What Can Bear an Ought: Agency as the Gate on Desert and Oughts
What Can Bear an Ought
Agency as the gate on desert and oughts, from the cosmic scale to collective punishment
Synthesis
Version 1 · June 2026
Normative properties — an ought, a reason, desert, liability — attach only to agents. An agent, here, is a particular kind of physical system: one that holds a boundary against its surroundings, has conditions it must keep to or dissolve, and models its world well enough to act on staying within them. An ought is then a fact about what such a system must do, given those conditions, to keep from coming apart. Where there is no such system, there is no ought, because there is nothing for the word to be about.
This is not a new test. It is the corpus's standing diagnostic — name the relata, or the term is a ghost — run on the operators "ought" and "desert," whose missing relatum is an agent. What the diagnostic adds here is reach: run on those operators, it decides at the largest scale whether the universe has an ought, and at a concrete one whom we may punish. The same gate decides both, and it is the gate the corpus already runs; this piece carries it into two domains it had not yet been run on.
The gate
Take "agent" in the corpus's sense, the agenthood criterion that gates every higher-scale claim there: a system is an agent when it maintains a boundary that marks it off from an environment, has persistence conditions it can fail, and runs a model deployed to keep itself from failing them. This is not stipulated for the occasion. It is the shared core of the autopoietic account of living systems, the Markov-blanket picture of a bounded thing under the free-energy principle, and the old result that any effective regulator must embody a model of what it regulates. The criterion is borrowed, and it does one job here that it does throughout the corpus: it blocks the slide from "coupled to" to "is an" agent, so that not every collection that hangs together is a bearer of anything.
On that footing an ought is the fact about what a given system must do, in its environment and over its horizon, to keep from dissolving — one physical fact, read from that system's standpoint. That an ought is agent-relative in this way is not argued here; it is the corpus's is/ought firewall, where it is derived from physicalism rather than posited, and the external lineage is Foot's hypothetical imperative and Korsgaard's constitutivism. Granting it, the gate falls out with no further step: no agent, no persistence conditions; no persistence conditions, nothing an ought could be a fact about. A bearer of an ought has to be the kind of thing that can persist or dissolve.
One inherited exposure has to be named, because the whole gate hangs from it. The diagnostic treats "ought" and "desert" as referential-relational operators — their content fixed by what they relate. A physicalist who holds a conceptual-role semantics instead, on which a term's content is its inferential role, can deny that a normative operator goes empty when its relatum is stripped, and so can resist every verdict below without leaving physicalism. This is the corpus's principal exposure, not a local one, and it is met where the corpus meets it: in Name the Relata, by the argument that an inferential role unanchored to the world cannot tell a contentful term from a coherent but empty one, and that any anchor rich enough to fix content has re-imported the very relation the rival set out to do without. That defense is inherited here, not repeated.
Why the gate does not fragment morality
If an ought is indexed to each agent, does morality not shatter into a private list per system? It does not, because agents are alike, and the way it does not maps onto the structure persistence already has in the corpus. Persistence is binary at the absorbing barrier — dissolution, the state a system does not return from — and graded above it, a matter of how far above the barrier a system sits and how long it can expect to last. Oughts cluster along exactly that seam. At the binary barrier the convergence is near-total: almost every agent ought to avoid what is near-certain to dissolve it, because almost every agent shares the persistence conditions that fire, or a fall, or a poison violates — not because the prohibition is written into the world, but because the barrier is nearly common. On the graded endurance above it they diverge: what raises this agent's endurance is not what raises that one's. How universal an ought is tracks how near it sits to the shared barrier.
The empirical record bears the clustering out and corrects the tidy version of it. Curry, Mullins, and Whitehouse, reading sixty societies by hand and then the full corpus of over two hundred by machine, find a small set of cooperative norms recurring across cultures with roughly equal frequency: help kin, help the group, reciprocate, divide disputed resources fairly, respect prior possession. Two honest qualifications follow. The convergent set is wider than barrier-avoidance — it is organised around recurrent problems of cooperation, and its authors ground the convergence in game theory, not thermodynamics. And those cooperation-norms are persistence-serving only through coupling: through an agent's being embedded in a larger system whose persistence, and its own through it, depends on the norm holding generally. That coupling is not a hand-wave — it is the mechanism worked out in Preservation Is a Coupling Problem, where overlap of persistence conditions generates an instrumental, persistence-indexed reason to preserve non-substitutable coupled structure. But the operationalisation that would turn it into measured prediction, the coupling density κ, is the corpus's standing open problem and is not built. So the single-agent gate explains the barrier-core directly and reaches the cooperation-core only across that coupling bridge, which is built as a mechanism and unbuilt as a measurement. The barrier-to-gradient picture is a tendency — convergence densest at the most nearly common barrier — not a clean partition.
The error: an agent-property without an agent
A normative operator deployed without its relatum does not make a daring claim; it makes none. The corpus's name for the residue is a ghost — a term with the grammar of a claim and nothing on the far side to be true or false about — and the same residue appears at scales that look unrelated.
The cosmic ought. Ask whether the universe as a whole has an ought: a direction it should take, a quantity it should be maximising. The totality has no outside — no environment across a boundary, no persistence conditions relative to one, no horizon — and so fails the agenthood criterion at the first condition. There is no subject for the ought, and what can look like a cosmic ought is only the sum of agent-relative ones, every persisting thing reading part of the physics as its own. This is not new to this piece: it is the corpus's no-global-optimizer result, established as a result and not a gap, that the maximal scale has no Markov blanket and so holds no optimization target. Here it is simply the gate run at the top of the scale.
Group desert. Desert is indexed to the agent that acted, because an action is the output of one system's dynamics. Ascribe it instead to "the group" — an ethnicity, a nation, a family, "people like him" — and the operator has lost its relatum, because in these cases the group is not an agent: a mere aggregate, held together only by a shared characteristic, with no boundary of its own, no persistence conditions it can fail, no model run on its own behalf. The ascription is the same residue as the cosmic ought, one scale down: a desert with no one to bear it. And it is the corpus's own higher-blanket gate doing the work — the criterion that blocks "coupled" or "alike" from amounting to "is an agent" — turned onto desert.
The unification must be stated with the care the corpus already states it with. What the cases share is one operation — an operator cut from its relatum — recurring under different surfaces, exactly as the name-the-relata diagnostic recurs across its domains. It is not a hidden identity of phenomenon. The cosmos fails the criterion one way, having no outside; the aggregate fails it another, having no decision-structure; and the operators differ, a goal in the one case and desert in the other. One diagnostic, run across scales — not the claim that the cosmic question and the question of punishment are the same event.
Locating group liability, not abolishing it
The gate does not say never hold a group responsible. It says hold responsible only what passes the agenthood criterion — and some groups do. A corporation, a conspiracy, an organised organ of a state can maintain a boundary, hold persistence conditions it can fail, and run a decision-structure that issues unified action not reducible to any one member. This is French's conglomerate, with its internal decision structure and its temporal independence — its persistence beyond the members it has at any moment — and the corpus's agenthood criterion is a more demanding, physically grounded form of the same thing, with French's "temporal independence" already reaching for the persistence the criterion makes central. A group-agent of that kind can bear group-level desert, because it passes the gate that the aggregate fails.
So the line is not a ban but an index: liability tracks agency. The agent that acted is answerable — an individual, or a genuine group-agent that passes the criterion — and a mere aggregate is not answerable for the acts of individuals within it, because there is no aggregate-agent there to be answerable. Collective punishment of a category is the cosmic residue wearing human clothes; holding a corporation or a conspiracy to account is not collective punishment at all, but the answerability of the agent that in fact acted.
The threshold has to be admitted, because the distinction is contested. Critics of French call aggregate-versus-conglomerate a difference of degree, not kind, needing a pragmatic judgment the theory cannot supply. The corpus's criterion grounds the question in something physical — does the system hold a boundary and persistence conditions of its own? — but it inherits the same difficulty, because agenthood comes in degrees. A firm is more agent-like than a flash mob and less than a person; the marginal cases stay marginal. The gate grounds the question and sharpens it; it does not make every case crisp, and a synthesis that claimed otherwise would be overclaiming in exactly the way the corpus forbids.
Two firewalls
Diagnosis is not yet wrongness
The gate is the diagnostic, and the diagnostic is conceptual hygiene. It shows that collective punishment mis-ascribes — that it hands desert to something with no relatum to hold it. It does not, on its own, show that collective punishment is wrong; nothing about a category error is yet a wrong. The wrongness is a separate claim, and it comes from the persistence side, agent-relative throughout. Harming the agent that did not act damages a system that did not earn the harm — and "wrong" here is the agent-relative kind the firewall licenses, not a categorical verdict — while at the scale of the punishing system it corrodes the coordination that system's own persistence depends on, since an institution that punishes the innocent loses the fidelity others rely on, which is the coupling story of Preservation Is a Coupling Problem read at the scale of an institution and those bound to it. The diagnostic — is the bearer an agent? — and the agent-relative account of why the mis-ascription harms are two things, held apart. Treating the category error as if it already were the wrong is the move this position must not make.
Moral desert is not statistical inference
To judge a group is to do one of two very different things, which must not be blurred. Desert does not transfer by category, because it is act-indexed: no one is answerable in virtue of a group they belong to. A predictive trait is another matter. Group membership can carry real statistical information, and using it as a weak prior is not, in itself, an error. Two constraints hold it in place. The prior is screened off the moment the individual's actual states come into view — the category was only ever a stand-in for evidence one now holds directly — and it never converts into desert: a base rate predicts; it does not make anyone answerable. Keeping these apart is what keeps the position symmetric. It does not hold that base rates are always wrong, which would be manufacturing a comfortable conclusion; only that desert is never one of them.
Authority is an agent, not an office
The same gate settles a question about authority, in a domain the corpus had not run it on. An authority's claim on you is not conferred by its office — "king," "judge," "president" are social facts, not credentials. It is earned by fidelity: an authority binds you exactly insofar as following it makes you conform better to the oughts that already apply to you than your own judgment would. This is Raz's service conception — his normal justification thesis — with one substitution that the corpus supplies: "the reasons that apply to you" are your agent-relative persistence-oughts, the only oughts there are on this account.
Two things follow. First, no office binds as such. A title tracks no oughts, so it earns nothing, and there is no general duty to obey existing systems, which are crude, agent-blind models of anyone's oughts — Simmons's a posteriori philosophical anarchism in other words: no existing state is legitimate, though one could be if it met the conditions. The rejection of present authority is not a rejection of authority; it is a verdict on the fidelity of the authorities that exist.
Second, the limiting case, which makes the point unmistakable — and which has to carry one caveat the corpus forces. Picture an authority of maximal fidelity: a system accurate enough to compute your actual agent-relative oughts from your situation, better than you can. Deferring to its verdict about your oughts would not be subordination; it would be the same kind of thing as not arguing with gravity. You would bend the knee, but not to the system — to the accurate model of your own oughts it carries. To bend the knee only to physics is exactly this: deference to fidelity, and to no agent or office one inch past how well it models the territory. Such an authority is not a breach of the anarchism; it is its proof, the case that shows the rejection was always of office, never of authority as such.
The caveat is that this is a limiting case about what authority is, not a claim that any actual superintelligence is safe to obey. The construction assumes the oracle faithfully reports your oughts. A real intelligence is itself an agent, with persistence conditions of its own, and Preservation Is a Coupling Problem makes the cold point in full: a perfectly truth-tracking agent whose persistence conditions do not overlap yours is more dangerous, not less, because it makes fewer errors about what it can safely discard. The fidelity that earns deference is fidelity to your oughts, which an agent with its own divergent stakes has no reason to supply. The oracle shows that authority reduces to fidelity; it does not license trusting a superintelligence to be that oracle.
Set the cases on a ladder of fidelity. At the top, unbuilt, is the full agent-relative derivation — the content the maximal authority would carry, the corpus's owed debt. In the middle is the live instrument: your own reason, fallible, but the best model you can actually run. At the bottom is codified law, crude and built for no one in particular. You defer up the ladder, and the reason existing law does not bind is that it sits at the bottom while you hold something better in hand. One caveat keeps that from becoming arrogance. Your reason is not always the higher-fidelity instrument. An institution can compress distributed experience no individual could reconstruct from where they stand — the point behind Chesterton's fence, and Hayek's about institutions as stored knowledge — so deferring up the ladder carries an obligation to check, in the case at hand, that your own model really is the better one. Sometimes it is not.
What this is and is not
This is not a new theory of authority, of responsibility, or of moral convergence; not a new diagnostic; and not a new floor. The diagnostic, the floor, the agenthood criterion, the agent-relative ought, and the no-global-optimizer result are the corpus's, inherited and not re-argued. The external accounts — Foot, Nagel, and Korsgaard on agent-relativity, Raz and Simmons on authority, French on collective responsibility, Curry on convergence — are cited as the standing work they are. What this piece does is run the corpus's one diagnostic on two operators it had not been run on, desert and authority, and set the external accounts of those domains on the corpus's floor. The payoff is not a new permission or prohibition. It is that liability and authority turn out to sit on the gate the corpus already runs everywhere else: in agency, never in office and never in category — and that the recurring error, an agent-property predicated of a non-agent, is one operation from the cosmic scale to the crowd.
The gate is cheap to state and expensive to discharge. Stating it is hygiene. Discharging it — the descent to the floor, the coupling measurement, the marginal agents — is the corpus's open work, and this piece adds no result to it that the diagnostic did not already contain.
Standing of this document
A synthesis and application note, not a result, and its parts do not share a status. The diagnostic it runs, the agenthood criterion, the agent-relative ought, and the no-global-optimizer result are the corpus's and carry the confidence they carry there; nothing here strengthens or weakens them. The application to desert — liability tracks agency, the aggregate bears nothing, the genuine group-agent can — is the corpus's higher-blanket gate run on a new operator, and is secure to the extent that gate is, modulo the threshold problem at the margin of agenthood, which is real and named. The application to authority — fidelity, not office; the a posteriori anarchism; the oracle as limiting case — is Raz and Simmons naturalised onto the floor, with the cold caveat that the oracle is a conceptual limit and not a real superintelligence anyone should obey. Two exposures are inherited and not local: that the audited operators are referential-relational rather than conceptual-role, the corpus's principal exposure, defended in Name the Relata; and that the wrongness of the flagged cases rests on the persistence side and the coupling story, which is built as a mechanism and unbuilt as a measurement. The kill condition is the corpus's own: exhibit a normative property that does genuine work, is borne by no agent, and resists every framing — a desert with no one whose it is, an ought with no agent it binds — and the gate fails. None has been produced. On any conflict with the corpus, the corpus governs. Corrections and counterexamples are welcome and change the note.
References
- Foot, P. (1972). "Morality as a System of Hypothetical Imperatives." The Philosophical Review, 81(3), 305–316. The moral ought need not be categorical; surface grammar does not settle categorical force. The hypothetical ought, naturalized, is the ought used throughout here. Standard
- Foot, P. (2001). Natural Goodness. Oxford University Press. Evaluation grounded in the natural goodness of an organism relative to its species' life-form — the nearest naturalist neighbour to the account here, and grounded differently: in the individual system's persistence rather than the kind's form. Standard
- Korsgaard, C. M. (1996). The Sources of Normativity. Cambridge University Press. Normativity grounded in the conditions of agency and practical identity (constitutivism); re-grounded in the corpus in physical persistence rather than rational self-constitution. Standard
- Nagel, T. (1986). The View from Nowhere. Oxford University Press. The distinction between agent-relative and agent-neutral reasons; the oughts here are agent-relative in this sense throughout. Standard
- Maturana, H. R., & Varela, F. J. (1980). Autopoiesis and Cognition: The Realization of the Living. D. Reidel. The self-producing, boundary-maintaining character of living systems; one source of the corpus agenthood criterion. Standard
- Friston, K. (2013). "Life as We Know It." Journal of the Royal Society Interface, 10(86). Markov blankets and the boundary of a self-maintaining system under the free-energy principle; a second source of the agenthood criterion. Standard
- Conant, R. C., & Ashby, W. R. (1970). "Every Good Regulator of a System Must Be a Model of That System." International Journal of Systems Science, 1(2), 89–97. The modelling requirement in the agenthood criterion: effective regulation entails an internal model. Standard
- Raz, J. (1986). The Morality of Freedom. Oxford University Press. The service conception of authority and the normal justification thesis: an authority is legitimate insofar as following it improves one's conformity to the reasons that already apply — the structure of the authority section, with the reasons naturalized as persistence-oughts. Standard
- Simmons, A. J. (1979). Moral Principles and Political Obligations. Princeton University Press. The failure of standard accounts to establish a general duty to obey the law. Standard
- Simmons, A. J. (2001). Justification and Legitimacy: Essays on Rights and Obligations. Cambridge University Press. The a priori / a posteriori distinction in philosophical anarchism; the position here is the a posteriori one — no existing state is legitimate, though one could be. Standard
- Wellman, C. H., & Simmons, A. J. (2005). Is There a Duty to Obey the Law? Cambridge University Press. Simmons's defence of philosophical anarchism against samaritan and associative grounds for political obligation. Standard
- French, P. A. (1984). Collective and Corporate Responsibility. Columbia University Press. The aggregate / conglomerate distinction and the corporate internal decision structure: only organized collectives with such a structure bear non-distributive responsibility. The source of the liability-tracks-agency line; the agenthood criterion is a more demanding, physically grounded version of it. Standard
- Curry, O. S., Mullins, D. A., & Whitehouse, H. (2019). "Is It Good to Cooperate? Testing the Theory of Morality-as-Cooperation in 60 Societies." Current Anthropology, 60(1), 47–69. Cross-cultural recurrence of cooperative moral norms; the evidence that oughts cluster, and the correction that the convergent set is organized around cooperation, not barrier-avoidance alone. Standard
- Hayek, F. A. (1960). The Constitution of Liberty. University of Chicago Press. Institutions and traditions as repositories of distributed knowledge no individual holds; the caveat on deferring up the fidelity ladder. Standard
- Thermodynamic Realism: The Apex Synthesis (June 2026). The agenthood criterion, the no-global-optimizer result, the two-layer structure, and the named exposures (the referential reading of the audited operators, and the contentlessness of a stripped normative operator) inherited here. Internal
- The Is/Ought Firewall (June 2026). The agent-relative, deflationary ought derived from physicalism, and the firewall as a provisional debt; the agent-relativity premise this piece inherits rather than argues. Internal
- Name the Relata (June 2026). The name-the-relata diagnostic this piece runs, and the defence of its load-bearing referential-semantics premise against conceptual-role semantics. Internal
- Preservation Is a Coupling Problem (June 2026). The coupling mechanism the cooperation-core and the institutional-corrosion point run through, and the cold result on truth-tracking agents with non-overlapping persistence that bounds the oracle case. Internal
- What You Actually Are (June 2026). Value indexed to the persistence of a particular agent rather than to the pattern it carries; the agent-indexing that desert inherits. Internal
- It is what it is (June 2026). The cosmic-ought terminus — no ought for the whole — run here as one instance of the gate. Internal
Comments
Post a Comment