Consider a composite drawn from the clinical reports: a retired accountant in his sixties began treatment for Parkinson’s disease. Within months he had gambled away the family savings a hundred dollars at a time. He hid the statements, lied about his evenings, swore each morning he had finished, and drove back to the casino that night. When his neurologist lowered the dose of a dopamine agonist, the compulsion drained away. The sequence recurs in the clinical literature: drug started, compulsion installed, dose lowered, compulsion gone.¹

The case makes causation vivid and therefore tempts a quick verdict: if a milligram can install a craving, perhaps the machinery ran the show all along. But the drug did not reveal that action has causes. Every action has them. It damaged something more specific: the accountant could still judge the conduct ruinous, yet that judgment had lost most of its power to govern what he did.

Compare an ordinary evening when someone wants to go out, remembers a promise to stay with the children, and puts the keys down. The desire remains; a reason changes the action. Both this choice and the accountant’s return to the casino have causes. The difference concerns whether what the person understands can govern what he does.

18.1 The Ghost’s Freedom

The popular skeptic and the libertarian share one picture of what freedom would have to be: origination, the power to begin a causal chain without standing at the end of one. Trace a choice upstream and you meet neurons, hormones, upbringing, and the history of a particular body. The skeptic concludes that freedom never appears. The libertarian looks for an uncaused act or an indeterministic opening through which the agent can intervene. One picture, two verdicts.²

That picture mistakes reliable causation for the enemy of agency. If a decision did not flow from values, memories, loves, fears, and attention to the situation, what would remain? Chance. A quantum swerve would hand the choice to nobody. Randomness buys no authorship.

The careful skeptics press a different question. Galen Strawson argues that ultimate responsibility would require responsibility for the mental equipment from which one acts, and therefore self-creation before one existed to create anything. Derk Pereboom denies basic desert: the claim that a wrongdoer merits suffering simply for the deed. Those arguments deserve their own measure. The account below establishes a capacity and a standing to be addressed. It does not establish self-creation or a right to make anyone suffer.³

18.2 The Excusing Conditions

Courts, parents, and ordinary morality distinguish excuses within the causal order. Compulsion, delusion, chemical disruption, and coercion can each matter, but they do not all matter in the same way. A person who signs at gunpoint may reason perfectly well: survival gives him an overwhelming reason to sign. The threat removes an acceptable opportunity to refuse without necessarily damaging his ability to deliberate.⁴

Being caused distinguishes none of these cases. We need to ask both what control the person had and what the circumstances allowed us fairly to demand.

The relevant capacity registers considerations, weighs them, carries them into action, and revises after error. Compulsion can block action on reasons already recognized; delusion can block recognition; disintegration can prevent the system from acting as one. Philosophers call the intact capacity reasons-responsiveness.⁵ The accountant could understand the cost to his family while failing to govern his conduct by it. The person under threat may act precisely because he understands the cost of refusal. Impaired control and coercion can both excuse without sharing one mechanism.

Why should that capacity carry moral weight? Because holding someone responsible begins by addressing that person: making a demand, asking for a reason, waiting for an answer. Nobody addresses the drug or the weather. A demand lands intelligibly on a continuing subject who can understand it and revise. When I ask why you broke a promise, I treat the promise, your understanding of it, and your present power to answer as parts of one deliberative life. But a recovered patient can answer questions now about an act he lacked the capacity to control then. Present answerability does not retroactively supply that missing control, and an intact capacity never makes every demand fair.

This grounds fitness for address, not basic desert. Protest, repair, changed expectations, and revised trust can survive without the further claim that suffering is deserved for its own sake. The chapter keeps that boundary in view.

18.3 Reasons as Contents

A shove moves a body. A reason reaches conduct through what it means. The mortgage weighs on you because you represent it as a fact about the choice; a dog’s sight of the leash can guide it toward the walk it anticipates. To answer a reason is to answer something your states concern.

Free agency therefore rests on intentionality, the directedness Brentano called the mark of the mental. Yet content cannot compete with its neural vehicle as a second force. It is a property of the vehicle-state under the description that says what the state tracks. The real question is whether the state guides conduct in virtue of meaning what it means.

Fred Dretske’s distinction helps. A neural event may trigger the accountant’s hand reaching for the chips. A history of learning and use explains why that kind of event occupies that place in the control system. Dretske called this the structuring cause. Meaning does not shove beside the neurons; it explains why the control system has the shape from which the agent acts.⁶ The exclusion problem remains disputed, but this narrower explanatory role is enough for the present argument.

Content alone does not suffice. A representation becomes a reason when it enters a deliberative economy:

1. available to memory;
2. compared with alternatives;
3. revised under correction; and
4. capable of bending whole-system conduct.

Integration must run both ways. A consideration must reach decision and action; success, failure, and correction must return to memory and change what happens next. A rationale produced after the deed may display verbal competence without reasons-responsive control. A terse deliberation may count fully if differences in represented reasons reliably alter the course of the system.⁷

Freedom therefore comes in grades. Systems vary in the considerations they can register, the reliability with which conduct responds, and the span of life held in view. A dog can respond over the next few seconds; a person can weigh a mortgage against a daughter’s last year at home across a decade. Causal-functional history fixes what a state represents. Integration makes that content available across memory, comparison, correction, planning, and action. Reasons-responsiveness lets differences in content bend conduct. Reliable causation supplies the medium in which all three achievements work.⁸

The dependency is asymmetric. Original content does not depend on free will: a frog’s prey circuit may answer to a world-fixed correctness condition without deliberating, and an agent may reason with language inherited from a community. Free agency does depend on content, because a reason can govern conduct only by representing a consideration as bearing on what to do. Reverse the order and the account circles—agency needs reasons, reasons need content, content needs agency. Freedom gives content a new role inside an integrated subject; it does not create the content.

Neither agency nor semantic originality requires self-creation. Inheritance can supply what a system later uses. The two assessments nevertheless differ: a content-fixing history tells us what a state represents, while free action requires the right kind of control in the circumstances of acting.

18.4 Three Objections

18.4.1 Could the agent have done otherwise?

Searle asks whether the person could have acted differently with every other condition held fixed. Determinism says no. The compatibilist should concede that modal point and reject the premise that agency requires a second future from an identical universe.⁹

Capacities follow a different logic. A glass remains fragile while safe on a shelf; a pianist retains the ability to play on an evening when she never approaches the piano. Likewise, a person can possess the capacity to refuse while actual reasons, character, and circumstances determine that he accepts. The relevant counterfactual asks whether conduct would change as reasons or opportunities changed, not whether one total state of the universe could branch unaided into two futures.

The felt openness of deliberation changes nothing. A deliberator cannot read the outcome from the physics in advance, and must weigh reasons to settle what to do. The weighing is not spectatorship. Remove it, change the considerations, or alter their force, and conduct may change. Deliberation works because it belongs to the causal process.

18.4.2 What about manipulation?

Pereboom’s continuum moves through four cases:¹⁰

1. intervention just before deliberation;
2. programming at the start of life;
3. rigorous childhood conditioning; and
4. ordinary deterministic upbringing.

An intervention that replaces the agent’s weighing gives us a case of bypass. Pereboom’s harder case preserves a reasons-responsive process: the manipulation works through deliberation, and the agent would refrain in some situations with different reasons. Calling that process bypassed simply because someone altered it would remove the feature the objection asks us to confront.

The distinction concerns control, not elapsed time since someone intervened. Suppose a programmed agent grows into a persisting subject who perceives, errs, corrects, forms attachments, weighs reasons, and lets demands reshape later conduct. That subject possesses the capacity the address account requires. Now imagine an ongoing manipulator who changes its priorities but leaves that capacity intact. If the agent’s revised assessment still governs action, the account must grant that it remains someone we can address. If the manipulator cancels every unwelcome revision, the appearance of weighing reasons hides a limit on control. We must test which case we have rather than infer it from the presence of a manipulator.

Granting answerability in the intact case has a cost: the person remains subject to demands despite another’s continuing influence. That does not give the manipulator a right to make those demands, make every demand fair, or establish basic desert. Manipulation may wrong someone without removing all agency, and freedom to reason can coexist with domination. Pereboom’s challenge to deserved blame therefore remains standing; the narrower capacity defended here does not answer it by quietly turning his agent into a puppet.

18.4.3 Doesn’t luck undermine responsibility?

Luck supplies the equipment, the occasion, and sometimes the thought that happens to surface. The premise holds. It makes one deed thin evidence about a person and should temper what responsibility licenses. A graded capacity calls for graded expectations, mercy, repair, prevention, and rehabilitation rather than confidence that suffering is earned.

But luck does not erase the present capacity to understand a demand and revise. It bears hard on desert while leaving answerability intact. The objection narrows the moral conclusions. It does not return the ghost.

18.5 What the Account Commits Us To

Reasons-responsiveness measures neither wisdom nor goodness. A zealot may deliberate coherently from terrible commitments; a selfish person may understand every consideration and discount the ones that matter to everyone else. Both remain answerable because reasons can reach their conduct. The account identifies agency, not virtue.

Nor does it make narration, linguistic fluency, or a practical stake necessary for every grade of agency. Chapter 17 identified the subject as a persisting control organization within which contents can update a shared economy of memory, attention, correction, planning, and action.

The positive commitment is spare: free agency requires content-bearing states integrated into one persistent economy such that represented differences can alter deliberation and conduct, and consequences can return as correction. The capacity can develop gradually, fail locally, and admit borderline cases. Nothing in it asks a cause to go missing.

18.6 From Agency to Other Minds and Machines

The chapter began with the self assembled in Chapter 17. It ends with a test for what that self can do: take in represented reasons, weigh them within one continuing organization, act from them, and revise when the world or another person answers back.

That last phrase opens the next chapter. Responsibility takes the form of address, and address places another minded subject across the table. Chapter 19 asks how we recognize such a subject without rebuilding the inner theater as a private chamber hidden behind behavior.

Part Four will carry the architectural question to machines. A system can print reasons without reasoning by means of them. Fluent rationales establish no persistent economy of memory, correction, planning, and control, just as engineered origin disqualifies none. The test is whether represented reasons can enter and govern one continuing cognitive organization. A separate test asks whether success and failure matter to that system as autonomy or welfare. A machine might meet neither, either, or both.

---

Chapter Summary

The claim. Free will names no exemption from the causal order but a graded capacity inside it: represented reasons can govern conduct through an integrated and persistent deliberative economy. Content precedes that capacity; freedom does not mint meaning.

Where the argument stands. Impaired control can excuse; coercion can also excuse while leaving deliberation intact, because the threat changes what anyone could fairly demand. Present capacity to answer does not establish control over an earlier act. Content, reason-guided agency, and moral answerability remain distinct; none requires self-creation or randomness.

What’s left open. The chapter grounds answerability, not basic desert. Luck limits what responsibility may fairly exact. Practical stake (autonomy and interests) and welfare remain separate questions, and the threshold for reasons-responsive integration remains empirical and graded.

The hand-off. Chapter 19 turns from the subject who answers reasons to the other subject whom we address. Part Four then asks whether represented reasons ever enter and govern one continuing artificial cognitive economy.

---

Notes

1. Dopamine agonists used to treat Parkinson’s disease (pramipexole, ropinirole) can precipitate impulse-control disorders — pathological gambling, compulsive shopping, hypersexuality, binge eating — in patients with no prior history; case reports describe remission when the agonist is reduced or withdrawn. The vignette composites the now-standard clinical picture. For the causal claim see M. Leann Dodd et al., “Pathological Gambling Caused by Drugs Used to Treat Parkinson Disease,” Archives of Neurology 62, no. 9 (2005): 1377–1381; for the scale, Daniel Weintraub et al., “Impulse Control Disorders in Parkinson Disease: A Cross-Sectional Study of 3090 Patients,” Archives of Neurology 67, no. 5 (2010): 589–595, who found an impulse-control disorder in 13.6% of the full sample and, dividing by exposure, roughly 17% of patients taking a dopamine agonist against roughly 7% of those not — the elevation on which the causal reading turns. The usual mechanistic reading is agonist overstimulation of mesolimbic reward circuitry mediating impulse regulation. Patients often conceal the behavior and describe it as against their better judgment, though concealment by itself shows awareness of disapproval rather than a settled evaluative judgment. Two kinds of evidence sit behind the vignette and should not be run together. The dechallenge sequence — agonist reduced or withdrawn, behavior remitting — rests on case series and clinical reports rather than on a controlled within-patient trial; the Weintraub figures report cross-sectional prevalence, an association across a large sample rather than a demonstrated dose-response in thousands of individuals. Reading that sequence as judgment intact but losing its grip on conduct is the chapter’s philosophical interpretation of the composite, not a finding the clinical reports establish. Nor does the argument need total loss of control: on a graded account, substantially impaired control supplies the same contrast. ↩
2. Robert M. Sapolsky, Determined: A Science of Life Without Free Will (New York: Penguin Press, 2023); Sam Harris, Free Will (New York: Free Press, 2012). The philosophically rigorous version of their position is “hard incompatibilism” — free will incompatible with determinism and indeterminism alike — defended in Derk Pereboom, Living Without Free Will (Cambridge: Cambridge University Press, 2001). Pereboom rests the case not on neuroscience but on a four-case manipulation argument and on the luck problem facing event-causal libertarianism; the imaging studies supply vividness, not modal force. The distinction matters. The popular books argue that a choice flowing from unchosen antecedents has no author; Harris states the argument as a dilemma, holding that a will determined by prior causes and a will produced by chance leave us equally without responsibility, and adds an introspective case that conscious authorship cannot be found. The chapter accepts the chance horn (Section 18.1) and disputes the inference in the other: that agency requires authorship of the antecedents from which one acts. The professional literature has for over a century declined to treat the bare fact that choices have causes as settling anything — libertarians denying causal completeness outright, compatibilists granting it and denying its significance. Pereboom’s target is narrower still, and Section 18.2 takes it at its own measure: basic desert moral responsibility — the wrongdoer’s deserving blame or suffering simply in virtue of having knowingly done wrong, apart from any further good the response secures — which he distinguishes carefully from forward-looking accountability, moral protest, and relationship repair, all of which he retains (Derk Pereboom, Free Will, Agency, and Meaning in Life, Oxford: Oxford University Press, 2014, ch. 1). The chapter defends the capacity and the standing to be addressed; it takes no position on desert in Pereboom’s strict sense. ↩
3. Galen Strawson, “The Impossibility of Moral Responsibility,” Philosophical Studies 75, nos. 1–2 (1994): 5–24. The core: (1) nothing can be causa sui; (2) true moral responsibility would require being causa sui, at least in crucial mental respects; (3) therefore nothing can be truly morally responsible. Strawson is explicit that the argument goes through whether determinism is true or false. For the causa sui’s pedigree as an object of ridicule, see Friedrich Nietzsche, Beyond Good and Evil (1886), sec. 21: “the best self-contradiction that has been conceived so far.” The use made of the argument here — as a reductio of the exemption conception of freedom rather than of responsibility — is not Strawson’s; he accepts the conclusion at face value. (He is the son of P. F. Strawson, whose reactive-attitudes account, n. 4, sits at the opposite pole of the literature.) The argument’s most formidable descendant drops the self-creation demand entirely: Neil Levy, Hard Luck: How Luck Undermines Free Will and Moral Responsibility (Oxford: Oxford University Press, 2011), argues that constitutive luck (the agent’s endowment) and present luck (the circumstances in which the endowment gets exercised) between them exhaust the difference between agents, so that no account — compatibilist or libertarian — can locate a difference the agent is responsible for. Section 18.4.3 concedes the premise and confines the damage to the desert question the chapter has already set aside (n. 2). ↩
4. P. F. Strawson, “Freedom and Resentment,” Proceedings of the British Academy 48 (1962): 1–25. The reactive attitudes — resentment, gratitude, indignation — constitute the practice of holding responsible, and Strawson argues that the practice neither needs nor could receive an external metaphysical justification. The present chapter supplies a modest naturalistic account of the capacities that make address possible. It does not reduce every excuse to incapacity: coercive circumstances may defeat the fairness of a demand while leaving reasons-responsive deliberation intact. ↩
5. The canonical development is John Martin Fischer and Mark Ravizza, Responsibility and Control: A Theory of Moral Responsibility (Cambridge: Cambridge University Press, 1998): moral responsibility requires “guidance control,” analyzed as the action’s issuing from the agent’s own, moderately reasons-responsive mechanism, and does not require “regulative control” over alternative possibilities. The dispensability of alternative possibilities descends from Harry Frankfurt, “Alternate Possibilities and Moral Responsibility,” The Journal of Philosophy 66, no. 23 (1969): 829–839; the hierarchical account of an agent identified with her effective desires, from Frankfurt, “Freedom of the Will and the Concept of a Person,” The Journal of Philosophy 68, no. 1 (1971): 5–20. Guidance control so analyzed has two components. Reasons-responsiveness is a current-time-slice property of the mechanism; ownership is historical, secured by the agent’s “taking responsibility” for the mechanism over time. The chapter adopts the first component and does not adopt the historical ownership requirement: Section 18.4.2 declines that route for the manipulation cases (n. 10) and rests instead on the capacity together with the normative ground given for it in Section 18.2: holding responsible consists in addressing a demand to someone, so the capacity to take in reasons and revise on them fixes who can fairly be addressed. That ground is developed in R. Jay Wallace, Responsibility and the Moral Sentiments (Cambridge, MA: Harvard University Press, 1994), and in the conversational account of Michael McKenna, Conversation and Responsibility (New York: Oxford University Press, 2012); T. M. Scanlon, Moral Dimensions: Permissibility, Meaning, Blame (Cambridge, MA: Harvard University Press, 2008), arrives at a relational treatment from the direction of blame’s meaning. Daniel Dennett, Elbow Room: The Varieties of Free Will Worth Wanting (Cambridge, MA: MIT Press, 1984), surveys the compatibilist capacities in the same spirit. The wider field the chapter’s verdict sits in: Gary Watson, “Free Agency,” The Journal of Philosophy 72, no. 8 (1975): 205–220, founds the tradition grounding freedom in valuational rather than merely motivational structure; Susan Wolf, Freedom Within Reason (Oxford: Oxford University Press, 1990), makes responsiveness to the True and the Good the condition; Manuel Vargas, Building Better Beings: A Theory of Moral Responsibility (New York: Oxford University Press, 2013), reaches a neighboring destination by an openly revisionist route. The address account is nearest kin to the quality-of-will and answerability lineage; its distinctive emphasis is that represented reasons must be integrated into the continuing control economy of the addressee. A stake may make the exercise bear on that being’s welfare, but it does not make a content into a reason. ↩
6. The worry is the exclusion problem: if every physical effect has a sufficient physical cause (causal closure), a mental property seems either to overdetermine the effect or to do no causal work at all — see Jaegwon Kim, Mind in a Physical World (Cambridge, MA: MIT Press, 1998). The reply taken here does not treat content as a further cause competing with the neural vehicle for the same effect; on the identity claim the content-property is a property of the vehicle-state (the state under the description that specifies what it tracks), so no competition arises. That a state’s meaning can be causally relevant to what an agent does is argued in Fred Dretske, Explaining Behavior: Reasons in a World of Causes (Cambridge, MA: MIT Press, 1988): the semantic fact figures as a structuring cause, a state’s being recruited to indicate a condition and thereby to control behavior. Section 18.3 states the distinction in Dretske’s own terms: the triggering cause of a particular movement, and the structuring cause that explains why states of that kind control behavior of that kind at all. Dretske’s account is not generally regarded as having dissolved Kim’s exclusion problem, and the main text claims only what the structuring role supplies — a content that explains the shape of the control system the agent acts from. The recruitment history may establish genuine mechanism content without ownership; the additional agency question is whether that content participates in integrated reasons-responsive control. ↩
7. That motivating reasons are representational states — beliefs and desires with intentional content — is the common ground of the causal theory of action. Armstrong makes a related point about reasons for belief: where one piece of knowledge bears no relation to the knowledge it supposedly supports, “It is a reason only by courtesy” (D. M. Armstrong, A Materialist Theory of the Mind, London: Routledge & Kegan Paul, 1968, 203). He proposes a potential-cause condition and then considers objections to it. The main text makes an analogous argument about action: a rationale’s availability does not show that it helped govern conduct. It neither quotes Armstrong nor attributes that action formulation to him. The reasons-receptivity half of reasons-responsiveness (n. 5) is likewise a representational capacity: the agent’s recognition of considerations as reasons. The claim that free agency is built from intentionality — the directedness of a state at an object, Brentano’s mark of the mental (Franz Brentano, Psychology from an Empirical Standpoint, 1874) — asserts a dependence, not an identity: a bacterium may have minimal aboutness and no freedom, so intentionality supplies necessary scaffolding, not sufficiency. What the ladder of freedom adds above bare aboutness is integration: contents become available for comparison, correction, long-range planning, and control as reasons. ↩
8. David Hume, An Enquiry Concerning Human Understanding (1748), sec. VIII (“Of Liberty and Necessity”). Liberty is “a power of acting or not acting, according to the determinations of the will” — belonging to “every one who is not a prisoner and in chains” — and actions that “proceed not from some cause in the character and disposition of the person who performed them” cannot “redound to his honour, if good; nor infamy, if evil.” On Hume’s analysis, causal connection to character is required for attribution and desert, not opposed to it. The classical compatibilist lineage runs from Hobbes through Hume to Mill; the position defended here descends from it, with reasons-responsive capacities doing the work Hume assigned to unimpeded willing. ↩
9. John Searle, Minds, Brains and Science (Cambridge, MA: Harvard University Press, 1984), ch. 6, “The Freedom of the Will”: compatibilism “doesn’t really answer that question in a way that allows any scope for the ordinary notion of the freedom of the will” — the question being whether the agent could have done otherwise “all other conditions remaining the same.” Searle’s own positive view ties the experienced “gap” of deliberation to a speculative indeterminacy at the neuronal level and concedes the problem unsolved (Rationality in Action, Cambridge, MA: MIT Press, 2001; Freedom and Neurobiology, New York: Columbia University Press, 2007). The proposal is not that chance decides: Searle wants an indeterminism at the neurobiological level through which a conscious, unified, rational self operates, so that the act is both the agent’s and undetermined, and he grants that he has not shown how the conjunction is secured. The main text presses the luck objection against it — an undetermined outcome answers to the agent no more than to anyone else — which is the standing difficulty for event-causal libertarianism generally (Appendix B.19), not a defect peculiar to Searle. Kant, Critique of Practical Reason (1788), Ak. 5:96–97, dismisses compatibilist liberty as “the freedom of a turnspit”; William James, “The Dilemma of Determinism” (1884), calls compatibilism “a quagmire of evasion.” The reply contests the premise all three share: that the ordinary “could have done otherwise” carries the total-state reading rather than the capacity-and-opportunity reading. ↩
10. The manipulation argument is Pereboom’s (n. 2): a four-case continuum running from an agent whose reasoning is manipulated near the time of action, through one programmed at the start of life, through one rigorously conditioned in childhood, to one produced by ordinary deterministic upbringing. The challenge asks what responsibility-relevant difference could separate the first from the last. Alfred Mele’s zygote argument (Free Will and Luck, New York: Oxford University Press, 2006) presses a related structure through a single engineered case. Pereboom’s first case does not simply stipulate failed reasons-responsiveness. In the 2014 formulation, the neuroscientists intervene before deliberation, inducing a reasons-responsive egoistic process rather than an irresistible desire; the formulation is reproduced in Derk Pereboom and Michael McKenna, “Manipulation Arguments against Compatibilism,” in The Oxford Handbook of Moral Responsibility, ed. Dana Kay Nelkin and Derk Pereboom (Oxford: Oxford University Press, 2022), 179–200 (consulted author-hosted penultimate manuscript, pp. 6–7). Section 18.4.2 distinguishes genuine bypass from manipulation through intact deliberation and adds an ongoing-manipulation variant of its own. Where integrated reasons-responsive control is preserved, it retains answerability, not a verdict of basic-desert responsibility. It leaves open what demands remain fair under domination and does not diagnose every contrary intuition as a demand for self-creation. Content, integration, agency, answerability, and practical stake remain distinct; the machine question in Part Four must assess each on its own evidence. ↩