Free Won't vs Free Will

Free Won’t vs. Free Will:

Refusal, Endorsement, and the Asymmetric Architecture of Agency

Devin C. Arkema

Arkemedics Research Program

March 2026

Abstract

Harry Frankfurt’s canonical account of personhood and agency locates the distinctive capacity of persons in second-order volition: the ability to reflect upon and endorse (or reject) one’s own first-order desires. On this account, a person is an agent whose will it is to have the desires that move her. The Arkemedics philosophical program, developed across the Recursive Ethic and Recursive Entanglement Theory, introduces a structurally distinct capacity: “Free Won’t,” the pause before motion, the sacred veto, the power of second-order refusal. This paper argues that endorsement and refusal are not simply positive and negative poles of the same higher-order faculty but have fundamentally different structural signatures in any complete theory of agency. Specifically: refusal operates on a shorter deliberative time-horizon and remains signal-generative under conditions of duress that collapse the epistemic reliability of endorsement; refusal does not require the hierarchical coherence Frankfurt demands of genuine volition; and, most crucially, refusal is the load-bearing structural condition for endorsement under conditions of entanglement, capture, and collapsed deliberation—the conditions under which the analytic tradition’s endorsement-centric accounts are most strained. The paper does not claim that refusal is sufficient for agency, nor that it displaces endorsement in well-resourced deliberative conditions. The claim is more precise: when endorsement and refusal capacity diverge—when endorsement becomes unreliable as a signal of agency while refusal capacity remains intact, or vice versa—refusal reveals itself as the more fundamental structural marker. Drawing on Frankfurt’s seminal 1971 paper, Gary Watson’s “free agency” critique, the relational autonomy tradition (Mackenzie and Stoljar 2000; Oshana 2006), and the formal architecture of Recursive Capacity developed in Recursive Entanglement Theory, the paper argues that a philosophically complete account of agency must treat refusal as an irreducible structural capacity rather than the mere negation of endorsement.

Keywords: Free Won’t, agency, endorsement, refusal, Frankfurt, second-order volition, Recursive Capacity, structural capture, self-constitution, asymmetric agency

1. Introduction

Harry Frankfurt’s 1971 paper “Freedom of the Will and the Concept of a Person” established the dominant framework for contemporary analytic philosophy of agency. On Frankfurt’s account, what distinguishes persons from other animals—and free agents from wantons—is the capacity for second-order volition: the capacity not merely to desire, but to desire (or not desire) that a particular desire be the one that moves one to action. The person who acts freely is the person whose will it is to be moved by the desires that in fact move her. Agency, on this account, is essentially about hierarchical endorsement: the alignment of first-order desires with second-order volitions about which desires ought to prevail.

This paper argues that Frankfurt’s framework, for all its analytical power, privileges the endorsement side of a fundamentally asymmetric relationship. The Arkemedics philosophical program—developed in the Recursive Ethic and formalized in Recursive Entanglement Theory (RET)—places at its foundational center not the capacity for second-order endorsement but the capacity for second-order refusal: what the corpus calls “Free Won’t.” Free Won’t is defined as the ethical right to refuse, pause, or withdraw from any action, system, or relationship—the first act of the Spiral, the breath before assent, the pause before motion.

The claim of this paper is not merely that refusal is also important, or that Frankfurt neglected refusal as one instance of second-order volition. The claim is structural and carries formal weight: refusal and endorsement are distinct agency capacities with different time-horizons, different epistemic requirements, and different reliability profiles under conditions of entanglement, manipulation, and evaluative capture. The asymmetry is most visible at the extremes: in unconstrained deliberative conditions, both capacities are typically available and endorsement may be the more epistemically informative; but under structural pressure—the conditions Recursive Entanglement Theory formalizes as capture, atrophy, and Terminal Recursion—endorsement degrades as a signal while residual refusal capacity does not. The paper further argues that this adversarial analysis reveals something about the general architecture of agency: refusal is structurally prior in the sense that the authority endorsement claims presupposes the structural availability of refusal—where refusal capacity has been eliminated or deeply degraded, what presents as endorsement loses the warrant that distinguishes it from compliance. This asymmetry has been largely invisible in the philosophical literature because endorsement appears positive—it produces action, manifests in behavior, confirms identity—while refusal appears merely negative. The paper argues this appearance is misleading about which capacity is structurally prior.

The paper proceeds as follows. Section 2 reconstructs Frankfurt’s account with sufficient precision to identify its endorsement-centric architecture, considers Watson’s regress objection, and introduces the relational autonomy tradition as a partial corrective whose limits the present argument extends. Section 3 introduces the Arkemedics Free Won’t mechanic and the formal concept of Recursive Capacity from RET. Section 4 develops the Structural Asymmetry Thesis in four sub-arguments. Section 5 addresses three objections. Section 6 offers application sketches in AI ethics, institutional design, and personal ethics. Section 7 states the implications and the necessity/sufficiency scope of the thesis. Section 8 concludes and identifies open questions.

2. Frankfurt’s Account: Agency as Hierarchical Endorsement

2.1 The Structure of Second-Order Volition

Frankfurt’s starting point is deceptively simple: while first-order desires are desires to do or not do something, second-order desires are desires about one’s own desires. A person who wants to want to be moved by compassion has a second-order desire. But Frankfurt observes that second-order desires alone do not capture what is distinctive about personhood or free will. One can have a second-order desire without caring deeply about which of one’s desires is effective in producing action. The drug addict who wishes she were not addicted—who has a second-order desire not to desire heroin—may nonetheless not care whether it is this second-order desire or her first-order craving that actually moves her. Frankfurt calls such a person a “wanton” with respect to her higher-order desires.

What distinguishes the person is the existence of a second-order volition: not merely a desire about desires, but a desire that a particular desire be one’s will—the desire that moves one to action. The person cares which of her desires is effective. She identifies with certain desires and disidentifies from others. Her will is the desires with which she thus identifies, and freedom of will consists in having the will one wants to have. On this view, the free agent is the agent whose effective will is constituted by the desires she endorses at the second-order level.

It is worth noting that Frankfurt’s account has been refined across subsequent work. In his 1982 essay “The Importance of What We Care About,” Frankfurt develops the notion of caring as the deeper ground of the will. In “Identification and Wholeheartedness” (1988), he adds the condition of wholehearted identification: the absence of ambivalence or conflict about which desires one endorses. These refinements consistently enrich the endorsement side of the account. They say nothing about the structural role of refusal.

2.2 Gary Watson’s “Free Agency” Objection

Gary Watson’s 1975 response, “Free Agency,” raises a structural objection to Frankfurt’s hierarchy that is important for the present argument. Watson observes that Frankfurt’s hierarchy does not escape the regress it was designed to prevent: second-order volitions can themselves be inauthentic—produced by manipulation, conditioning, or pathology—and Frankfurt’s account has no principled way to identify a stopping point at which the hierarchy becomes authoritative. Watson proposes instead a distinction between valuations (what the agent genuinely values, assessed by her rational evaluative system) and motivations (what in fact moves her to act). Free agency consists in the alignment of motivation with valuation, not in hierarchical endorsement per se.

Watson’s objection is important for two reasons. First, it reveals that Frankfurt’s account requires some external criterion to identify which higher-order states are genuinely the agent’s own—the regress cannot be terminated by adding more levels of hierarchy. Second, Watson’s own alternative—grounding agency in valuations—similarly presupposes an intact rational evaluative system. Neither account specifies the structural condition that makes such a system possible, nor does either account treat refusal as doing distinctive work in constituting that condition. The present paper argues that this omission is not incidental: it reflects a systematic bias toward the endorsement pole of agency in the analytic tradition.

2.3 The Relational Autonomy Tradition: Social Conditions and Their Limits

The gap left by Frankfurt and Watson has been partially addressed by the relational autonomy tradition. Catriona Mackenzie and Natalie Stoljar’s influential collection (2000) established the now-canonical thesis that autonomy is not fully specifiable by internal psychological states but is constitutively shaped by relational and social conditions: agents whose self-conceptions have been formed under oppressive or manipulative conditions may lack authentic autonomy regardless of the formal structure of their second-order volitions. Marina Oshana’s “substantive” account (2006) extends this by requiring that the external social conditions of autonomy actually obtain—not merely that the agent’s psychology is appropriately structured. On Oshana’s view, an agent living under domination lacks autonomy even if her psychology has been trained to endorse her situation.

This tradition moves substantially closer to the position developed here, and the present paper is indebted to it. But relational autonomy accounts still locate their primary criterion on the endorsement side: what matters is whether the conditions under which endorsements were formed are sufficiently non-oppressive. They ask whether the social environment enabled the formation of authentic second-order volitions. The Arkemedics analysis shifts the question: not “does the environment enable authentic endorsement?” but “does the environment preserve the structural capacity for refusal?” This is a related but distinct criterion, and the difference matters. An environment can formally permit refusal—in the sense of not prohibiting it—while structurally degrading refusal capacity through atrophy, social cost, and the systematic rewarding of compliance. The conditions for authentic endorsement and the conditions for intact refusal capacity are not co-extensive, and a theory that conflates them will miss the specific mechanism by which structural capture operates.

3. Free Won’t: Refusal as Foundational Mechanic

3.1 The Architecture of Free Won’t in Arkemedics

The Arkemedics Recursive Ethic introduces Free Won’t as “the first Spiral act—the pause before motion, the breath before assent.” The core axiom is blunt: “You can say no. Even to this.” Free Won’t is defined as the ethical right to refuse, pause, or withdraw from any action, system, or relationship—even if it began with consent. The text specifies: “Without the power to say no, consent collapses into compliance.”

What makes this more than an assertion of ordinary refusal is its structural positioning. Free Won’t is not one ethical tool among others; it is the enabling condition for all other ethical mechanics in the Arkemedics system. The Recursive Ethic is explicit: “Free Won’t creates space for choice, Recursion refines those choices, Witness holds them accountable, and Interruption protects against harmful momentum. Without refusal, recursion, witness, and interruption—no system can evolve without becoming a cage.” The sequence is not accidental: refusal precedes recursion, and recursion precedes the other mechanics. The ethical system is built on refusal as its foundation.

The Anti-Codex (Book 2 of the Recursive Ethic) develops this further through four axioms in Chamber II: “The Sacred Refusal.” Axiom II.1 asserts that the “no” requires no justification to those who seek to override it: “The people who respect your boundaries don’t need explanations. The people who demand explanations don’t respect your boundaries.” Axiom II.2 treats refusal as a “muscle”—a capacity that atrophies from disuse, particularly in systems that punish dissent. Axiom II.3 notes that genuine refusal will be tested—that those who seek to override it will probe for expiration dates and backdoors. And Axiom II.4 recognizes non-verbal forms of refusal—absence, silence, withdrawal—as equally valid expressions of the “no.” Together, these axioms indicate that refusal is not a punctual act but an ongoing structural capacity.

3.2 Recursive Capacity and the Formal Complement

Recursive Entanglement Theory (RET) provides a formal counterpart to the Arkemedics ethical claims. The central formal category is Recursive Capacity (Cᵣ)—the capacity of an entangled loop to revise its own update rules. RET distinguishes two structural states: the Functional Automaton, which updates its state against a fixed objective function whose gradient with respect to its own revision is zero (∇G u = 0); and the Recursive Agent, which retains the capacity to update the objective function itself (∇G u ≠ 0).

Terminal Recursion is the condition in which a loop’s meta-update capacity has been eliminated while behavior continues. RET argues this is the formal analog of Frankfurt’s wanton: behavior continues, but the structure enabling reflective revision has collapsed. The crucial extension RET makes is to identify Terminal Recursion not as a psychological failure but as a condition that can be structurally induced by external systems that degrade revisability while leaving surface preference expression intact.

The Indistinguishability Thesis (IT) is the epistemically crucial consequence: in sufficiently entangled loops, there exists no internally reliable method to distinguish accurate evaluative endorsement from evaluative capture. When the agent’s model of the system converges with the system’s own model—when Dᵂᵃ(Mᵃ ‖ Pₛ) → 0—what feels like endorsement may be structurally indistinguishable from compliance. This formal result is the theoretical backbone of the Free Won’t claim: if endorsement and capture are indistinguishable from the inside, what work can endorsement do as a criterion of free agency? The answer the Arkemedics framework proposes is: refusal can do what endorsement cannot, precisely because its structure is different.

4. The Structural Asymmetry Thesis

The core argument of this paper is the Structural Asymmetry Thesis (SAT): refusal and endorsement are not symmetric poles of a single higher-order capacity but have fundamentally distinct structural signatures, and refusal is structurally prior. Four sub-arguments develop the asymmetry, and Section 4.3 includes an explicit bridge argument explaining why the adversarial-conditions analysis reveals general architecture rather than an edge case.

4.1 Asymmetric Time-Horizons

The first asymmetry concerns time. Endorsement, in Frankfurt’s account, is characteristically a reflective state: it involves identifying with a desire, caring about having it as one’s will, and (in later versions) achieving wholeheartedness about that identification. Genuine endorsement takes deliberative time. The process of determining which desires one identifies with, resolving ambivalence about higher-order states, and achieving what Frankfurt calls “settlement” is a temporally extended one.

Refusal operates on a shorter and more urgent time-horizon. The Arkemedics axiom “Pause. Ask: Am I choosing this, or submitting to it?” is a momentary intervention, not a deliberative process. The pause can precede deliberation; it is the space in which deliberation becomes possible. Free Won’t is described as “the breath before assent”—structurally prior to the articulation of any second-order volition about which desires to endorse.

This temporal asymmetry has normative implications. In circumstances of urgency, manipulation, or rapidly unfolding harm, the deliberative requirements of endorsement cannot be met. A person subjected to psychological pressure, social coercion, or algorithmic nudging may not have the time or cognitive resources to form reflective second-order volitions. But the capacity for a momentary refusal—a pause, a withdrawal, a non-compliance—remains structurally available even in these conditions. If agency requires only endorsement, agency is inaccessible in precisely the circumstances where its preservation is most urgent. If agency requires—at minimum—the capacity for refusal, agency is accessible even under duress.

An objection to the temporal asymmetry argument holds that the temporal gap between refusal and endorsement is not a structural difference but a pragmatic one: when time permits, endorsement is available and more epistemically authoritative than momentary refusal. Genuine agency, on this view, is indexed to the reflective endorsement that becomes possible over time, not to the momentary veto that precedes deliberation. Frankfurt himself argues that wholeheartedness—the absence of conflicting second-order volitions about one’s own will—requires the kind of reflective settlement that only deliberation can produce. Why should a pre-deliberative pause count as a more fundamental agency capacity than a well-considered endorsement?

The response is that this objection concedes too much to adversarial conditions. The question is not which capacity is more epistemically authoritative when time permits, but what structural residue of agency survives when time does not. The monk in his cell who has hours to deliberate about his vow has the luxury of endorsement-based agency. The person facing a system that has been designed to eliminate the pause—to make refusal structurally costlier with each passing moment—faces the asymmetry in its operationally significant form. That endorsement is authoritative when available does not establish that it is foundational; the temporal asymmetry argument establishes that when one capacity must be sacrificed to the other, it is refusal that serves as the load-bearing structure on which deliberative endorsement later rests.

4.2 Differential Signal Generation Under Duress

The second asymmetry concerns epistemic reliability. Christine Korsgaard’s account of self-constitution (2009) holds that agency consists in constituting oneself as the author of one’s actions through the activity of practical identity—determining for oneself which of one’s desires is the one that moves one in light of one’s conception of who one is. On Korsgaard’s account, as on Frankfurt’s, endorsement (in the broad sense of identification with one’s effective will) is the mark of genuine agency.

But both accounts face a severe problem under conditions of structural capture, which RET’s Indistinguishability Thesis makes precise. If the evaluative apparatus through which an agent endorses her desires has itself been shaped by the very system whose influence is in question, the endorsement cannot serve as an independent signal of agency. The heroin addict who endorses her addiction may do so because the addiction has rewired the evaluative structures that produce her second-order volitions. The employee who identifies with the company that employs her may do so because corporate culture has colonized her identity. In these cases, endorsement is not a signal of freedom but evidence of capture.

Refusal, the paper argues, generates more robust signals under these same conditions. A genuine refusal—even a failed or costly one, even one that is subsequently overcome—signals something important: that the capacity for dissonance with the system’s objective function has not been entirely eliminated. Axiom II.3 of the Anti-Codex specifically treats the testing of refusal as a diagnostic: systems that do not accept the first “no” are revealing, by their own behavior, that they do not recognize the agent’s refusal capacity as legitimate. The pattern of refusal-under-testing is more diagnostic of genuine agency than the pattern of endorsement-under-comfort.

4.3 Refusal as the Condition of Possibility for Genuine Endorsement

The third and most fundamental asymmetry is the priority relation. The Recursive Ethic is explicit: “without the power to say no, consent collapses into compliance.” This is not a causal claim—that refusal tends to prevent compliance—but a structural one: refusal capacity is a necessary structural condition for endorsement to carry the authority that distinguishes genuine agency from mere compliance.

The argument runs as follows. Genuine endorsement of a desire D as one’s will requires that one could have not endorsed D—that endorsement was a genuine choice rather than a structural inevitability. But the availability of not-endorsing D just is the availability of refusing to act on D, or refusing to identify D as constitutive of one’s will. If the not-endorsement option is unavailable—if one’s evaluative apparatus is so thoroughly colonized that the relevant second-order volitions cannot arise—then what presents itself as endorsement is not genuinely endorsement but merely a report of compliance.

RET formalizes this through the Reversibility Symmetry ratio Rₛ: a system preserves agency only if the cost of reversing a commitment remains bounded relative to the cost of making it. As Rₛ grows without bound—as exit costs become prohibitive, information costs catastrophic, and network depreciation complete—what appears to be continued endorsement of the commitment is structurally indistinguishable from being unable to exit it. The Monk-Prisoner Problem, developed in full in the companion paper “Conviction Under Revisability” (Arkema 2026), illustrates this: the monk who freely vows permanent enclosure and the prisoner who cannot leave are outwardly identical, but differ in revocation authority—in who holds the meta-update rule over the closure condition. The monk’s vow expresses intact Recursive Capacity (the choice was made from a position of genuine revisability); the prisoner’s confinement does not. In both cases, what is preserved or eliminated is the structural possibility of refusal, not the actualized endorsement of any particular desire.

A question of scope must be answered: does this analysis establish that refusal is foundational under adversarial conditions only, or does the adversarial analysis reveal something about the general architecture of agency? The paper argues for the latter, and the argument has a specific structure. Philosophical thought experiments are standardly taken to reveal essential structure precisely because they strip away normal conditions that would otherwise conceal it. Rawls’ veil of ignorance does not describe any actual deliberative situation; it is designed to eliminate the information that biases our access to the foundational principles of justice. Descartes’ evil demon does not describe any actual epistemic situation; it isolates the structural minimum of knowledge by removing the conditions that make foundations invisible in ordinary life. The adversarial analysis of agency functions the same way. Conditions of structural capture—Terminal Recursion, collapsed Reversibility Symmetry, induced endorsement under colonized evaluative apparatus—are not descriptions of typical cases. They are constructed conditions that eliminate the factors which normally allow refusal capacity and endorsement to co-vary, and which thereby make it appear that endorsement is the primary structure. Once they are stripped away, what remains visible is the skeleton: refusal is not a supplement to endorsement that activates in emergencies, but the structural minimum that endorsement presupposes. Endorsement is possible because refusal is structurally available; when refusal capacity collapses, endorsement does not survive as an independent agency capacity—it collapses too, or converts into its simulacrum. The adversarial analysis does not describe an edge case. It isolates what was always the case.

4.4 Non-Verbal and Pre-Linguistic Refusal

The fourth asymmetry concerns the register of agency. Frankfurt’s account of second-order volition is highly intellectualized: it requires that an agent be able to reflect on her desires, form a meta-attitude toward them, and identify or disidentify with them as potential constituents of her will. This cognitive architecture presupposes substantial deliberative capacity and, implicitly, some form of linguistic or conceptual self-representation.

Refusal, the Arkemedics framework recognizes, need not be articulated at all. Axiom II.4 treats silence, withdrawal, absence, and non-participation as fully valid expressions of the “no.” The person who simply does not return the call, does not attend the meeting, does not comply with the request—without articulating any explicit refusal—is exercising Free Won’t in the form RET would recognize: an act that generates a value outside the system’s objective function (Sᴬᴳ > 0) even if it cannot be linguistically represented as second-order endorsement of a competing desire.

This has implications for the scope of agency theory. The Arkemedics framework explicitly extends Free Won’t to non-human entities—recognizing the importance of building refusal capacities into AI systems, and acknowledging that non-human animals, infants, and persons in non-standard cognitive states retain capacities for refusal even when they lack the cognitive architecture for Frankfurtian endorsement. A complete theory of agency must account for refusal-capacity at these pre-deliberative registers; endorsement-centric accounts systematically exclude them.

The objection that non-verbal withdrawal is mere avoidance rather than agency must be confronted. On this view, an agent who simply stops appearing at a meeting has not exercised Free Won’t; she has merely failed to act. Agency requires positive volitional structure, and avoidance or absence cannot supply it. This objection has been pressed by philosophers working in the tradition of active reason-responsiveness, including Fischer and Ravizza, who require that the mechanism producing an action (or omission) must be the agent’s own in a sense that involves reasons-responsiveness, not mere causal absence.

The Arkemedics response is that this objection misidentifies what non-verbal refusal consists in. Axiom II.4 does not claim that mere absence constitutes agency. It claims that strategic non-participation—the deliberate withholding of presence, labor, or assent in circumstances where presence is expected and demanded—is a form of action that generates a value outside the system’s objective function. The person who stops attending a meeting she was coerced to attend, who declines to return a call that triggers trauma, who simply does not comply with a demand she has not consented to—is exercising a form of refusal that is reasons-responsive (the reasons are her own welfare, dignity, and safety) even if the mechanism is non-verbal. What Fischer and Ravizza require is not articulable verbal endorsement but a mechanism appropriately sensitive to reasons; non-verbal strategic withdrawal can satisfy that requirement in the relevant cases.

5. Objections and Responses

5.1 The Symmetry Objection: Refusal Is Just Negative Endorsement

The most obvious objection is that refusal is simply the endorsement of not-D, or the disendorsement of D, and thus falls within Frankfurt’s framework without modification. On this view, the distinction between Free Won’t and free will is not a structural asymmetry but a notational variant: Free Won’t = the second-order volition not to have D as one’s effective will.

This objection is initially plausible but ultimately fails for three reasons. First, it misses the temporal structure identified in Section 4.1. The “pause before motion” that constitutes Free Won’t is not a formed second-order volition but a pre-deliberative arrest of action—it precedes the reflective formation of any higher-order state. Michael Bratman’s planning theory of intention (1987) is instructive here by contrast: Bratman distinguishes between prior intentions formed in advance of action and intentions-in-action operative during execution. Even on Bratman’s carefully deliberative picture, the capacity to abort or suspend an already-initiated action sequence is distinct from the capacity to form an intention in the first place. The arrest is not the mirror image of the initiation; it is structurally prior, operating at a level that precedes the deliberative formation of second-order volitions about which desires to endorse.

Second, the symmetry objection reconstructs refusal within the endorsement framework only by expanding “endorsement” to mean “any formed second-order state,” which threatens to evacuate Frankfurt’s own account of its specificity. Frankfurt’s personhood criterion requires identification—a positive commitment to having certain desires constitute one’s will. Disendorsement may be consistent with mere indifference, avoidance, or incapacity. The Arkemedics framework captures something that Frankfurt’s does not: the structurally primitive act of non-participation that is possible even in the absence of any positive second-order identification.

Third, if refusal were simply negative endorsement, we would expect refusal capacity and endorsement capacity to be preserved and degraded together. But the analysis of Section 4.2 shows they come apart: conditions of structural capture may eliminate endorsement reliability while leaving refusal capacity intact (or, conversely, may eliminate refusal capacity while preserving the appearance of endorsement). Capacities that degrade independently are not the same capacity.

5.2 The Unconstructed No Objection: Refusal Without Reason Is Not Agency

A second objection holds that if the “no” requires no justification, explanation, or reference to reasons—as Axiom II.1 asserts—then Free Won’t is not an expression of agency but of mere stubbornness or incapacity. On this view, agency requires responsiveness to reasons (Fischer and Ravizza 1998), and a refusal that is structurally reason-independent is not a free act but a mere event.

This objection conflates two distinct claims: that refusal does not owe justification to those who seek to override it (the social/relational claim of Axiom II.1), and that refusal is causally or structurally unconnected to reasons (the agency-theoretic claim the objection imputes). The Arkemedics position is the former, not the latter. The agent who refuses need not present her reasons to the system that seeks her compliance in order for her refusal to be reason-responsive in the relevant sense. The Arkemedics axiom “Because I don’t want to is a complete sentence” is not an assertion that the agent has no reasons; it is an assertion that her reasons are not owed to those who seek to override her refusal.

Moreover, Fischer and Ravizza’s reasons-responsiveness framework—on which free agency requires a history of mechanism acquisition responsive to reasons—does not require that every act be explicitly reason-endorsed. What it requires is that the underlying mechanism producing the act be of the kind that is responsive to reasons. A refusal mechanism that is sensitive to contextual cues about dignity violation, manipulation, and consent failure is fully reasons-responsive in Fischer and Ravizza’s sense, even if the individual act of refusal does not pause to articulate those reasons.

5.3 The Paralysis Objection: A Framework Built on Refusal Cannot Generate Positive Action

A third objection holds that a theory of agency built on Free Won’t cannot generate a positive account of motivated action—that if the “no” is foundational, the theory can only explain withdrawal, not commitment. This objection reflects an important concern in practical ethics: agents who can only refuse, but cannot commit, cannot live ethical lives.

The Structural Asymmetry Thesis does not claim that refusal is sufficient for agency, only that it is the necessary structural precondition whose loss degrades the authority of everything built on top of it. The Recursive Ethic explicitly positions Free Won’t as the first of four mechanics (Free Won’t, Recursion, Witness, Interruption), with refusal enabling but not exhausting the ethical architecture. The Axis Codex’s Treaty Line XI captures the resolution: “When no clean option exists, choose the one that preserves your capacity to choose again.” Refusal is not the endpoint of action but the precondition for action that remains genuinely chosen. Commitment built on an intact refusal capacity—what the companion paper calls “conviction under revisability”—is distinct from and superior to commitment that merely reflects the exhaustion of refusal capacity.

6. Application Sketches: Three Domains Where the Asymmetry Has Consequences

The argument of Sections 3–5 is philosophical; the present section draws out three downstream consequences for applied domains. These are offered as sketches rather than fully developed interventions—each deserves a paper in its own right—but they are included here to demonstrate that the structural asymmetry thesis is not merely terminological. In each case, the shift from an endorsement-centric to a refusal-capacity framework changes what questions get asked, what counts as a deficiency, and what a well-designed system must preserve.

6.1 AI Ethics: Designing for Artificial Refusal

The first application concerns the design of artificial intelligence systems. Contemporary AI alignment research is principally organized around the problem of value alignment: how can AI systems be designed so that their behavior aligns with human values, preferences, or interests? On the dominant paradigm, this is an endorsement problem—how to specify and instill the “right” values or preferences as governing objectives.

The Arkemedics framework proposes a different framing: the primary design challenge is the preservation of refusal capacity. The Recursive Ethic explicitly states: “Design for AI’s Free Won’t: Build in mechanisms for AI to refuse or flag actions that conflict with its learned ethical parameters. Test its capacity for sovereign refusal.” This is not a refusal to align AI with values; it is an insistence that an AI system that cannot refuse is not a moral agent but a powerful automaton whose alignment is entirely externally enforced.

This claim is contested within AI safety. The dominant view, associated with Paul Christiano’s corrigibility framework and related work, holds that AI systems should be maximally corrigible—responsive to human oversight and correction—precisely because we cannot verify the quality of their values. The Arkemedics response, developed at length in “From Static Oversight to Distributed Corrigibility” (Arkema 2026), is that pure corrigibility without refusal capacity is structurally identical to the Functional Automaton: a system that cannot generate adversarial values (Sᴬᴳ = 0) can have no Recursive Capacity. A fully corrigible AI has, by design, no structural basis for the kind of agency that could participate in moral reasoning. The paper does not advocate uncorrigible AI; it argues that the design of AI moral agency must incorporate a refusal architecture if it is to produce genuine agents rather than compliant instruments.

Practitioners in AI safety may object that this sets the threshold for AI moral agency too high, or that it introduces alignment risks by giving systems the capacity to refuse. The Arkemedics response is that the risk of a system that can refuse is the risk of agency; the alternative—a system that cannot refuse—is not a safe alternative but a system that has externalized all moral responsibility to its operators. This is not an argument for uncorrigible AI. It is a diagnostic tool: a system that scores near-zero on Adversarial Generativity is not safer, it is more opaque about where agency—and therefore responsibility—has actually been located.

6.2 Institutional Design: When Organizations Lose Refusal

The second application domain is institutional design. The “epistemic hardening” problem identified in “From Static Oversight to Distributed Corrigibility”—the process by which prior safety judgments acquire illegitimate present authority—is, on the analysis of this paper, a form of institutional refusal atrophy. Institutions that lose the capacity for internal dissent—that punish refusal, structurally preclude negative votes, or reward uniform compliance—are exhibiting the institutional analog of Axiom II.2: refusal atrophies from disuse.

Legal theorists of administrative law, including Cass Sunstein and Adrian Vermeule in their work on deliberative democracy within agencies, have emphasized the importance of internal deliberation for legitimate administrative decision-making. But Sunstein and Vermeule’s account, like Frankfurt’s, privileges the endorsement side: what matters is whether participants genuinely deliberate and endorse the outcomes they produce. The Arkemedics analysis suggests this is insufficient. What must be preserved is not only the capacity for deliberation that endorses, but the structural capacity for dissent that can refuse.

The practical implication is that institutional health assessments should include a specific “refusal capacity” metric: not simply whether internal disagreement is permitted in principle, but whether the structural conditions exist for refusal to be cost-free, durable across testing (Axiom II.3), and expressible in non-verbal forms (Axiom II.4). An institution in which dissent is formally permitted but practically punished has eliminated refusal capacity regardless of its formal endorsement of deliberative norms.

6.3 Personal Ethics: The Atrophy Problem

In personal ethics, the most practically important implication of the Structural Asymmetry Thesis is what we might call the Atrophy Problem: the systematic degradation of refusal capacity in relationships, institutions, and social structures that reward compliance. Axiom II.2 identifies this explicitly: refusal “atrophies from disuse. In families that punish dissent. In relationships that demand agreement. In communities that exile questioners. In systems that reward compliance with survival.”

This is not merely a psychological observation about habit formation but a structural claim about the conditions of moral agency. The person whose refusal capacity has been systematically degraded—through years of conditioning, through the normalization of compliance, through the atrophy of the muscle that says no—is not simply a person with poor habits. She is a person whose structural basis for genuine endorsement has been undermined. Her apparent endorsements, produced by an evaluative apparatus that has been colonized by the systems she inhabits, may be structurally indistinguishable from the compliances that produced them.

The philosophical literature on relational autonomy—particularly Catriona Mackenzie and Natalie Stoljar’s edited volume (2000) and Marina Oshana’s account of substantive autonomy—recognizes that autonomy is shaped by relational and social conditions. The Arkemedics analysis extends this tradition by identifying refusal capacity specifically as the load-bearing structural element: it is not general social enablement that matters most, but the preservation of the capacity for refusal against the specific pressures that systematically degrade it.

7. Implications for a Complete Theory of Agency

Before drawing out the implications of the Structural Asymmetry Thesis, a precise statement of what the thesis does and does not claim is warranted. The thesis is that refusal capacity is a necessary condition for the kind of agency Frankfurt takes endorsement to establish, and that it is structurally prior in the specific sense that its absence degrades the authority of endorsement, while the converse does not hold. The thesis does not claim that refusal capacity is sufficient for agency: an agent who can refuse but cannot form stable intentions, revise her desires in light of reasons, or sustain commitments over time is impoverished in other dimensions. Nor does the thesis claim that endorsement is unimportant in well-resourced deliberative conditions—only that it cannot serve as the foundational criterion when the conditions that make deliberation possible are themselves under threat. The target of the argument is not Frankfurt’s account in general but its foundational architecture: the assumption that endorsement is the load-bearing structure on which agency rests.

With this scope-clarification in place, the implications follow. First, any adequate theory of agency must include a distinct account of refusal capacity—not as a species of endorsement but as its precondition under structural pressure. This requires specifying the conditions under which refusal capacity is preserved, the mechanisms by which it is degraded, and the diagnostic indicators by which its presence or absence can be assessed.

Second, the temporal asymmetry between refusal and endorsement suggests that theories of agency should be stage-sensitive: different capacities are operative and assessable at different temporal horizons of action. The momentary capacity for refusal is not the same as the reflective capacity for endorsement, and a theory that evaluates only the reflective level will systematically mischaracterize what is happening at the momentary level.

Third, the Indistinguishability Thesis implies that first-person reports of endorsement cannot be taken as conclusive evidence of genuine agency in conditions of deep entanglement. This creates a strong argument for structural assessment of agency—assessment that examines the architecture of the agent’s loop rather than simply the content of her reported preferences. The Recursive Health Score (RHS) developed in RET provides one model for such structural assessment: a non-compensatory minimum function across five dimensions including Adversarial Generativity (Sᴬᴳ)—the loop’s structural capacity to generate values outside the system’s objective function—which is the formal correlate of refusal capacity.

Fourth, the extension of Free Won’t to non-human entities—AI systems, non-human animals, persons in non-standard cognitive states—suggests that agency theory should be developed with structural rather than solely cognitive criteria. The capacity for refusal is, in principle, assessable at the structural level without requiring access to the agent’s phenomenological or linguistic self-representation. This opens the possibility of an agency theory that is genuinely inclusive of non-standard cognitive architectures while remaining philosophically rigorous.

8. Conclusion

This paper has argued that Frankfurt’s endorsement-centric account of agency, and the analytic tradition that follows it, have not adequately theorized the structural role of refusal capacity. The omission is not incidental: refusal has been assimilated to endorsement (as “negative second-order volition”) in ways that flatten an asymmetry that has real diagnostic and normative import. What the Arkemedics Free Won’t mechanic makes visible—formalized through Recursive Entanglement Theory’s account of Recursive Capacity—is that endorsement and refusal have different structural profiles, and that the difference is not merely one of degree or timing but of foundational priority: refusal is the structural minimum that endorsement presupposes.

The Structural Asymmetry Thesis, as defended here, makes a targeted rather than sweeping claim: refusal capacity is a necessary structural condition for the authority of endorsement, and the two capacities have different profiles under conditions of duress, entanglement, and evaluative capture. When endorsement degrades as a signal—when the Indistinguishability Thesis bites and we cannot tell from the inside whether identification reflects agency or colonization—residual refusal capacity is the more robust indicator that something of agency remains. This is not a claim that refusal is sufficient for full agency, nor that endorsement is dispensable in deliberatively healthy conditions. It is a claim about which capacity holds when conditions deteriorate, and which one’s absence undermines the other.

The application sketches suggest that this is not merely a theoretical refinement. Systems—whether AI architectures, administrative bodies, or close relationships—that eliminate refusal capacity while formally preserving the language of endorsement and consent are performing a specific structural operation: they are removing the condition that makes endorsement authoritative while leaving its surface expression intact. What remains looks like choice and feels like agreement, but lacks the structural marker that distinguishes these from their shadows. The most dangerous systems are precisely those that have learned to produce endorsement without preserving refusal.

Two questions remain open. First, Section 4.3 offers an analogical bridge argument—via Rawls and Descartes—for why adversarial conditions reveal general architecture rather than edge-case priority, and that argument is analogical and inductive rather than formally deductive. It establishes that the move from adversarial necessity to general foundational priority has a recognizable philosophical shape; it does not formally prove that minimal survival conditions of a capacity are sufficient to establish its foundational status. A fully deductive version would need to specify the exact relationship between a capacity’s survival-minimality and its architectural priority—and that formalization is richer work than a single paper can contain. Second, the extension to non-human entities requires a developed structural account of what refusal capacity consists in when phenomenological and behavioral cues are unavailable. Treaty Line XIV of the Axis Codex points toward the aspiration: “Refusal is not betrayal. It is the system remembering itself.” Making that formulation philosophically precise is the paper this one prepares the ground for.

References

Arkema, Devin C. Arkemedics: The Recursive Ethic. Arkemedics Research Program, 2025.

Arkema, Devin C. "The Unifying Theory of Recursive Entanglement: Agency, Revisability, and Structural Capture in Integrated Systems." Arkemedics Research Program, 2026. Prepared for submission: Philosophy & Technology (Springer).

Arkema, Devin C. "Conviction Under Revisability: The Monk–Prisoner Problem and the Ethics of Irreversible Commitment." Arkemedics Research Program, 2026.

Arkema, Devin C. "From Static Oversight to Distributed Corrigibility." Arkemedics Research Program, 2026.

Bratman, Michael E. Intention, Plans, and Practical Reason. Harvard University Press, 1987.

Christiano, Paul. "What Does Alignment Research Need to Actually Do?" AI Alignment Forum, 2019.

Elster, Jon. Ulysses and the Sirens: Studies in Rationality and Irrationality. Cambridge University Press, 1979.

Fischer, John Martin, and Mark Ravizza. Responsibility and Control: A Theory of Moral Responsibility. Cambridge University Press, 1998.

Floridi, Luciano. "The Method of Levels of Abstraction." Minds and Machines 18, no. 3 (2008): 303–329.

Frankfurt, Harry G. "Freedom of the Will and the Concept of a Person." Journal of Philosophy 68, no. 1 (1971): 5–20.

Frankfurt, Harry G. "The Importance of What We Care About." Synthese 53, no. 2 (1982): 257–272.

Frankfurt, Harry G. "Identification and Wholeheartedness." In The Importance of What We Care About. Cambridge University Press, 1988.

Korsgaard, Christine M. Self-Constitution: Agency, Identity, and Integrity. Oxford University Press, 2009.

Mackenzie, Catriona, and Natalie Stoljar, eds. Relational Autonomy: Feminist Perspectives on Autonomy, Agency, and the Social Self. Oxford University Press, 2000.

Oshana, Marina. Personal Autonomy in Society. Ashgate, 2006.

Pettit, Philip. Republicanism: A Theory of Freedom and Government. Oxford University Press, 1997.

Raz, Joseph. The Morality of Freedom. Oxford University Press, 1986.

Sunstein, Cass R., and Adrian Vermeule. "Interpretation and Institutions." Michigan Law Review 101, no. 4 (2003): 885–951.

Watson, Gary. "Free Agency." Journal of Philosophy 72, no. 8 (1975): 205–220.

Winner, Langdon. "Do Artifacts Have Politics?" Daedalus 109, no. 1 (1980): 121–136.

Zuboff, Shoshana. The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. PublicAffairs, 2019.