The Course We Carve
Take a group’s values and its knowledge as far as they’ll go, and one thing is still missing: the act of choosing. Many wills flow together into a current that is no single stream — and that current then cuts the banks that will channel every drop to come. This essay is about that carving: how a we makes a choice it can own, why there’s no clean machine for it, and how the same river that widens a valley can be dammed to a standstill.
The question, and why it isn’t a tally
Ask how a community decides and the mind reaches, almost at once, for a count. Put the question, collect the votes, read off the majority: the motion carries, fifty-two to forty-eight. It’s a natural picture, and — like the poll and the library before it — it’s too small. A tally can tell you how a heap of people, asked one at a time, came down on a Tuesday. It can’t by itself tell you what the we decided, because a collective choice is no more the sum of its members’ choices than a we’s values were the average of theirs, or its knowledge the contents of their heads.
So this essay makes one claim about that choosing, the last of these three essays on the collective mind: a we’s decision is emergent judgment — a real collective choice that needn’t match any single member’s choice — and turning many wills into one is not a solved problem. It’s provably not solvable by any clean rule, as we’ll see; which is why deciding is a craft and not a formula, and why so much rides on how it’s done. Values were the first essay’s subject and knowledge the second’s, but neither becomes an act until it passes through here. What a we cares about and what it can see just sit there until the we chooses — and in the choosing, a current forms that none of the tributaries is.
A verdict no one voted for
Start where the collective character of the thing is sharpest, because it’s the claim the rest leans on — and it was planted, twice, for this moment.
Picture a panel of three judges deciding a single case that turns on two questions: was there a valid contract, and was it broken? To be liable you need yes to both. The first judge finds a contract and a breach — liable. The second finds a contract but no breach — not liable. The third finds a breach but no valid contract — not liable. Now count. On the outcome, the panel splits two-to-one for not liable. But look at the two questions on their own: a contract was found by two of the three, and a breach by two of the three — so on the reasons, the panel says yes and yes, which is exactly what makes for liable. The same three people, each perfectly rational, hand you one verdict if you add up their conclusions and the opposite verdict if you add up their reasons.1 There’s no member you can point to whose judgment the group’s judgment simply is.
This is the discursive dilemma, and it’s the plainest proof in the book that a we’s stance isn’t its members’ stances added together. The group’s judgment is emergent in the exact sense the essay on what a we values gave the word: a real property of members-in-an-arrangement, made by how their votes are combined, belonging to none of them alone. Tributaries meet, and the current runs a direction no single stream was running.
And here the trilogy’s sharpest case of the downward reach comes with it. Once the panel picks a rule — decide by the reasons, or decide by the bottom line — and the verdict is fixed, every member is bound to a choice that, on the other rule, a majority of them reject. The collective doesn’t just emerge from its members; it turns and commits them, hands them a result they now have to own and act on, though not one of them, deciding alone, would have reached it. The river, once it’s formed, cuts the banks that carry every drop. That’s emergent judgment and downward causation in a single motion, and it’s why deciding together is a genuinely different problem from deciding alone.
No perfect map from many to one
If the group’s verdict can flip on nothing more than a choice of counting rule, the natural hope is that some better rule dodges the trouble. The twentieth century’s great, bracing discovery is that none does.
Kenneth Arrow proved it first and most sweepingly: no way of turning individuals’ rankings into one collective ranking can satisfy, all at once, a short list of conditions any fair method should want — count everyone, respect unanimity, let no single member dictate, and keep the ranking of two options from flipping just because a third one is in the room.2 Something always gives. The Marquis de Condorcet had glimpsed the shadow of it two centuries earlier: majorities can go in circles, preferring A to B and B to C and then, around the ring, C to A — so that “what the group prefers” has no stable answer at all. And the trouble isn’t only in the ranking; it’s in the honesty of the vote. Gibbard and Satterthwaite showed that every reasonable system with three or more options can be gamed: there are always situations where a voter does better by misstating what she wants than by saying it straight.3 Even the thinnest demand for individual liberty turns out to clash with the mildest demand for efficiency — Amartya Sen’s small, stubborn paradox of the Paretian liberal.4
It’s tempting to read all this as a scandal, proof that collective choice is broken at the root. It reads better the other way, and the framework has met this shape before. The book has already turned down the search for a final moral standard, on the grounds that any fixed target can fairly be asked to justify itself, and so was never final after all. The impossibility theorems are that same refusal, showing up in the mathematics of the vote: there’s no fixed, neutral, un-gameable rule that turns many wills into one and settles the matter for good.5 Deciding, like the morality it serves, is a direction you travel, not a machine you build once and trust forever. What the theorems retire is only the dream of a perfect device that would let a we choose without judgment. What they hand back is the judgment — and the responsibility that comes with it.
Then design is all
If no rule is perfect, then the choice among imperfect rules stops being a technicality and becomes a moral question — because each rule is a different bet about whose voice reaches the current, and how loudly.
The plainest ballot is the crudest: mark one name, and a whole ordered world of preference is flattened to a single dot, the runner-up worth exactly as much as the last-place finisher. The families of reform all try, in their own ways, to widen what the voter is allowed to say. Ranked ballots let her order the field, so a second choice isn’t simply lost. Approval ballots let her name everyone she can live with. Condorcet methods hunt for the option that would beat every rival head to head. And quadratic voting lets her spend a budget of “voice” to register not just which way she leans but how hard — where piling votes onto one question costs steeply more, so a committed minority can protect the thing it can’t bear to lose without simply overruling everyone else. It’s not a toy: a chamber of legislators has used it to sort its budget priorities, surfacing intensities a show of hands would have crushed flat.6
There are deeper redesigns than the ballot. Sortition — choosing deciders by lot, the way we still choose juries — builds a body that mirrors the people rather than the parties, and shields it from the short horizon of the next election; a citizens’ assembly drawn this way has broken deadlocks a legislature couldn’t, talking its way to a settlement on questions too raw for the floor.7 But no design is neutral, and the ones built to widen can quietly narrow. Liquid schemes, which let each member either vote directly or hand her vote to a trusted proxy and take it back whenever she likes, look like pure enfranchisement — and in practice have tended to concentrate power in a handful of “super-voters” who pile up the borrowed weight of the passive many, the very concentration the design was meant to dissolve.8 And drawing the banks of the river is itself a way to fix its course: gerrymandering — cutting the districts so the lines pick the winners before a single vote is cast — is the counter-dynamic in map-making, a rigged procedure wearing the face of a fair one. Representation is the channel a we’s wills flow through to reach its choice; every way of building that channel is a wager about which waters get through.
Deciding changes the deciders
A vote is only the last motion of deciding; the longer part is the talking, and the talking does something a tally can’t. Deliberation doesn’t just collect preferences — it forms them. Give a representative sample of a public a fair briefing, real access to opponents and experts, and time to argue it out, and their considered judgments move, reliably, away from the first gut answer toward something they can actually defend.9 A preference walked through the fire of other viewpoints is a different thing from one polled cold; this is the very “made in the asking” the first essay found in the world of value, now in the world of choice.
Which is why the mark of a good decision isn’t only that it be right but that it be owned — that the members can see themselves as its authors, not just its subjects. And here the reflexive loop the earlier essays traced does its work once more: the procedures a we practices shape the members it makes. A people long practiced in real deliberation becomes a deliberating people; one long trained to rubber-stamp becomes a deferring one. We build our institutions, and then they build us.
The case that a wider circle decides better is real, but it’s easily oversold, and being honest about that matters. A celebrated theorem once claimed that cognitive diversity flat-out beats individual ability on hard problems; a closer look showed it proved far less than the slogan promised, leaning more on the good of plain randomness than on diversity as such.10 What survives the scrutiny is the narrower, sturdier claim the book has leaned on all along: on genuinely hard ground — rugged, deceptive, the kind moral questions almost always are — a diverse body protects a search that a uniform one gives up on too soon. The payoff is real where the ground is rough, and claimed only there. That’s all the argument needs. It’s not a license to think that merely adding voices is, by itself, proof of wisdom.
Even granting all that, a hard limit remains, and it’s fatal to the easy hope that letting a chosen few deliberate can stand in for the whole. A sortition assembly widens the context for the hundred citizens inside the room — and leaves the millions outside untouched, asked to obey a conclusion they didn’t reach and can’t fully check. Make the assembly’s verdict binding on the rest and you’ve bought a coherent center by asking the edges for blind deference.11 The honest answer isn’t one enlightened chamber deciding for all, but deliberation nested — Ostrom’s lesson, that a large order governs itself best as many overlapping, semi-independent bodies, each deciding what it competently can, bound to the others by shared and revisable rules.12 Coherent pluralism, met in the essays on scale, is the shape of legitimate deciding too.
Deciding without a chooser
It helps to see that emergent judgment needs no central judge at all — that a current can find its channel with no hand on the tiller.
When a bee colony outgrows its home, it has to choose a new one, and it chooses well, settling on the best of a dozen sites with an accuracy that would shame most committees — and with no chooser anywhere in it. Scouts find candidates and dance their quality; the better the site, the longer the dance, the more recruits it draws to dance in turn; when the crowd at one site crosses a threshold, the swarm commits and flies.13 No bee holds the comparison; the comparison happens in the arrangement, a decision spread across thousands with no one deciding. Human markets show a paler version — prices, and the prediction markets built to sharpen them, boiling scattered guesses down to one moving number that often beats the experts.14
But the swarm carries a warning as plainly as a lesson, and it’s the honest boundary on the whole comparison. Bees are near-identical kin with a single stake; their disagreement is purely factual — which site is best — never a clash of values, because they have, in effect, only one. A human we is the opposite case: its hardest decisions are exactly the ones where members want different things, not merely where they guess differently about the same thing. Drop the bees’ quorum straight onto a value-divided people and you don’t get wisdom, you get its counterfeit — a coordinated faction tripping the threshold and calling the swarm’s momentum a consensus. Decentralized choice is a real and beautiful power; it comes together honestly only when what’s being decided is a question of fact against a shared aim. The moment the values themselves genuinely differ, the easy mechanisms stop being enough, and the difficulty this whole essay circles comes back.
The river dammed
Which brings the moral fork the essay exists to draw. A we can carve its course by widening — letting more of its members’ actual wills reach and shape the current — or it can narrow, manufacturing the look of one will by damming the tributaries that would trouble it. From the bank, a broad river and a walled sluice can both run smooth. The difference is everything.
The institutional forms of the narrowing are familiar: the voter turned away, the count suppressed, the district drawn to a foregone result, the referendum staged not to ask a question but to rubber-stamp an answer already chosen — each an agreement made by shrinking the demos until the inconvenient are outside it. The philosophical form is older, and runs through Rousseau’s general will: the seductive idea that a people has one true will above the mere sum of its members’, which — as Isaiah Berlin warned — becomes, in the wrong hands, a license to override the actual wills of actual people and call the coercion freedom, recasting the dissenter not as a citizen who disagrees but as an enemy of the whole.15
The subtlest form is the most instructive, because it shows how hollow a manufactured unanimity can be. Timur Kuran named it preference falsification: when the cost of dissent is high, members voice the opinion they’re expected to hold and hide the one they actually have — and because each of them reads the others’ masks as faces, a whole society can lock into a public consensus almost no one privately shares.16 Such a we looks supremely coherent — and is brittle to the touch, its unity a shared lie waiting for the moment the cost of honesty drops. That’s why the regimes of Eastern Europe seemed monolithic right up until the week they didn’t, and then dissolved almost overnight: the consensus had been empty all along. This is the counter-dynamic in the world of deciding, exact and complete — coherence bought not by widening the circle but by frightening it into silence, the will of all forged by cutting away the wills that compose it.
The newest version is built of software, and it’s worth naming precisely because it can wear the face of help. An algorithm set to maximize agreement will, left to itself, learn to trim the tails — to quietly drop the fervent minority view that spoils an otherwise agreeable summary, and hand back a smooth center no one quite holds.17 That’s the counter-dynamic on autopilot. But the same tools can be built the other way, and this is the hopeful note the essay won’t oversell: bridging systems that surface only the statements winning approval across factions who usually disagree, and selection rules held to keep a minority’s fair share in the room instead of rounding it away.18 The lesson isn’t that machinery narrows and humans widen. It’s the same test that has run through all three essays: whether the course is carved by taking in more of the members’ real wills, or by damming them — and the instrument, like the institution, can be built for either.
The honest edges
Three cautions, because the thing is real and, like the others, not tidy.
First, a decided course isn’t a delivered one. A we can hold clear values, see its world well, and choose wisely, and still get nothing done — because deciding and doing are different capacities, and the second can be hollow while the first looks sound. States can build the whole apparatus of a choice carried out — the ministry, the statute, the reported metric — while delivering none of the actual thing, mistaking the look of a functioning state for its function.19 A widened reach of concern that never reaches the world is a concern in name only; the link from decision to delivery is where many a fine choice quietly dies, and the framework earns no credit for a coherence that never touches the ground.
Second, legitimacy isn’t correctness. Everything this essay has praised — wide enfranchisement, real deliberation, a choice the members can own — makes a decision legitimate, theirs by right. None of it makes the decision right. A we can carve its course fairly and carve it straight toward a cliff; the counter-dynamic’s test tells you whether the choosing widened or narrowed, not whether the world will prove the choice out. That gap isn’t a flaw in the account but the same honesty the whole book keeps: no procedure certifies an outcome true, only a direction that keeps the deciding open to correction.
Third — and this is where the trilogy closes — a decision is a steering, not a destination. The river never stops moving. A choice a we can own is a choice it can also revisit when the world answers back; a we that treats any verdict as final has stopped carving its course and started walling it, trading the living channel for a stagnant certainty. To decide well together isn’t to reach the end of deciding. It’s to keep the current moving in the widening direction, one revisable choice after another.
The course we carve
Gather it up, and gather the three. A we’s decision is emergent, not added up; it binds the members it’s made of; it answers to no perfect rule, and so is a craft rather than a formula; it lives in the banks a we cuts and is cut by; and, like every choice a living thing makes, it’s never final — a steering still answerable to the world, or else a channel hardened into a wall. That’s how a we carves its course.
And that finishes the small map these three essays set out to draw. A collective has values that are more than the sum of its members’ — what it cares about. It has knowledge and capability that reach farther than any member can see — how it knows and acts. And it has a way of turning both into a choice it can own — how it decides. Values, methods, decision: the collective mind entire, with the same two-way traffic running through all of it — members composing a we, and the we turning back to shape the members. None of the three is something a we simply has. Each is work a we does, and can do widening or narrowing, and is more itself the more it does them in the open. The arrow that runs through a single life runs through the life we make together — and it points, as it always has, outward.
Sources & further reading
This essay leans on its sources directly, rather than through the book’s per-chapter endnotes. A citation-level pass (Zotero keys) is still owed; the works below were web-verified on 2026-07-07.
The formal limits of aggregation. Kenneth Arrow, Social Choice and Individual Values (1951), on the impossibility of a fair rank-aggregation rule; the Marquis de Condorcet on cyclical majorities and the jury theorem; Allan Gibbard, “Manipulation of Voting Schemes” (Econometrica, 1973), and Mark Satterthwaite (1975), on the universality of strategic manipulation; Kenneth May (Econometrica, 1952) on majority rule for binary choices; Amartya Sen, “The Impossibility of a Paretian Liberal” (Journal of Political Economy, 1970).
Emergent judgment — the discursive dilemma. Lewis Kornhauser and Lawrence Sager, “The One and the Many: Adjudication in Collegial Courts” (California Law Review, 1993), on the doctrinal paradox; generalized by Christian List and Philip Pettit, “Aggregating Sets of Judgments: An Impossibility Result” (Economics and Philosophy, 2002). The dilemma was planted in More Than Its Notes and paid off here.
Designing the vote. Eric Posner and E. Glen Weyl, “Voting Squared: Quadratic Voting in Democratic Politics” (Vanderbilt Law Review, 2015); on delegative schemes and their drift to oligarchy, the study of LiquidFeedback in the German Pirate Party (Christoph Kling and colleagues, 2015) and Blum and Zuber, “Liquid Democracy” (Journal of Political Philosophy, 2016); on sortition, David Farrell and Jane Suiter on the Irish Citizens’ Assembly (Reimagining Democracy, Cornell, 2019).
Deliberation, legitimacy, and the diversity question. James Fishkin, When the People Speak (2009), on deliberative polling; Hélène Landemore, Democratic Reason (2012), on the epistemic case for inclusive democracy — which leans on Lu Hong and Scott Page, “Groups of diverse problem solvers can outperform groups of high-ability problem solvers” (PNAS, 2004); read together with Abigail Thompson’s rebuttal, “Does Diversity Trump Ability?” (Notices of the AMS, 2014), which is why this essay rests the diversity dividend on the narrower, rugged-terrain result (Kevin Zollman; see Coherence at Scale) rather than on the Hong–Page theorem. Cristina Lafont, Democracy Without Shortcuts (2020), on the “blind deference” problem of binding minipublics; Jürgen Habermas on the ideal speech situation (Moral Consciousness and Communicative Action, 1990).
Deciding without a chooser. Thomas Seeley, Honeybee Democracy (2010), on quorum-sensing swarm decisions (with Pratt et al., 2002, for the quorum mechanism); Elinor Ostrom, Governing the Commons (1990), on polycentric self-governance; on markets as decentralized decision, Wolfers and Zitzewitz, “Prediction Markets” (Journal of Economic Perspectives, 2004).
The counter-dynamic in the decision register. Timur Kuran, Private Truths, Public Lies (1995), and “Sparks and Prairie Fires” (Public Choice, 1989), on preference falsification and the sudden collapse of manufactured consensus; Jean-Jacques Rousseau, The Social Contract (1762), and Isaiah Berlin, “Two Concepts of Liberty” (1958), on the general will and its abuse. On automated consensus and its alternatives: the “Habermas Machine” (Michael Tessler and colleagues, Science, 2024); bridging-based ranking (Aviv Ovadya, “Bridging-Based Ranking,” Belfer Center, 2022, as in X’s Community Notes; and the Polis / vTaiwan platform, Christopher Small and colleagues, 2021); and representation-preserving opinion selection (Salim Hafid and colleagues, Algorithmic Approaches to Opinion Selection for Online Deliberation, arXiv 2602.15439, 2026).
Decision and delivery. Lant Pritchett, Michael Woolcock, and Matt Andrews, “Looking Like a State: Techniques of Persistent Failure in State Capability for Implementation” (Journal of Development Studies, 2013), on isomorphic mimicry; Timothy Besley and Torsten Persson, Pillars of Prosperity (2011), on state capacity.
Within the framework: Chapter 3 — What Matters and More Than Its Notes (what a we values); Chapter 4 — What Works and Farther Than One Can See (how a we knows and acts); Chapter 5 — Selves Made of Selves (the emergent we); Coherence at Scale (nesting, and the rugged-terrain diversity result this essay relies on); The Tree of Agreement (convergence below the surface); Held Together, Held Apart and Coherent Pluralism (the counter-dynamic, and nested self-governance as coherent pluralism).
The premise-versus-conclusion split is the doctrinal paradox (Kornhauser and Sager, 1993), later generalized as the discursive dilemma (List and Pettit, 2002). Real appellate courts run into it: the judgment comes out differently depending on whether the court adds up the votes issue by issue or on the bottom line.↩︎
Arrow’s conditions, informally: unrestricted domain, Pareto efficiency, independence of irrelevant alternatives, and non-dictatorship. His theorem is that no rank-aggregation rule satisfies all four at once.↩︎
The Gibbard–Satterthwaite theorem: every deterministic, non-dictatorial voting rule over three or more options is manipulable — there are situations where a voter does better by misrepresenting her preferences.↩︎
Sen’s paradox: if even two individuals each have a single matter over which their own preference is decisive (minimal liberty), a society honoring both can be forced into an intransitive collective preference, colliding with the Pareto principle.↩︎
The same pattern shows up well beyond the vote. In formal epistemology, attempts to boil the coherence of a set of beliefs down to a single measure run into their own impossibility and non-uniqueness results — coherence turns out to need a vector rather than a single number, and no one coherence ranking holds up across sets of every size (Bovens and Hartmann, 2005, 2006). Wherever a single, complete, neutral measure or ranking is demanded of a plural whole, one provably isn’t to be had; see Measuring Coherence.↩︎
The Democratic caucus of the Colorado state House used quadratic voting in 2019 to rank budget priorities, letting legislators concentrate credits on the bills they cared about most.↩︎
The Irish Citizens’ Assembly (2016–2018), ninety-nine citizens chosen by lot, deliberated on constitutional questions including the Eighth Amendment, producing recommendations that led to a national referendum.↩︎
The German Pirate Party’s LiquidFeedback experiment (c. 2009–2013) is the standard cautionary case: a power-law distribution of delegated weight concentrated influence in a few “super-voters.”↩︎
James Fishkin’s deliberative polling: a representative sample is polled, briefed, given deliberation and expert access, then polled again — and considered judgments shift, often substantially.↩︎
Hong and Page’s “diversity trumps ability” (2004) was rebutted by Abigail Thompson (2014), who argued the result, corrected, is close to a tautology and better read as “randomness trumps ability.” The robust survivor is the narrower Zollman result on transient diversity in rugged epistemic landscapes, already used in Coherence at Scale.↩︎
Cristina Lafont’s critique of binding minipublics: because the wider public does not undergo the assembly’s deliberative transformation, granting the assembly decisional authority asks the public for “blind deference,” undermining self-government.↩︎
Elinor Ostrom’s polycentricity: overlapping, semi-autonomous decision centers, bound by shared and revisable rules, outperform both the single central authority and full fragmentation — the decision-register form of coherent pluralism.↩︎
Thomas Seeley, Honeybee Democracy (2010): scouts advertise sites by waggle-dance with intensity proportional to quality; a quorum at one site triggers commitment. No individual bee compares the options.↩︎
Prediction markets aggregate dispersed, self-interested estimates into a price that tracks probability, frequently outperforming individual experts — decentralized judgment with no central forecaster.↩︎
Rousseau’s volonté générale distinguishes the “general will” from the “will of all” (the mere sum of private interests); Berlin’s “Two Concepts of Liberty” traces how the former licenses coercion “for the citizen’s own good.”↩︎
Timur Kuran, Private Truths, Public Lies (1995): under social pressure, private and publicly expressed preferences diverge; cascades of falsification can lock in an equilibrium almost no one privately supports — and shatter without warning when the cost of dissent falls (his reading of 1989).↩︎
Consensus-maximizing selection algorithms tend to discard high-conflict (long-tail) opinions; Hafid and colleagues (arXiv 2602.15439, 2026) document this and propose selection constrained to preserve balanced representation (their “DiverseBJR”).↩︎
Bridging-based ranking surfaces content approved across normally opposed groups (as in Polis / vTaiwan and, at scale, community-note systems), rewarding cross-factional assent rather than raw majority.↩︎
Pritchett, Woolcock, and Andrews (2013) call this isomorphic mimicry: adopting the outward forms of a capable state to camouflage the absence of function — the decision-to-delivery link severed. Note the deliberate echo of the legibility critique in Farther Than One Can See.↩︎