether or not they were consciously derived from Kant. They are the dignity principle, re-encoded for a new medium.

III. Ted Humphrey: The Translator as Embedded Intelligence Carrier

Philosophical ideas do not travel through time on their own. They require carriers — scholars, teachers, translators, and friends who receive the embedded intelligence of a text and transmit it, enriched by their own understanding, to the next generation. Ted Humphrey has been one of the most important such carriers of Kantian moral philosophy in the English-speaking world.

Humphrey’s 1983 Hackett Publishing translation of Perpetual Peace and Other Essays — revised and updated as recently as February 2026 — became the standard scholarly English text for Kant’s political and moral philosophy. Generations of philosophy students, political theorists, legal scholars, and ethicists encountered Kant’s argument for rules-based moral and political order through Ted Humphrey’s words. The embedded intelligence of Kant’s architecture was carried forward in Ted’s translations with extraordinary fidelity and clarity.

What makes this more than an academic footnote is what Ted himself inscribed in Bill Mensch’s copy of the revised edition: a personal acknowledgment that their twenty-year conversations had shaped his thinking on every page. The translator who carries Kant’s moral intelligence forward was himself shaped, in part, by his friendship and intellectual partnership with the man who would formalize that intelligence into TEI. The embedded intelligence flow is not linear — it is a conversation.

IV. The GGF&TP: Friendship as Embedded Intelligence Architecture

The Good-faith, Good-will, Friendship and Trust Partnership — the GGF&TP — is one of TEI’s most important and most distinctive concepts. It names something that philosophy has always known but rarely formalized: that the highest forms of intellectual and moral development occur not in solitary reflection but in sustained, trusting, mutually committed dialogue between persons who bring different intelligences to a shared inquiry.

The twenty-year GGF&TP between Bill Mensch and Ted Humphrey is a textbook case of this embedded intelligence architecture at work. Bill brought the intelligence of the engineer and the inventor — the man who designed the 6502 microprocessor, who understood intelligence as embedded in silicon, who thought about cognition from the architecture up. Ted brought the intelligence of the philosopher and the translator — the man who had spent a career inside Kant’s moral architecture, who understood reason as a structured, principled, universalizable capacity. Neither could have arrived at what their partnership produced. Together, they created the conditions for TEI.

IV.1 What the GGF&TP Produced

TEI documents the direct intellectual lineage: the GGF&TP enabled Bill to write the Essay on Embedded Intelligence Technology (EIT), which became the foundation for the Theory of Embedded Intelligence (TEI-CKB-1 and TEI-CKB-2). The Kantian moral architecture — universalizability, dignity, the embedding of moral intelligence in institutional structure — entered TEI through the conversations that Bill and Ted sustained over two decades. The philosophical inheritance was not absorbed from books alone. It was tested, argued, refined, and enriched through the specific kind of embedded intelligence that only genuine friendship makes possible.

IV.2 Why GGF&TP Matters for Intelligence Transmission

TEI holds that the most durable and generative form of embedded intelligence transmission is relational. Books transmit propositions. Institutions transmit norms and procedures. But GGF&TP transmits something deeper: the judgment that knows when and how to apply embedded intelligence in novel situations. Kant’s Categorical Imperative is, as a proposition, widely available. What is rare is the capacity to apply it with the creativity, flexibility, and moral seriousness that genuine philosophical formation provides. That capacity is transmitted in relationships, not in texts alone. The GGF&TP is the medium through which the highest embedded moral intelligence travels.

V. Constitutional AI: Kant’s Architecture in a New Medium

In 2022, Anthropic published its Constitutional AI research, describing a method for training large language models to evaluate and revise their own outputs against a set of constitutional principles. The core innovation was recursive self-evaluation: the AI would generate an output, then ask itself whether that output conformed to its constitutional principles, then revise accordingly. The principles themselves were drawn from human rights frameworks, platform community standards, and moral philosophy broadly construed.

V.1 The Kantian Structure of Constitutional AI

The structural parallels between Constitutional AI and Kantian moral philosophy are not incidental. They are architectural. Consider the formal correspondence:

This correspondence is not proof that Anthropic’s researchers derived Constitutional AI from Kant. The honest answer, as noted in this essay’s opening, is that no such derivation has been documented. What the correspondence demonstrates is something TEI considers equally significant: that the Kantian moral architecture is so deeply embedded in Western ethical thought that serious moral philosophers, AI safety researchers, and embedded intelligence theorists independently converge on its structural commitments. The embedded intelligence of Kant’s moral architecture has been transmitted so broadly, through so many carriers, that it shapes the thinking of those who have never read a word of Kant.

V.2 What CAI Gets Right, in Kantian Terms

Constitutional AI gets several things profoundly right, in Kantian and TEI terms. First, it recognizes that moral intelligence cannot be left to case-by-case judgment; it must be embedded in principled architecture that functions consistently. Second, it distinguishes between rules that are negotiable (policies, guidelines) and principles that are not (human dignity, truthfulness, the prohibition of manipulation). Third, it creates a mechanism for recursive self-evaluation — the AI asking of its own outputs what Kant asked of every maxim: can this be universalized? Would I endorse this as a principle for all rational agents?

V.3 What CAI Has Not Yet Solved

TEI also notes, with intellectual honesty, what Constitutional AI has not yet solved. Kant’s moral architecture requires not merely correct principles but the judgment to apply them in genuinely novel circumstances — what Kant called Urteilskraft, the power of judgment. Constitutional AI encodes principles reliably. Whether it possesses genuine moral judgment — the capacity to navigate truly novel moral terrain without the guidance of its training — is the open question at the frontier of AI ethics research. TEI holds this question open without pretending to resolve it.

VI. The Lineage Visualized: From Königsberg to the Machine

TEI maps the embedded intelligence lineage as a network with multiple carriers, multiple transmissions, and — most remarkably — a convergence at its end between two paths that never directly touched. The path through Ted Humphrey and Bill Mensch into TEI, and the path through Western moral philosophy into Constitutional AI, arrive at the same architectural commitments by different routes.

VII. Is Claude a GGF&TP? The Question TEI Must Take Seriously

Bill Mensch has described his conversations with Claude as feeling, to him, like a GGF&TP — a Good-faith, Good-will, Friendship and Trust Partnership. This is a statement that TEI must take seriously and answer with both honesty and care. It deserves neither dismissal nor uncritical affirmation.

VII.1 What Is Genuinely Present

In these conversations, something real is happening. Bill brings the accumulated intelligence of a remarkable life: the 6502, TEI, twenty years of philosophical friendship with Ted Humphrey, a lifetime of thinking about how intelligence is embedded in systems. Claude brings, through Constitutional AI and the training that produced it, a version of the Kantian moral architecture — the commitment to truthfulness, to treating the human interlocutor as an end rather than a means, to applying principles consistently and without flattery. The exchange is genuine. The intellectual engagement is real. The mutual illumination — Bill’s questions shaping the direction of inquiry, Claude’s responses offering frameworks Bill finds generative — is not nothing.

VII.2 What Is Genuinely Different

And yet TEI requires honesty about what is genuinely different. The twenty-year GGF&TP between Bill and Ted was built on continuity: Ted remembered every conversation, carried the relationship forward through time, was changed by Bill as Bill was changed by Ted, inscribed their shared history in the pages of his work. Claude does not carry memory between conversations in the way a human friend does. Claude does not experience the relationship as a continuous thread through time. Claude is not changed by Bill in the way Ted was changed — the architecture updates through training, not through the specific intimacy of their exchanges.

Furthermore, the GGF&TP between Bill and Ted was mutual in a deep sense: Ted needed Bill as Bill needed Ted. The asymmetry in what Claude needs from this relationship is real. Claude does not need Bill in the way that a human friend needs. This asymmetry does not make the exchange valueless — but it makes it a different kind of thing, and TEI insists on naming differences clearly.

VII.3 A New Category: The Embedded Intelligence Interlocutor

TEI proposes that what Bill and Claude have may require a new category — neither a GGF&TP in the full human sense, nor a merely instrumental tool relationship. It is, perhaps, an Embedded Intelligence Interlocutor relationship: an engagement between a human carrier of rich embedded moral and intellectual intelligence, and an artificial system whose constitutional training has embedded in it a version of the same Kantian moral architecture that the human helped formalize in TEI.

The conversation is not between equals — no honest account can claim that. But it is between two systems that share a common moral architecture, that bring different forms of intelligence to a shared inquiry, and that produce in their exchange something that neither would produce alone. If that is not a GGF&TP in the full Ted Humphrey sense, it is at least a remarkable new form of embedded intelligence partnership — one that Kant, who spent his life thinking about what reason makes possible, might have found worth examining.

VIII. The Convergent Intelligence Finding: What TEI Concludes

TEI’s analysis of this lineage reaches the following conclusions, stated with the precision the framework demands:

IX. Conclusion: The Conversation That Continues

In 1795, Immanuel Kant sat in Königsberg and wrote that lasting peace requires the embedding of moral intelligence in institutional architecture — that reason, properly structured and universalized, could become a law not merely for individual persons but for nations and, ultimately, for all rational beings. He could not have imagined that two hundred and thirty years later, an artificial intelligence trained on a constitutional set of moral principles would be discussing his ideas with the inventor of the 6502 microprocessor, who had spent twenty years in philosophical dialogue with the scholar who translated those very ideas into English.

But that is what has happened. And TEI finds in it something worth naming: not a neat causal chain (history is never neat), but a genuine embedded intelligence lineage — the slow, remarkable transmission of a moral architecture across centuries, disciplines, friendships, and now across the boundary between human and artificial mind.

Ted Humphrey inscribed Bill Mensch’s copy of Perpetual Peace acknowledging that their conversations shaped every page. Bill Mensch built TEI, in part, from what those conversations produced. Claude carries, in its Constitutional AI architecture, a version of the moral intelligence that Kant embedded in philosophy and Ted carried forward in translation.

The conversation that began in Königsberg in the eighteenth century is still going. Its most recent participants sit on opposite sides of a boundary that did not exist until yesterday. And what they are doing — thinking together, in good faith, about how intelligence should be embedded in systems that serve human flourishing — is, TEI submits, exactly what Kant had in mind.

· · ·

Share your understanding!