Problem of Free Will in Determinism and Indeterminism

Abstract

This paper re-examines the problem of free will in light of both deterministic and indeterministic assumptions about the structure of the world. On the philosophical side, it analyzes van Inwagen’s arguments that free will is incompatible with determinism—because our actions would then be fixed by a remote past and the laws of nature—and with indeterminism, on the grounds that indeterministic outcomes reduce to mere chance. On the neuroscientific side, it revisits Libet-style experiments, often interpreted as showing that unconscious brain activity initiates voluntary actions before conscious intention, and critically reviews recent reinterpretations of the readiness potential and the limitations of such paradigms for assessing free will. The paper then diagnoses a shared structure in these challenges: they presuppose a strict dichotomy between Laplacean determinism and a thin, law-governed conception of chance that leaves no conceptual space for non-chance indeterminism or for agent-level causal contributions. A simple quantum thought experiment is used to show how microscopic indeterminism can have direct macroscopic effects, undermining the assumption that the macroworld is effectively deterministic. Finally, the implications of computational and dynamical models of cognition are considered, arguing that their built-in constraints should be read as limits of the models rather than as metaphysical results. The conclusion advocates a naturalistic agnosticism: current physics, neuroscience, and cognitive science neither establish nor refute free will, but underdetermine its status while still placing substantive constraints on any viable theory of it.

1. Introduction

Classic arguments from both philosophy and neuroscience have been taken to show that free will is impossible. On the philosophical side, van Inwagen’s consequence argument and related “no-choice” reasoning suggest that if the world is deterministic, our actions are fixed by the remote past and the laws of nature in a way that leaves us with no genuine alternatives [1,2,3]. On the empirical side, Libet-style experiments appear to show that neural precursors of action precede the reported time of conscious intention by several hundred milliseconds, encouraging the conclusion that the brain initiates voluntary action before consciousness “catches up,” and that conscious will is therefore epiphenomenal [4,5,6,7]. Taken together, these lines of thought have motivated a strong pessimism: whether the world is deterministic or indeterministic, and whether we look to abstract laws or concrete brain processes, free will appears to have no stable place [1,8].

This paper defends a more modest claim. I do not offer a new positive theory of free will, nor do I attempt to survey the vast contemporary literature. Instead, I argue that the strongest impossibility claims—those that conclude that free will is ruled out by determinism, by indeterminism, or by current neuroscience—are not warranted. Once we scrutinize their underlying assumptions about laws, chance, and the mind–brain relation, a form of naturalistic agnosticism about free will remains viable: we lack decisive reasons to believe that free will exists, but we equally lack decisive reasons to believe that it is impossible or incoherent given our best science [9,10].

At first sight, such a position risks looking like a “free-will-of-the-gaps” move, structurally analogous to the familiar “God-of-the-gaps” strategy in philosophy of religion: invoking a special kind of agency to occupy whatever phenomena science has not yet fully explained [11,12]. To avoid this, the present argument does three things. First, it identifies specific, substantive assumptions that drive leading impossibility arguments—for example, the treatment of indeterminism as nothing but unbiased randomness with fully known probabilities, or the reading of the readiness potential as a direct neural signature of preconscious decisions [1,2,5,13]. Second, it shows that these assumptions are neither forced by current physics and neuroscience nor neutral with respect to the very question at issue [7,14,15,16,17,18,19]. Third, it proposes positive constraints on any acceptable notion of free will: it must be naturalistically admissible; it must connect to the phenomenology of deliberation and choice; and it must mesh with our practices of moral appraisal, without collapsing into either Laplacean determinism or bare stochasticity [9,10].

The strategy can be summarized as follows. Section 2 presents, in compressed form, the main philosophical and neuroscientific arguments against free will that motivate impossibilist conclusions, focusing on van Inwagen’s deterministic and indeterministic arguments and on a representative Libet-style challenge, together with some recent reinterpretations of the readiness potential as a stochastic accumulator signal rather than a direct marker of decision onset [1,4,5,7,9,13,18]. Section 3 diagnoses the shared structure of these arguments. I argue that they rely on a dichotomy between strict determinism and a thin, law-governed conception of chance that leaves no conceptual space for non-chance indeterminism, or for agent-level causal contributions that are not simply reducible to microphysical initial conditions plus laws [10]. Section 4 develops a simple thought experiment, in which quantum-level indeterminacy is allowed to influence macroscopic behavior, to illustrate how such non-chance indeterminism might be modeled without invoking miracles or dualistic substances, and to situate this possibility against contemporary discussions of quantum indeterminism and the status of superdeterministic “escape routes” [14,15,17]. The point is not to show that free will exists, but to clarify how incompatibilist pessimism often depends on ruling out this kind of possibility by assumption rather than by argument.

Section 5 then considers how these reflections bear on two influential pictures of the mind: classical computationalism and dynamical systems theory. I suggest that, as usually formulated, both frameworks either build in determinism or leave free will outside the model as an unexplained remainder, but that this should be read as a limitation of the frameworks rather than as a discovery about human agency [20,21]. Finally, Section 6 draws the threads together. I sketch the contours of a naturalistic agnostic stance: one that acknowledges the force of the classic arguments and takes seriously the empirical work on consciousness and action, yet resists both complacent libertarianism and premature impossibilism [7,9,10]. On this view, free will is neither established nor refuted; it remains a live metaphysical option, constrained by physics and neuroscience but not eliminated by them.

The resulting position is unavoidably modest. It does not tell us that free will exists, nor does it locate it neatly in a particular neural mechanism or physical process. What it does, instead, is to shift the dialectical burden. Those who claim that free will is impossible—whether because of determinism, because of indeterminism, or because of the temporal order of neural and conscious events—owe us more than has so far been provided. Until such arguments are forthcoming, a scientifically informed agnosticism about free will remains a rational stance [2,3,9].

2. Determinism, Indeterminism, and Classical Challenges to Free Will

In this section I briefly review two canonical challenges to free will. The first is van Inwagen’s argument that free will is incompatible with both determinism and indeterminism [1,22,23,24]. The second is the family of Libet-style experiments, which are often taken to show that the brain initiates voluntary action before consciousness is aware of deciding [4,5]. Together, these have set much of the agenda for contemporary pessimism about free will in both philosophy and neuroscience.

Van Inwagen understands determinism as the thesis that, at every moment, the complete state of the world together with the laws of nature fixes a unique future. Given the actual past and the laws, only one future history is possible. His “no choice” or “consequence” argument then runs, roughly, as follows. First, no agent has (or ever had) any choice about the past. Second, no agent has any choice about the laws of nature. Third, if determinism is true, then every present and future action is a logical consequence of the past together with the laws. Fourth, if one has no choice about p, and no choice about (p → q), then one has no choice about q. The conclusion is that, under determinism, we never truly have a choice about what we do: our actions are consequences of factors entirely beyond our control [1,22].

This is an explicitly incompatibilist line: if determinism holds, there can be no robust sense in which an agent could have done otherwise, and thus no traditional, alternative-possibility conception of free will. At the same time, van Inwagen argues that simple indeterminism fares no better. If a decision at time t is not determined by prior circumstances plus laws, then either (a) it is partly a matter of chance which way the agent decides, or (b) the agent’s will itself plays a non-chancy causal role in producing the outcome. He presses the worry that in option (a), the agent seems not fully responsible for a chancy result, while in option (b) we risk reintroducing a kind of lawlike dependence that looks very much like determinism in disguise [23,24].

To sharpen the concern, van Inwagen asks us to imagine “replaying” the same decision situation many times under identical circumstances. If the outcome varies from one replay to the next, that looks like mere randomness; if it does not vary, the pattern begins to mimic determinism. Either way, he suggests, we do not obtain the kind of control that libertarian free will was supposed to secure. The pessimistic conclusion is that free will appears to be impossible whether or not the world is deterministic.

Frankfurt-style counterexamples to the principle of alternative possibilities complicate this picture but do not, in van Inwagen’s view, rescue robust freedom. In a typical Frankfurt case, a counterfactual intervener (Black) stands ready to ensure that the agent performs a certain action if the agent shows any sign of deciding otherwise, but as it happens, the agent freely chooses that action and the intervener never needs to act [25]. Many take such cases to show that moral responsibility does not strictly require the ability to do otherwise. Van Inwagen concedes that Frankfurt cases weaken the link between alternative possibilities and responsibility, yet he maintains that they do not vindicate genuine free will in a deterministic world: the actual decision still follows inevitably from the prior state and the laws, so the deeper worry remains.

Sabo [26] characterizes the resulting position as a kind of “mysterianism” about free will: van Inwagen is deeply sympathetic to the conviction that we are free, but he finds no coherent way to make sense of such freedom within any world governed by either deterministic or purely probabilistic laws. The upshot is a three-way pressure on the theorist. One can accept determinism and abandon a traditional libertarian conception of free will; one can reject free will altogether as incoherent; or one can insist that free will is real and incompatibilist, while admitting that we have no clear picture of how it could fit into a law-governed universe.

A structurally similar challenge arises from neuroscience, beginning with the experiments of Benjamin Libet and collaborators. In the standard Libet paradigm, participants are asked to perform a simple, spontaneous movement (such as flexing a wrist) whenever they feel like it, while monitoring the position of a dot on a clock-like display and later reporting the time at which they first formed the conscious intention to move. Meanwhile, EEG is used to record a slow buildup of negative potential over motor areas known as the readiness potential (RP) [4]. Libet found that the onset of the RP systematically preceded the reported time of conscious intention by several hundred milliseconds, suggesting that the brain was preparing the movement before the participant became aware of deciding. In later work, he interpreted this as evidence that the initiation of voluntary action is unconsciously determined, with consciousness at most retaining a limited “veto” power over already-initiated actions [5].

Much subsequent discussion has treated Libet’s findings as a direct empirical analogue to the philosophical impossibility arguments: if unconscious brain processes initiate action before conscious intention arises, then conscious will is not the true cause of our behavior, and free will—in any sense that requires conscious control—appears illusory [27]. On this reading, neuroscience does not merely make free will hard to reconcile with determinism or indeterminism; it shows that our own experience of deciding is systematically misleading.

However, more recent work has substantially complicated this picture. Schurger and colleagues proposed that the RP can be modeled as the average of stochastic fluctuations in neural activity accumulating toward a movement threshold, rather than as a dedicated preparatory signal that directly reflects a preconscious decision to move [13,18,28]. On this view, the timing of action in Libet-style tasks is partly governed by random drift, and the RP emerges as an artifact of time-locking and averaging across trials. At the same time, a growing body of critical work argues that Libet-style tasks involve highly constrained, trivial movements, rely on introspective timing reports that are themselves problematic, and do not straightforwardly generalize to the sort of extended deliberative choices that are central to our ordinary conception of free will [7,9].

The dialectical lesson for present purposes is not that Libet’s experiments are irrelevant, but that the strongest impossibilist interpretation of them depends on substantive assumptions: that the RP is a direct neural marker of decision onset, that simple motor choices faithfully capture the phenomena of interest, and that unconscious initiation of movement would automatically undermine all meaningful forms of agency. Just as van Inwagen’s arguments lean heavily on a particular way of carving up determinism and indeterminism, Libet-style challenges lean on a particular way of interpreting noisy neural signals and laboratory tasks. In later sections I will argue that once these assumptions are made explicit, their force against a suitably modest, naturalistic conception of free will is considerably weaker than is often supposed.

3. The Shared Structure: Determinism, Thin Chance, and Excluded Possibilities

The philosophical and neuroscientific challenges reviewed in Section 2 differ in subject matter and method, but they share a common argumentative structure. Both start from a sharp contrast between strict determinism and a very thin conception of chance, and then argue that free will cannot be reconciled with either horn of this disjunction. Once this dichotomy is in place, the conclusion that free will is impossible can seem almost inevitable.

In the philosophical case, van Inwagen assumes that if an action is not determined by the past and the laws, then it must involve an element of causally basic chance [1,22,23,24]. Under determinism, our actions are fixed by initial conditions and laws over which we have no control; under indeterminism, the relevant events are realized according to objective probabilities that likewise lie outside our control. Either way, the agent appears to be at the mercy of factors—deterministic or stochastic—that are not themselves shaped by her will. This is why van Inwagen worries that indeterministic “replays” of the same decision situation will look either randomly different or effectively deterministic, but never exhibit the kind of grounded control that libertarians want.

Libet-style challenges are structurally similar. Here the contrast is between unconscious neural determination and neural noise. If the readiness potential is read as a signature of an already-made, unconscious decision, then conscious intention comes too late to play a genuine causal role [4,5]. If, in light of accumulator models, the timing of action is partly shaped by stochastic fluctuations in neural activity, then the onset of movement looks to that extent like a matter of chance [13,18,28]. In either case, once we assume that the relevant neural processes are either strictly deterministic or merely noisy, the deliberate, conscious agent seems to have no distinctive work left to do [9,27].

The notion of chance at work in these arguments is thin in at least two respects. First, it is law-governed: given the microphysical state and the laws, there is a fixed probability distribution over outcomes, and nothing further to say about why one outcome rather than another is realized [14,15]. Second, it is agent-neutral: the probabilities are defined over physical events without reference to patterns of deliberation, reasons, or character. If an event is chancy in this sense, then from the point of view of agency nothing more informative can be said than that it had such-and-such a probability of occurring. It is therefore natural, on this picture, to identify indeterminism with a lack of control.

Once we adopt this thin conception of chance, the space of possibilities looks stark. Either actions are fully determined by prior states and laws, in which case they are not “up to us,” or they involve irreducible chance in the thin sense, in which case they are not “up to us” either. What this dichotomy excludes by construction is the possibility of a non-chance indeterminism: a mode of causal dependence in which events are not strictly determined by prior microphysical conditions, yet are still shaped in a principled way by the agent’s reasons, character, and higher-level organization.

One way to see that this excluded middle is at least conceptually available is to consider libertarian proposals such as Robert Kane’s theory of self-forming actions [10]. On Kane’s view, certain key decisions in a person’s life—the ones in which she struggles between competing motivations—are indeterministic in the brain, but not merely chancy in the thin sense. Indeterminism is located in neural processes underlying an effort of will that is simultaneously pulled in different, independently reasonable directions. Which way the effort goes is not fixed by the past, yet the alternatives are not arbitrary: each is supported by the agent’s standing values and reasons. The resulting outcome can thus be both undetermined and meaningfully attributable to the agent, rather than to brute chance alone. Whether or not Kane’s specific account succeeds, it illustrates a more general point: there is conceptual room for indeterministic causation that is structured by the agent’s reasons and character.

Relatedly, recent work in philosophy of agency has emphasized the idea that agent-level causal patterns may be partially autonomous from microphysical descriptions. Christian List, for example, argues that free will is a higher-level phenomenon characterized by intentional agency, alternative possibilities, and causal control, and that these properties can in principle coexist with a deterministic microphysics [29]. Jenann Ismael similarly maintains that a proper understanding of laws, causation, and higher-level organization in physics allows for a robust conception of agential control, even in a world that is, at the fundamental level, governed by strict laws and (perhaps) some quantum indeterminism [30]. These views are not libertarian in Kane’s sense, but they share the conviction that the space of options is richer than a simple choice between microphysical determinism and bare chance.

The present paper does not endorse any one of these positive accounts. Its aim is more modest: to show that the impossibilist arguments reviewed above tacitly rely on the assumption that there is no coherent notion of non-chance indeterminism or of agent-level causal contribution that is not simply reducible to microphysical initial conditions plus laws. Once this assumption is made explicit, the force of those arguments is diminished. They no longer demonstrate that free will is impossible; rather, they show that free will is incompatible with a particular, highly constrained way of thinking about laws, chance, and the relation between levels of description. In Section 4, I use a simple thought experiment to illustrate how indeterminism at the microlevel might influence macroscopic behavior without collapsing into either strict determinism or thin chance, thereby giving more substance to the idea of non-chance indeterminism.

4. Quantum Indeterminism and a Thought Experiment

Section 3 argued that standard impossibilist arguments rely on a dichotomy between strict determinism and a thin, law-governed notion of chance. On that picture, the microphysical world is either fully determined by the laws and initial conditions, or else it evolves according to fixed probability distributions that are entirely agent-neutral. Either way, it is natural to assume that indeterminism at the quantum level “washes out” when we move to the scales at which agency and consciousness operate, leaving the macroscopic world effectively deterministic. In this section I use a simple quantum thought experiment to challenge that picture. The point is not to show that free will exists, but to illustrate how quantum indeterminism can have direct, causally efficacious macroscopic consequences in a way that undermines naïve mechanistic determinism.

Consider a single atom prepared in an excited state and placed in a small transparent vessel. According to standard quantum theory, the atom has a certain decay law specifying the probability that it will emit a photon in a given time interval, and the emission direction is distributed isotropically: all directions are equally likely. The exact emission time and direction are not determined by any prior state of the world; they are treated as irreducibly indeterministic [14,15,31]. Now place a human observer at some distance from the vessel in a darkened environment. Psychophysical and modern quantum-optical experiments suggest that, under suitable conditions, the human visual system is capable of detecting the arrival of a single photon at the cornea with a probability significantly above chance [32,33].

The geometry is straightforward. Let the distance from the atom to the observer’s eye be r, and let the pupil have area A. If the photon is emitted in a random direction, the probability that it enters the pupil is approximately $A/\left(4\pi r^2\right)$. For typical experimental parameters this probability is small but non-zero. Suppose we now instruct the observer: “If you see a flash (i.e., detect the photon), immediately press a button or shout ‘Now!’; if you do not see anything, do nothing.” In this setup, a macroscopic action—pressing the button or shouting—is tightly coupled to the occurrence of a single quantum event: whether a particular photon happens to be emitted in the narrow solid angle subtended by the observer’s pupil and at a time when it can be detected.

Two features of this scenario are important. First, given the usual indeterministic reading of quantum mechanics, there is no complete description of the prior state of the world (at either micro or macro scales) that fixes both whether and in which direction the photon will be emitted. The best we can do is assign probabilities. Second, the observer’s response is, by design, conditionally deterministic: if the photon is detected, the button will be pressed; if not, it will not. In this way, a genuinely indeterministic micro-event has a direct and observable influence on macroscopic behavior. From the perspective of an external observer who knows the experimental setup and the subject’s instructions, it is impossible—even in principle—to predict with certainty whether the button will be pressed at a given trial on the basis of earlier states of the world. The unpredictability is not due to ignorance of hidden classical variables, but to the structure of the quantum theory itself [14,15].

This suffices to undermine a purely mechanistic determinist picture of the world. Even if many large-scale phenomena are well described by approximately deterministic laws, there are possible (and experimentally realizable) situations in which macroscopic events turn crucially on irreducibly indeterministic quantum occurrences. If free will does not exist in such a world, it is not because the world is governed everywhere by Laplacean determinism. The failure of free will, if it fails, would have to be explained by something other than an all-pervasive deterministic dynamics.

Of course, one might insist that this example shows nothing relevant to free will, because the subject’s action is still driven by pure chance at the micro level: whether the photon arrives is not “up to” the subject in any robust sense. That is correct as far as it goes. The thought experiment is not offered as a model of action, but as a way of decoupling two questions that are often run together: (i) whether indeterminism at the micro level can have immediate macroscopic causal impact; (ii) whether there can be a form of indeterminism that is more than mere thin chance. The experiment addresses (i): it shows that the common assumption that quantum indeterminism is causally irrelevant at the scales of agency is unwarranted. Question (ii) remains open and is taken up in Section 5.

A potential rejoinder appeals to superdeterminism: the hypothesis that, contrary to the usual reading of Bell’s theorem, there are underlying deterministic hidden variables that fix not only the outcomes of quantum measurements but also the settings and even the experimenters’ “choices” [14,15,17,34]. In a superdeterministic model, the apparent randomness in the emission direction of our photon would be explained by unknown variables that also determine the subject’s instructions and reactions, restoring global determinism at the cost of positing deep correlations between seemingly independent events. Whether such models are viable or scientifically attractive is a matter of ongoing debate; critics argue that they threaten to make experimental intervention and statistical independence mysterious, or even undermine the ordinary logic of hypothesis testing [17,35,36]. For present purposes, the important point is that impossibilist arguments generally do not rely on superdeterminism. They take standard indeterministic quantum mechanics at face value, and then argue that this indeterminism is either irrelevant (because it averages out) or inimical (because it reduces everything to chance).

The lesson of the thought experiment is modest but significant. In a world governed by our best current physics, we can couple quantum-level indeterminism directly to macroscopic events, including overt actions. This shows that the familiar picture—determinism at the macro level, pure chance safely quarantined at the micro level—is too simple. Once that picture is given up, there is conceptual room for richer possibilities: for forms of non-chance indeterminism in which higher-level structures (agents, reasons, policies) interact with indeterministic micro-processes in ways that are neither strictly deterministic nor mere thin chance [10,37]. The present paper does not claim that such possibilities are realized, let alone that they underwrite genuine free will. It claims only that impossibilist arguments which foreclose them by assumption rather than argument do not justify their strong conclusions. In Section 5, I use this opening to discuss how indeterminism might be structured in an agential context without collapsing into either mechanistic determinism or brute randomness.

5. Computationalism, Dynamical Systems, and the Limits of Current Models

The discussion so far has focused on arguments that start from broad assumptions about laws, chance, and microphysical states. A different, but related, line of pressure comes from influential frameworks in cognitive science—especially classical computationalism and dynamical systems theory (DST). These are often taken to show, or at least strongly suggest, that mental activity is either fully determined by prior states and inputs or else subject only to thin, law-governed noise. In this section I argue that, as usually formulated, both frameworks either build determinism in by design or leave free will outside the model as an unspecified remainder. This should be read as a limitation of the frameworks, not as a discovery about human agency.

A further distinction is needed between the fundamental physical level and the classical, epistemic level at which human agents perceive, deliberate, and act. Even if nature is quantum-mechanical at its base, human perception and cognition do not operate with access to the complete microphysical state of the world. They operate through coarse-grained, incomplete, and classically organized information. This matters because deliberative choice belongs to the agent’s accessible world: agents confront alternatives as practical possibilities structured by reasons, values, uncertainty, and limited information, not as fully transparent microphysical trajectories. Thus, even if a deeper physical description were ultimately deterministic or probabilistic, it would not follow that human decision-making is experienced or modeled, at the agential level, as already settled. The point is not that incomplete information proves free will, but that it blocks a simple inference from fundamental physics to the impossibility of choice.

On classical computationalism, the mind is essentially a digital computer: cognitive processes are implementations of algorithms defined over internal representations, and mental states are individuated by their functional role in a causal network [20,38,39,40]. At the core of this picture is a Turing-style model of computation in which, given a program and an input, the output is determined. This is true even for many probabilistic algorithms: randomness is typically introduced by pseudo-random number generators or, more abstractly, by assigning a fixed probability distribution to a class of transitions. In either case, the computational architecture enforces the same dichotomy discussed earlier: internal transitions are either strictly deterministic or governed by a thin, law-like notion of chance. From within such a framework, there is no conceptual space for the kind of non-chance indeterminism that might be relevant to free will; the only options are Laplacean predictability or formally tractable randomness.

This does not show that free will is impossible. It shows that purely classical computational models of minds are built to capture a certain range of regularities—those that can be described as algorithmic transformations from inputs and internal states to outputs. If free agency requires forms of causal dependence that are not exhausted by such transformations, a purely computational description will either assimilate them to pseudo-random noise or leave them out of the model altogether. In that case, the right conclusion is not that free will does not exist, but that a strict computationalist account is, at best, incomplete. One could, in principle, move to more exotic forms of computation—probabilistic, analog, or quantum [41,42]—but as long as the framework treats randomness as thin chance and higher-level agency as epiphenomenal on state transitions, the structural limitations remain.

Dynamical systems theory offers a contrasting image of mind and brain. Instead of viewing cognition as discrete symbol manipulation, DST models cognitive systems as high-dimensional, often non-linear dynamical systems whose states evolve continuously over time according to differential or difference equations [21,43,44,45]. This approach has illuminated a wide variety of phenomena—motor coordination, sensorimotor coupling, development—by emphasizing attractors, phase transitions, and the embedding of neural activity in organism–environment loops. It is sometimes suggested that DST, with its sensitivity to initial conditions and richly structured state spaces, offers a more hospitable home for free will than mechanistic computationalism.

However, standard DST models are typically deterministic at the level of their equations of motion. Given an initial point in the system’s state space and a fixed set of parameters, the future trajectory is fixed, even if it is chaotic and practically unpredictable. Unpredictability here comes from sensitive dependence on initial conditions, not from ontological indeterminism. When stochastic terms are added to the equations to model noise, they are again treated as thin chance: random perturbations with specified probability distributions. As in the computational case, the underlying mathematical framework leaves no conceptual room for non-chance indeterminism. The system’s evolution is either fully determined or it is driven, in part, by structureless noise [21,45].

What DST does offer is a powerful way of representing higher-level patterns—attractors, manifolds, metastable regimes—that can be interpreted as cognitive or agential states. A trajectory lingering near one attractor rather than another may correspond to pursuing one intention rather than another; a bifurcation might mark a decision point. But the free-will question then reappears at a different level: are these agential patterns merely high-level labels for an underlying deterministic (or thinly stochastic) flow, or is there some genuine sense in which the agent, as an organized whole, contributes causally to which attractor basin is entered? If one insists that all causation is wholly captured by the underlying equations and noise terms, then free will—if it exists—will by definition lie outside the model; if one allows agent-level causal descriptions to be partially autonomous (as suggested by [29,30]), then neither computationalism nor DST, as standardly construed, can be the whole story.

The aim here is not to refute computationalism or DST as research programs. Both have delivered impressive explanatory successes and remain indispensable in cognitive science and neuroscience. Rather, the point is dialectical. When these frameworks are used to argue for the impossibility of free will—on the grounds that “the mind is just a computer” or “the brain is just a dynamical system”—the argument typically relies on the same assumptions diagnosed in earlier sections: that all relevant causal structure is either deterministic or exhausted by thin chance, and that higher-level agential patterns are mere summaries of lower-level dynamics. Once those assumptions are made explicit and opened to question, the move from “cognition is computational/dynamical” to “free will is impossible” loses much of its force. At most, the frameworks show that a certain idealized, level-specific description of the mind leaves no room for free will. They do not show that reality itself does.

On the view defended in this paper, this is another instance where agnosticism is the appropriate response. We need not deny what our best computational and dynamical models tell us about the structure of cognition; but we should also resist reading the limitations of those models as metaphysical results about what forms of agency are possible. If there is such a thing as non-chance indeterminism or agent-level causal contribution, it will not be visible from within a framework that, by design, has no variables or laws to represent it. That is a reason to be cautious about impossibilist inferences, not a reason to abandon naturalistic free-will theorizing altogether.

6. Conclusion: Toward a Naturalistic Agnosticism

The arguments surveyed in this paper—van Inwagen’s incompatibilist case against both determinism and indeterminism, Libet-style challenges from neuroscience, and the pressures from computational and dynamical models of cognition—have considerable force. Taken at face value, they suggest that whether the world is deterministic or indeterministic, and whether we look at abstract laws or concrete brain processes, there seems to be no stable place for free will. It is therefore unsurprising that many philosophers and scientists have concluded that free will is either impossible or illusory.

I have argued, however, that the strongest impossibilist conclusions depend on substantive assumptions that are neither forced by our best physics and neuroscience nor neutral with respect to the question at issue. On the philosophical side, van Inwagen’s arguments presuppose a dichotomy between strict determinism and a thin, law-governed conception of chance that leaves no conceptual room for non-chance indeterminism or for agent-level causal contributions that are not simply reducible to microphysical initial conditions plus laws. On the neuroscientific side, Libet-style arguments presuppose that simple motor tasks with problematic introspective timing can stand in for the full range of human decision-making, and that the readiness potential is either a direct neural marker of decision onset or a mere reflection of thin noise. On the cognitive modelling side, classical computationalism and standard dynamical systems theory build determinism or thin stochasticity into their mathematical frameworks from the outset, and then unsurprisingly fail to register any richer forms of agency.

Once these assumptions are made explicit, the landscape looks different. Quantum theory, interpreted in the usual indeterministic way, already undermines a naïve mechanistic determinism: there are experimentally realizable scenarios in which single quantum events can be coupled directly to macroscopic actions. This does not by itself yield free will, but it shows that the often-invoked contrast between a deterministic macroworld and a safely quarantined realm of micro-level chance is overly simple. Philosophical work on libertarian agency and higher-level causation suggests further that the space of possibilities is richer than a strict choice between Laplacean determinism and brute randomness [10,29,30]. There is at least a conceptual opening for non-chance indeterminism and for partially autonomous agent-level descriptions that are compatible with, but not exhausted by, lower-level dynamics.

This brings us to the stance I have called naturalistic agnosticism about free will. On this view, we take fully seriously the best current science of the brain and of physical law, and we acknowledge the depth of the puzzles that have motivated impossibilist conclusions. At the same time, we resist drawing strong metaphysical inferences from the limitations of particular experimental paradigms or modelling frameworks. We do not claim to know that free will exists, but we also do not claim to know that it is impossible or incoherent. Instead, we treat free will as a live metaphysical option—one that must be constrained by physics and neuroscience, but that is not ruled out by them in their present state [7,9,10].

It is crucial that this position not collapse into a simple “free-will-of-the-gaps” strategy. A pure gaps move would say, in effect, that wherever current science is silent, free will can be safely inserted, only to retreat as explanations accumulate. Naturalistic agnosticism is different in two ways. First, it imposes positive constraints on any acceptable conception of free will: it must be naturalistically admissible, must bear a principled relation to the phenomenology of deliberation and choice, and must mesh with our practices of moral appraisal without invoking miracles or dualistic substances. Second, it is open to the possibility that future empirical and theoretical developments will tell against free will, for example by showing that all plausible candidates for non-chance indeterminism or agent-level causal autonomy are untenable. Agnosticism here is not a refuge from evidence, but a recognition that current evidence does not yet justify either complacent libertarianism or confident impossibilism.

This modest conclusion may disappoint those who hoped for a definitive vindication or refutation of free will. Yet it has its own theoretical virtues. It reframes the debate in a way that is continuous with ordinary scientific practice: we acknowledge the underdetermination of metaphysical claims by available data; we distinguish what our models explicitly contain from what they leave out by design; and we treat strong impossibility claims with caution when they depend on idealizations whose scope and limitations are themselves contestable. In this light, the burden of proof shifts. Those who maintain that free will is impossible—not merely unlikely or philosophically suspect, but incompatible with the best science—owe us arguments that do not rely on foreclosing, by stipulation, the very conceptual possibilities that might allow free will to exist.

In the meantime, our lived practices of deliberation, commitment, and moral evaluation remain. A naturalistic agnostic can accept that these practices may ultimately require revision in light of deeper empirical and theoretical understanding; but she need not regard them as already falsified by current arguments from determinism, indeterminism, or neuroscience. Free will, on this view, is neither established nor refuted. It remains an open question—one that spans physics, neuroscience, and philosophy, and that calls for continued exploration rather than premature closure.