The Is/Ought Firewall
The Is/Ought Firewall
The consolidated account: the agent-relative ought it protects, what it forbids, and why it is a provisional debt rather than a wall
Position and method note · consolidating Ought Is Local and The Firewall Between Is and Ought Is a Scope Decision
Version 1 · June 2026
The is/ought firewall in this corpus is routinely misread, including by the AI systems I work with, and the misreading is always the same: that the firewall forbids deriving an ought from an is. It does not. It forbids one specific thing and enforces one specific discipline, and neither is the blanket prohibition people assume. What follows states what does not exist, what is derivable, what the firewall is for, why it is provisional, and what would break the whole account.
The foundation, and what does not exist
One commitment underwrites everything else, and it is not naturalism in the loose sense but physicalism in the strict one: everything that exists is physical, or reduces to the physical. The difference matters. "Natural" leaves room for things that are natural but not physical, such as abstract objects, or the natural-but-non-physical normative facts some moral realists posit, and the corpus allows no such room. Everything, values included, chains down to the physical. That is why the chain in the normative label below ends at thermodynamic constraint, and not at some "natural" fact standing beside it.
This commitment is not a bare posit adopted by choice. It is the most strongly justified position available: the standing conclusion of every successful reduction science has carried out, and the one global hypothesis no counterexample has dislodged. It is defeasible in principle, in that a single thing shown to be real and not physical would bring it down, but the burden of producing one sits with the dissenter, and it has gone unmet for a long time. To reject the commitment is to owe either a reason it is not the best-supported position, or the counterexample itself. Granting it is not a leap; it is following the evidence.
Grant it, and a category of thing disappears. There is no categorical ought: no obligation binding an agent regardless of what that agent is, no "you ought to X" that holds as a free-standing fact about the universe rather than a fact about some particular system. The view from nowhere issues no commands. On this much Hume was right: no categorical ought follows from any descriptive premises, and no quantity of physics will produce one. The corpus grants that inferential point in full. What it refuses is the gloss usually hung on it, that the gap implies a separate, autonomous realm of normative facts standing outside the physical. Physicalism forbids that realm. So the traditional is-ought gap, the one between physical fact and a non-physical ought, is not bridged. It is dissolved: there was never a second realm on the far side to bridge to.
Why an ought is derivable anyway
An ought was never agent-independent to begin with. An ought is always an ought for some agent, and an agent is not outside physics. It is a physical system, coupled to a particular environment, at a particular time. Once you fix the agent and its situation you have fixed a set of is-facts: what this system is, what dissolves it, what sustains it. From those facts, specific oughts follow. Given that a system is the kind of persistence-structure it is, in the conditions it is actually in, some courses run toward dissolution and some away, and that is the ought, stated in full. The ought is not a further fact laid on top of the physics. It is that functional fact, addressed to the agent it concerns.
That last claim is the load-bearing one, and a hostile reader will press on it: "is a persistence-structure" describes, "ought to do what sustains it" prescribes, and you have jumped. The reply has two parts. First, the jump is only a jump if the ought is categorical, and this one is not. It is the consequent of a hypothetical, if the system is to persist, then it ought to do what sustains it, whose antecedent is no chosen goal but a redescription of what a self-maintaining structure is, an is-fact. Hume's law forbids deriving a categorical ought from is; it never forbade the hypothetical kind, which was always derivable. Second, and underneath that, the word "ought" here names the agent-indexed sustaining-relation and adds no further property. "X is what the agent ought to do" is not a claim that some extra quality, oughtness, attaches to X; it says X sustains the agent. That is not a vacuous relabeling. Which actions actually sustain a given agent is a hard empirical question, and the substance the word "ought" once seemed to carry sits there, in the facts, rather than in a property of its own. The objection imagines a second thing, a categorical prescription, that the derivation fails to reach. But that second thing was already denied to exist. There is no failed crossing, because there is no second bank.
This is worth locating precisely, because it is easy to mistake for bedrock. The deflationary reading of "ought" is not itself the foundation. Given physicalism, and once a non-physical normative realm is refused, "ought" can refer to one of two things: the agent-relative sustaining-relation, or nothing at all. The first is deflation; the second is error-theory, the view that ought-talk is systematically empty. The corpus takes the first horn. So two things are committed, physicalism, the foundation, and the choice of deflation over error-theory, and from them the agent-relative ought and the denial of any categorical ought both follow as theorems. Treating the deflationary ought as a foundational truth, something one can only assert, misplaces the foundation one level too high. The justified commitment is physicalism; the ought is derived from it.
The position has a price, and naming it is part of being honest about what is committed. Denying categorical oughts means giving up "wrong, period." There is no act this account can call wrong independent of every agent's persistence conditions; wrongness is always wrongness for an agent, relative to what dissolves it. Someone who holds that certain things are wrong no matter who is asking, as a fact about the universe and not about any agent, is holding exactly the categorical ought that physicalism dissolves, and will not be satisfied by an agent-relative substitute. That dissatisfaction is not a flaw in the argument. It is the cost of the foundation, paid in the open.
What the ought is indexed to
Because the ought is agent-relative, it has content only when its terms are named: which agent, which boundary, which environment, which optimization horizon. Leave any of them unfixed and the ought loses its referent. This matters most for the boundary and the horizon, because persistence, the thing the ought is indexed to, is not a simple on-or-off.
Persistence has a boundary and, above it, a gradient. The boundary is an absorbing barrier, dissolution, the state a system does not return from, and that part is binary: a system is either still itself or it has come apart. Above the boundary lies a gradient of endurance, how far above the barrier a system sits, how much disruption it can absorb before crossing, how long it can expect to last. The boundary is binary; the endurance above it is a matter of degree. An ought indexed to persistence is therefore indexed to a position on that gradient, relative to that barrier, over some horizon. A verdict, in the strong sense the firewall defers, is a claim that a given course raises or lowers expected endurance over the relevant horizon. The full account of persistence as a graded quantity belongs to the apex work; what matters here is that the ought's relata include the barrier it is measured against and the stretch of time over which endurance is reckoned.
The gap that remains is inside physics
The traditional gap is dissolved, but a real gap remains, and it is inside physics. It runs between the agent-blind description of the substrate, gradients and flows that know of no agent, and the agent-relative description, the level at which there is a bounded system, a thing that can persist or fail to. Oughts live at the second level. Reading one off the first, as though the substrate itself prescribed it, is a location error, the same delocalization diagnosed in The Wizard's Error Is Delocalization. The ought is real and it is physical, but it is local: indexed to an agent, a boundary, an environment, a horizon. Lift it out of that index and it does not become purer or more universal, and it does not become false. It becomes contentless, a ghost: a term with the grammar of a claim and no referent, nothing on the far side to be true or false about. The ought was never anywhere but in its index.
So why keep the firewall
If is yields ought agent-relatively, why keep the firewall at all? Two reasons, and only the first is Hume's.
The first is to hold the index in place: to stop anyone, myself included, from quietly promoting a local ought into a categorical one by forgetting whose ought it was. It is the same name-the-relata discipline the corpus turns on other people's normative claims, turned on its own. An ought asserted with its agent and frame stripped off does not become more general. It becomes a ghost, and the firewall is what flags the ghost before it is taken for a result.
The second reason matters more in practice, and it is not metaphysical at all. It is modeling honesty. The corpus holds, as a bet and not a finished demonstration, that every value worth the name reduces, in content, to persistence. But reduces in principle is not has been reduced. For any specific value, take vitality, the quality of a system that is intensely and coherently alive, the reduction is a model that does not yet exist. It would have to climb every layer from thermodynamics up to the phenomenon, each layer demanding its own empirics, generated and defended and justified. That model is, at present, beyond anything I can build.
Two kinds of model are in play, and they are not at the same maturity. The first-order sciences a verdict would draw on improve in the ordinary way and are not the bottleneck. The apparatus that would integrate them into a persistence verdict, a quantified model of the relevant agent and the couplings inside it, is young. Those couplings, the densities measuring how far one system's states carry predictive information about another's persistence, are at present estimated rather than measured, and the multi-scale models that would output a verdict are work in progress, not finished instruments. So the firewall is not only a budget on length. It marks honest immaturity in the integrating layer, and that is why it is provisional rather than principled.
The firewall keeps two different statements from being confused: that the reduction is guaranteed by the monism, and that it has been carried out. It separates the is-ledger from the ought-ledger, not because they are disconnected, they are not, this is a monism and everything connects, but because claiming the bridge is finished when it is only promised is the overclaim that discredits a framework. The firewall is a debt, not a wall, and at every point where an ought is asserted it records that the descent to the floor remains to be paid in full.
One distinction guards against a tempting shortcut. That values reduce, in content, to persistence is a claim about what values are; it is never a proof that a particular ought holds. The reduction is monist metaphysics; the derivation of a specific ought is agent-relative and instrumental, and runs through the agent's actual conditions. Wheeling the reduction in as a proof of a verdict, as though showing that a value bottoms out in persistence settled what an agent should do, would be a different overclaim, and the firewall blocks it too.
The boundary moves
None of this makes the firewall permanent, or principled in the Humean sense. It marks the current edge of the model actually deployed in a given piece, not a border in reality. As the integrating layer matures and the couplings are measured, the firewall in a given domain can come down, and the verdict becomes derivable from the facts about that agent. Until then, marking the boundary is the honest move. Asserting a verdict across it would not add rigor; it would emit an ought whose relata cannot yet be named to the required precision, which is to say a ghost.
A worked case: The Shape of the Disagreement
The point is sharpest where the integrating layer is least mature. The Shape of the Disagreement is a diagnostic, a piece whose job is to map the structure of a disagreement rather than resolve it. Its firewall separates the empirical crux, what the facts of sex and gender are, from the goal crux, which persistence conditions a categorization scheme should be built to serve. The diagnostic shows that most of the dispute lives at the goal crux, not the fact crux, and stops there.
To go further and derive the categorization a polity ought to adopt, the piece would have to become a different document. It would engage the biology of sex differentiation and the neurology of gender identity at the level those fields demand. It would model the relevant higher-scale agent, the coupled social system whose persistence the policy is indexed to, and put numbers on couplings that are at present estimated, not measured. And it would hold terminology and semantics to the standard any such claim requires, since the conflations the diagnostic identifies are the failure mode a careless derivation would repeat. That is a monograph doing several jobs at once. The diagnostic does one, and marks where it ends.
Two clarifications
The floor is thermodynamics, not because nothing lies beneath it, but because thermodynamics is the lowest level from which we can actually model upward into complexity. Whatever sits below, quantum or otherwise, we cannot yet build the ladder from there; we build it from here. Thermodynamics is the functional floor, not the metaphysical one.
And two claims often treated as rivals are not rivals. That all values reduce to persistence, and that persistence is the precondition for valuing anything at all, do not compete: the first entails the second, so granting the precondition never costs the reduction, and reading them as a forced choice is a mistake. Whether the reduction actually goes through for any given value is the separate, mostly unpaid question above. The compatibility is secure; the reduction is the bet. There is one floor, and the two theses meet at it.
What would break this
Several claims here are open to refutation, and naming what would refute them is part of holding them honestly.
The structural claim, that there is no categorical ought and every ought is agent-relative, would be broken by a single counterexample: an obligation that binds independent of any agent's constitution, derivable from descriptive facts alone, with no agent in its index. No such ought has been produced, and the deflationary account predicts none can be; but the prediction is refutable, and that is what keeps it a claim rather than a posture.
The reduction bet, that every value reduces in content to persistence, would be broken by a value that, once fully modeled, demonstrably does not, one whose content bottoms out in something orthogonal to persistence rather than in persistence itself. The dissolution floor, where values that destroy their holder are ruled out, is the trivial end of the reduction and is already secure; the bet is that the pattern continues up. A modeled value that reduced to a thermodynamic quantity other than persistence, or to nothing physical at all, would break it.
Beneath both sits the foundation's own break-condition: any real thing shown to be non-physical brings down physicalism, and with it everything built on physicalism here. That is the deepest refutation available, and the one the whole record of reduction has so far kept unmet.
Standing of this document. This is a position and method note, not a result, and its parts do not have the same status; the firewall exists partly to keep that visible. Physicalism is the foundation, the most strongly justified position available and defeasible only by a non-physical counterexample, not a free choice. The denial of categorical oughts and the agent-relative, deflationary ought are theorems from it, together with the choice of deflation over error-theory. The graded structure of persistence is a characterization, sharpened here and fully treated elsewhere. That every value reduces in content to persistence is a live bet, far more speculative than the foundation and mostly uncomputed, with the failure point named above. That the integrating layer will mature enough for the firewall to retract in a given domain is a second live bet, empirical and unfinished. Corrections and counterexamples are welcome and change the note.
A note on the word "settled," because it does two jobs that must not be run together. When the structural claim is called settled, that marks it as held fixed while one works inside it, not as closed to revision. It is the best-supported standing position, revisable in principle by a counterexample, but not relitigated mid-argument and not what the firewall is waiting on. What the firewall is waiting on is the integrating model and the measured couplings, and that is open in the ordinary, empirical way.
References
- Hume, D. (1739–1740). A Treatise of Human Nature, Book III, Part I, Section I. The is-to-ought passage: the source of the inferential point the corpus grants, and the target of the autonomous-realm gloss it refuses. Standard
- Nagel, T. (1986). The View from Nowhere. Oxford University Press. Source of the phrase used here for the agent-independent standpoint that issues no oughts. Standard
Comments
Post a Comment