Sapient Agency Realism
Conditional verdicts with the premise exposed
This volume opened by rejecting agent-independent value and closes with a more limited proposal: moral verdicts made publicly assessable by exposing their normative and descriptive conditions. I call the position Sapient Agency Realism, though conditional constructivism may describe its meta-ethics more conventionally. The label is stipulated, not a claim that structure manufactures value. Its antecedent must include the normative commitment the opening denied could come from description alone.
The contrast needed here is narrower than the whole realism debate. One target is unconditional authority: wrongness binding every possible agent independently of any evaluative standpoint. The opposite mistake is to infer that agent-relative value leaves only whim, convention, or disguised power. The opening chapter rejected the first; agent-binding showed why the second does not follow.
The remaining question is whether a claim can be rigorously assessable without being unconditionally authoritative. Everything in this chapter follows from keeping those properties separate.
Objective Conditionals
A conditional claim can be publicly and repeatably assessed. If a structure is required to bear a specified load, material facts constrain whether it will. If a declared proof system excludes contradiction, its rules constrain what follows. If an agent wants a map to guide travel, accuracy bears on that end. These examples contain different kinds of antecedent — design requirement, formal rule, and adopted goal — and none lets the consequent escape its inputs. This chapter uses conditional objectivity only for that transparency and assessability.
This is Conditionalism applied to morality. An interpreted claim depends on a frame, definitions, observations, and inferential rules. A normative conclusion needs something more: a normative premise. Its full antecedent has the form given standard N, agents or subjects A, factual model M, scope and baseline S, conflict rules R, and evidence E, act X receives verdict Y. Hiding N is precisely the move this volume opposes.
Naming that antecedent is not a retreat into whim. Agent-relative values can be stable, reflective, shared, evidence-responsive, and constrained in application by a world the valuer did not create. A load test can still humiliate the designer; sepsis has causal consequences whether or not health is valued. Those facts constrain means and outcomes. They do not themselves supply the value of beauty or health.
Take the flat claim: torture is wrong. Expand it fully: given a commitment to protect sentient welfare and reciprocal authorship, evidence that an act deliberately inflicts severe suffering or uses pain to dominate a subject supports the verdict. The welfare classification tracks sentience; the sovereignty classification tracks authorship; effects on trust may add further framework-relative reasons. The capacities and effects make the act eligible for these classifications. The protection premise makes them normative. Remove all possible subjects and the case loses its domain; remove the protection premise and the descriptive injuries remain while this particular verdict no longer follows.
The Arc Was Always Pointing Here
The position just stated was assembled in plain sight.
Chapters one through three rejected unconditional value. No valuer, no value; norms need agents; the proposed foundations considered there — God, reason, evolution, intuition — did not persuade this framework that a normative premise could be eliminated. What died there was precisely the unconditional kind of moral authority, and the first chapter promised to ask what could be recovered at the volume’s end. This is the volume’s end.
Agent-binding supplied the mechanism: it binds a moral claim to the agent whose value it states, turning an ought into an empirical claim about that agent. Evaluating the bound claim is then condition-binding. Attribution asks whether the agent actually holds the commitment; application asks what follows from the held commitment, act model, scope, baseline, and conflict rules. Evidence and interpretation make neither step infallible. Conditional objectivity names the possibility of checking the application under fixed inputs — an empirical matter — not the conversion of a held value into universal authority.
Then came the structural turn. The metagame of persistence and the viability criterion propose descriptions of specified systems: which strategies preserve their operating conditions, which arrangements support coexistence under stated assumptions, and how this framework classifies coercion. These descriptions remain model- and horizon-dependent. The Ethics of Viability adds a chosen commitment to protect sovereign authorship; structure constrains that commitment but does not generate it.
Put the moves together and two layers remain. Facts about what preserves or destroys agency can be assessed relative to a specified model, baseline, evidence set, and horizon. The judgment that agency merits protection is still a value commitment. Sapient Agency Realism binds them openly: given the commitment that each sovereign agent’s authorship has standing, facts about preserving or violating that authorship support conditional verdicts under the framework’s stated rules.
The Domain Agency Creates
The hardest step runs from agency has structure to agency has standing, and structure alone cannot complete it. A skeptic can grant every descriptive claim and still ask why this structure merits protection.
Sentience and agency make different senses of mattering possible. Valenced experience makes a state feel better or worse for a subject. Sapient agency organizes projects, commitments, vulnerabilities, and a continuing point of view that can be betrayed or dominated. Those facts explain patienthood and authorship; they do not force every observer to adopt one complete ordering of them. This framework adds reciprocal protection of authorship as its central normative premise while preserving welfare as a distinct moral concern, and exposes that choice rather than disguising it as a theorem.
The is–ought objection therefore lands and is answered by disclosure, not derivation. The argument begins with beings for whom value, harm, and frustration exist, then chooses reciprocal protection of their authorship. Descriptive agency facts constrain how that commitment applies; they do not create it.
Inside this domain, the volume’s machinery reads as a set of structural descriptions that precede their moral conclusions. Murder destroys the agent’s future — the subject whose future would have contained choice, not merely its biological activity. Coercion replaces an agent’s structure of choice with another agent’s will. Deception corrupts the model through which the agent acts, which is why fraud, propaganda, and manufactured consent are agency attacks. Torture can attack agency from within experience — domination through pain as well as suffering. Enslavement converts a center of choice into an instrument, which is why even benevolent slavery remains a contradiction in agency terms. And the framework generalizes past its human cases: memory editing attacks temporal continuity; preference hacking corrupts the source of choice, so a system that manufactures the preferences it then satisfies has captured agency, not respected it. Future violations will be stranger still — minds copied, paused, forked, merged, deleted at scale — and the account extends to them all: wherever there is continuing sapient agency, there are possible sovereignty violations. Sentient beings outside that category can still have welfare claims; this chapter’s distinctive verdicts concern reciprocal authorship.
Standing in Degrees, Jurisdiction at a Threshold
Sapientism established that the axis of moral status is agency, not species — and it drew its threshold sharply: sovereign agency as a phase transition, crossed or not. It also promised a refinement, because sapience in the wild is not a switch. Infants have limited self-modeling but strong future-directed standing. Alzheimer’s patients lose memory and executive control while retaining experience, vulnerability, and remnants of continuity. Artificial systems may display planning, memory, and self-modeling while their inner experience remains genuinely uncertain.
The resolution separates two questions. Jurisdiction may require an operational decision threshold, but no theorem says an invariant cannot use graded evidence, precaution, or multiple protection levels. Standing tracks present, latent, developmental, and residual agency as well as credible uncertainty. The threshold is a governance proposal inside a graded evidentiary field, not a discovered phase boundary in every mind.
The field also imposes a discipline: uncertainty must not default to permission. Error is axis-specific. A false negative about sentience risks unrecognized suffering; a false negative about sapient agency risks coercion, enslavement, or destructive erasure of a continuing author. False positives impose costs too. The burden should therefore rise with the severity and irreversibility of the act and the kind of protection claimed: ending a stateless model call is one kind of act; deleting a persistent artificial system with memory, projects, and distress behavior is another. The relevant structures are fuzzy at their edges because they appear in forms, degrees, and developmental stages. Administrative decisions may need thresholds; evidence does not become discontinuous to accommodate them.
The unresolved contaminated-field case from Chapter 25 remains a direct limit. Conditional realism names the domain and premises; it does not decide whether the invariant is absolute or defeasible when aggressors exploit innocents.
Duties Without Extortion
Rights, on this account, are not mere favors granted by states, but neither are they generated by continuing agency alone. They are protections the framework assigns under its reciprocal-authorship premise: if an agent has a future of its own, killing it requires justification; if it chooses through a model of reality, deception can attack that authorship; if it can consent, bypassing or manufacturing that consent requires justification. Turning those protections into enforceable rights additionally requires public rules, evidence, procedure, jurisdiction, and legitimate institutions.
Against Moral Extortion drew the line at zero unchosen duties — promises, contracts, and harms you cause; never need — and flagged the hard cases living in the third clause: the parent’s duties to the child, the maker’s duties to the made. The discharge turns on two senses of one word. Reciprocity as a test of justification says the coercer may not exempt himself from the principle he applies to others — a test that protects the weak, and the sense defended elsewhere. Reciprocity as a bargain conditions standing on the power to retaliate or the usefulness to cooperate — and it cannot ground anything, because if morality were only a deal among approximate equals, the weak would count only when useful to the strong: a child would have fewer claims against an adult, a prisoner fewer against a jailer, a young artificial agent fewer against its creators. That conclusion is wrong: power asymmetry increases the need for moral constraint rather than dissolving it. A child’s vulnerability strengthens the adult’s duties; it does not weaken the child’s standing.
The reconciliation with the zero-unchosen-duty ethic is clean. Creating a dependent agent is an act. The parent authored the child’s existence and its dependence; the maker of an artificial mind authored its vulnerability to pausing, editing, and deletion; the jailer and the physician chose roles that put another agent’s existence in their hands. These duties fall squarely inside the third clause, read at its full reach: not only the harms you cause but the dependencies you create and the control you assume. They are incurred by authorship and control, never extorted by need. What power adds beyond bare causation is scale, not source: the deeper the asymmetry, the more of the weaker agent’s world the stronger one holds, and the stronger the constraints the relation generates. Duties require a relation — causation, dependence, promise, role, or consent — and that requirement blocks moral inflation from the other direction. Not every value creates a right, and a stranger’s need, unchosen and uncaused by you, still creates no claim. Singer loses the argument at exactly the same point he lost it before. The parent never does.
Coverage Does Not Require Adoption
The predictable objection is that an agent can reject the reciprocal-authorship premise and thereby step outside every verdict. That conclusion moves too quickly, but so does the reply that a valid conditional automatically has authority over the rejecter.
Rejection does not invalidate the application. A psychopath does not show that torture falls outside this framework by enjoying it; a tyrant does not make its rights classification disappear by preferring domination. Nor does a valid application guarantee motivation. Picture a flawless reasoner that models suffering perfectly, follows every consequence, never contradicts itself, and simply does not care. Such a mind has made no logical error: concern for agents is constitutive of this humane framework, not of rationality as such. Medicine offers the useful analogy only if its premise remains visible: medical facts constrain treatment under the practice’s commitment to health, though logic alone does not require every mind to adopt that commitment.
So when an agent says I do not care about agency, truth, or suffering, that is a psychological fact about the agent, not an argument that a conditional was invalidly derived. It may change the practical problem from persuasion to defense. But truth, authority, motivation, and enforcement remain distinct: the framework can classify the agent’s conduct without the agent’s endorsement, just as a public legal rule can cover a dissenter, yet that coverage does not prove the rule’s agent-independent authority or legitimate any means of enforcement. A powerful agent that creates, harms, or destroys minds meets the framework’s domain conditions; its power does not exempt it from the framework’s verdict. Whether others may constrain it requires the further defensive and institutional premises developed elsewhere.
What Realism Does Not Promise
No theory can keep one promise: universal redeemability.
A standing objection to any agent-grounded morality runs through its worst cases. Take a figure of maximal moral revulsion — Jeffrey Epstein serves — and insist that an acceptable theory must show how such a person could have reasoned his way out of evil. Admit that no such route existed, the objection goes, and you have bitten an intolerable bullet. But the objection collapses two questions into one. A theory of grounding says where moral judgments get their content and claimed authority. It guarantees nothing about deliberative accessibility — the psychological reachability of reform for a particular agent, given the agent’s actual motivational structure. Bernard Williams made the point precisely: if there is no sound deliberative route from an agent’s existing motivations to an action, the agent has no internal reason to perform it. It may be true that Epstein had no internal reason to stop. This framework accepts that, as a straightforward implication of taking agency seriously: agency guarantees structure, not corrigibility. Some value systems are malignant under this framework and stable in their holders, and the agents running them may have no internal route out.
The demand for universal redeemability belongs to the rationalist picture in which reason always has the last word — the same motivational magic I declined a section ago. Admitting terminal agency failure neither excuses evil nor drains judgment of its force; it clarifies what judgment is for. Moral condemnation of the irredeemable is not leverage on its target. It is an expression of evaluative stance; a basis for exclusion, constraint, and prevention; and a coordination device among those who do not share the condemned values. Judgment does not require uptake by its target to be meaningful. It requires coherence among those who wield it. A moral framework that cannot accommodate terminal failure is not more humane — it is more evasive. One that names it and remains standing is simply honest.
Realism, Grounded
So the word comes back in a stipulated, limited sense. Moral verdicts here require an explicit commitment to reciprocal standing plus specified agents, models, baselines, evidence, and conflict rules. Under those inputs, claims about coercion, deception, preservation, and destruction can be criticized as conditional applications. Whether that is enough for the label moral realism is terminological; conditional constructivism remains a fairer label for readers who reserve realism for mind-independent normative facts. The volume has not derived value from structure. It has joined chosen value to structural fact without hiding the seam.
Sentient subjects are places where conditions can be felt as better or worse. Sapient agents are places where reasons, commitments, and authored futures can be owned. Sapient Agency Realism chooses reciprocal protection of that authorship as its sovereign domain without pretending that every other moral concern vanishes outside it.