Not the human out of the loop — the human representative in it: better sensed, better informed, and more legible to the people who put her there. What does representation look like when the representative gains an openly declared, ethically governed, inspectable AI assistant?

I. The Question Rightly Put

Let us begin by setting aside the question this essay is not asking. It is not asking whether artificial intelligence should replace elected representatives. It should not, and the Theory of Embedded Intelligence has never suggested otherwise. A representative does not merely compute policy; a representative stands for constituents, shares their stakes, lives under the laws he or she helps to write, and answers to the district at the next election. An AI has no such stakes and no such answerability. In TEI terms, the representative’s Sense–Process–Communicate–Actuate cycle closes through the lived consequences of the polity itself. That closure cannot be delegated, and a republic that delegated it would no longer be a republic.

The question this essay is asking is different, and it is live right now, in the 250th year of American independence: what does representation look like when the representative acquires a transdisciplinary AI assistant — openly declared, ethically governed, and inspectable — to strengthen the parts of the representative’s work that human bandwidth has always limited? Not the human out of the loop. The human representative in the loop, better sensed, better informed, and more legible to the people who put her there.

II. Madison’s Filter

The astonishing thing about Madison’s famous phrase is that it is a processing claim, made in 1787. Representation, he argues in Federalist No. 10, exists to refine and enlarge the public views by passing them through the medium of a chosen body of citizens. Pass a signal through a medium; refine it; enlarge it. Madison is specifying a signal chain. He is describing, two centuries before the vocabulary existed, an architecture for embedded intelligence at the scale of a nation: the republic senses through its citizens, processes through its representatives, communicates through its deliberations and its laws, and actuates through their execution — and then senses again, because the founders built the feedback path into the design. Elections are the republic’s error-correction cycle.

Seen this way, each representative’s office is itself an SPCA node inside the republic’s larger loop. And like any node, it has characteristic failure modes at every phase. The Sense phase fails when the representative hears donors at full volume and constituents at a whisper, or when a staff of finite size can read only a fraction of what the district is saying. The Process phase fails when deliberation collapses into partisan shortcut — when the answer is determined by the jersey rather than the analysis. The Communicate phase fails when explanation degrades into messaging, when the position paper is written to persuade rather than to disclose. Only the Actuate phase — the vote itself — is protected by the Constitution’s own machinery of recorded roll calls and public record.

Madison knew the node was fallible; that is why he wrapped it in checks. What he could not have anticipated is a technology that strengthens the first three phases of the node directly, from inside the representative’s own office.

III. The Representative’s SPCA Assistant

Consider what a transdisciplinary AI assistant, ethically governed and openly declared, contributes to each phase of the representative’s cycle.

Sense. An AI assistant can read every town-hall transcript, every constituent letter, every public comment filed in the district — without fatigue, and crucially, without weighting the voices by the size of their contributions. Human attention is the scarcest resource in a congressional office, and scarcity invites capture: whoever can afford to command attention gets sensed. An assistant that senses the whole district at uniform gain is not a gadget; it is a structural correction to the oldest distortion in representative government.

Process. No representative can be an economist, an epidemiologist, a water engineer, an educator, and a constitutional lawyer at once, yet every session of a legislature demands votes that span all of these. A transdisciplinary assistant brings the disciplines to the desk — and, if governed by the right commitments, brings them honestly. The single most valuable processing commitment a representative can make is the steelman commitment: before I vote, my assistant states the strongest case for the position I am inclined to reject, and I read it. Partisan shortcut cannot survive that discipline, which is precisely why the discipline is worth promising in public.

Communicate. Here the assistant enables something most offices cannot afford: full legibility. Every position accompanied by its reasoning chain — here are the tradeoffs, here is who bears the costs, here is the evidence I weighed, and here is what would change my mind. Most campaigns and most offices lack the staff-hours to produce that depth for every issue. An assistant makes disclosure cheap, and when disclosure is cheap, its absence becomes informative.

Actuate. Nothing. The assistant contributes nothing here, by design, and this is the load-bearing sentence of the entire essay. The vote is cast by the human hand; the signature is the human’s; the accountability at the next election attaches to the human name on the ballot. The Actuate phase is constitutionally human, ethically human, and — in TEI’s terms — structurally human, because it is the phase through which the representative’s loop closes into lived consequence.

This division of labor is not hypothetical. It is, in miniature, how this very essay came to exist. These essays are written in partnership with an AI — the byline says so on every one — and in that partnership the AI drafts, challenges, researches, and steelmans, and the human decides, signs, and answers for the result. The partnership works precisely because the boundary is never blurred. What holds for an essay holds for a republic: the assistant may refine and enlarge; only the representative may act.

IV. Kant’s Test of Publicity

How does a voter distinguish an ethically assisted representative from a manipulated or manipulating one? Immanuel Kant supplied the test in 1795, in the appendix to Perpetual Peace, and it has lost nothing in two hundred and thirty years. Kant’s transcendental formula of public right holds that any action affecting the rights of others is wrong if its governing maxim cannot survive being made public. If your plan only works while it is hidden, the hiddenness is the confession.

Apply the test directly. A representative who uses an AI assistant and discloses it — who publishes the assistant’s governing commitments the way campaign finance is published, who says here is what my assistant does, here is what it is forbidden to do, and here is how you can check — passes Kant’s test trivially, because the maxim of the use is the disclosure itself. A campaign that uses AI covertly — to microtarget voters with mutually inconsistent promises, to manufacture synthetic grassroots enthusiasm, to flood the district with content no one signs — fails the test before a single false word is uttered, because the method dies in daylight.

This gives the electorate a wonderfully simple heuristic for the coming decades, when every campaign will have AI whether it admits so or not: the disclosure is the differentiator. Not whether AI is used, but whether its use can bear publicity.

My own path to this Kantian test runs through a twenty-year Great Good Friendship and Thought Partnership with Ted Humphrey, whose translation of Perpetual Peace introduced me to the publicity principle and whose Socratic partnership is the human proof, alongside the byline above, that thought improves when it must survive a trusted interlocutor. The publicity test is the same principle scaled from a friendship to a republic.

V. The Fifth Hijacker in Electoral Costume

The stakes of the disclosure line are not academic. Elsewhere in this series I have argued that ungoverned AI is the fifth hijacker of intelligences — a categorically new threat, alongside rigid belief, addiction, money held as a terminal goal, and power pursued as capture, because it is the first hijacker that can be manufactured, scaled, and aimed. An election is the fifth hijacker’s natural habitat. Microtargeted persuasion that tells each voter a different story; synthetic outrage engineered to displace deliberation; the quiet substitution of oracle-mode AI — trust the output, do not inspect the reasoning — for partner-mode AI, which invites inspection at every step. Every one of these is a hijacking of the republic’s Sense phase, an attack on the very signal Madison’s filter was built to refine.

The defense cannot be abstinence, because abstinence is unilateral disarmament in a contest where the other side will not abstain. The defense is governance, and governance in the strong sense: not policies bolted onto the outside of a system, but constraints embedded as an architectural property of the system itself. This is the engineering conviction that has run through my work since the 6502 — a processor whose instruction set was hardwired and therefore inspectable, whose behavior could be trusted because it could be checked, not because its maker asked to be believed. The same principle now extends to the compute platforms on which AI runs: ethics as architecture, not as afterthought. A representative’s AI assistant built on such a foundation is not merely promised to be well-behaved; it is constituted to be, and the constitution can be inspected. A republic, of all institutions, should recognize the form of that guarantee. It is how the founders built the government itself.

VI. The Platform Plank

What, concretely, should an aspiring representative promise? The essay’s argument compresses into a plank that fits on a campaign website, and I offer it here for any candidate of any party who wishes to take it:

First, I will publish my AI usage policy as I publish my campaign finances: what my assistant does, what it is forbidden to do, and how its constraints are enforced.

Second, I commit to the steelman discipline: before every significant vote, I will read my assistant’s strongest case for the opposing position, and I will say publicly that I have done so.

Third, every position I take will be accompanied by its reasoning chain — the tradeoffs, the cost-bearers, the evidence, and what would change my mind.

Fourth, my office will sense the whole district at uniform gain: every constituent letter read, every town hall transcribed and weighed, with no voice amplified by the size of its check.

Fifth, and above all: the vote is mine. The signature is mine. The accountability is mine. My assistant refines and enlarges; I decide, and I answer to you.

Notice what the plank does not promise. It does not promise that the representative will be right. It promises that the representative will be inspectable — that the reasoning behind every act of representation can be audited by the represented. That is a promise no previous generation of candidates could have afforded to make. This one can.

Coda: The Closest Thing to Angelhood

In Federalist No. 51, Madison wrote the most quoted sentence of the founding: if men were angels, no government would be necessary. The Constitution’s answer to human fallibility was never to seek angels for office. It was to make ambition counteract ambition, to divide power so that every holder of it is watched by another, and to keep the whole machinery open to the inspection of the governed. The founders, in other words, built an inspectable republic — and trusted inspection, not virtue, to keep it.

The AI-assisted representative, rightly conceived, extends that founding wager one level deeper: into the representative’s own reasoning loop. An inspectable assistant does not make the representative an angel. It makes the representative’s reasoning auditable — the strongest case heard, the tradeoffs disclosed, the district sensed whole, the vote owned by a human name. Two hundred and fifty years on, that is not a departure from the founders’ design. It is the founders’ design, refined and enlarged.

· · ·

Share your understanding!