Nobody was holding a gun to your head when you started reading this. You made a choice. Surely it felt that way, at least. A sense of agency—of control over our actions, of continual decision-making—is part of the experience of being human, moment by moment and day by day. True, we sometimes just drift, like robots or zombies, but at other times we gird our loins and exert our will. David Hume defined will nearly three centuries ago as “the internal impression we feel and are conscious of, when we knowingly give rise to any new motion of our body, or new perception of our mind.” The feeling was universal then and it’s universal now.
Yet a peculiar fact about the state of the sciences in the early twenty-first century is that many authorities—physicists, neuroscientists, and even philosophers—will tell you that this sense of agency is an illusion. In their daily lives, these same experts pick out clothing, choose wallpaper, and order from restaurant menus, but when they study the matter professionally they doubt that they have chosen freely. They understand “free will” to be a feeling people have, but no more than that.
For physicists, the problem is that we are made of matter, like every particle and planet in the universe, and matter is governed by physical laws. According to the physicist and best-selling author Brian Greene, “We need to recognize that although the sensation of free will is real, the capacity to exert free will—the capacity for the human mind to transcend the laws that control physical progression—is not.” We do not and cannot cause anything; we are caused. “Our choices are the result of our particles coursing one way or another through our brains,” he writes.
Our actions are the result of our particles moving this way or that through our bodies. And all particle motion—whether in a brain, a body, or a baseball—is controlled by physics and so is fully dictated by mathematical decree.
It is a stern and final lesson: “We are no more than playthings knocked to and fro by the dispassionate rules of the cosmos.” Nothing to see here. Move along.
Brain scientists, too, doubt free will and look for root causes—mechanisms underlying behavior—in the material substrate of what we like to call our minds. This is reductionism: as physicists begin at the bottom, with elementary particles, so neuroscientists look to neurons. They tend to reach the same conclusion: volition is the end, not the cause, of a chain of electrical and chemical activity. Our desires, intentions, and plans float above the engine room, the systems of the brain where the real work is done.
Some people resist the arguments of physics by resorting to stubborn faith, and it’s hard to blame them. The mathematical philosopher Martin Gardner remained persuaded that “somehow, in a way utterly beyond our ken, you and I possess that incomprehensible power we call free will,” but he gave up trying to explain. “Like time, with which it is linked, free will is best left—indeed, I believe we cannot do otherwise—an impenetrable mystery. Ask not how it works because no one on earth can tell you.”
The term “free will” carries a lot of baggage; the constraints of nature and nurture, our genes and our unconscious habits, our family histories and social conditions all help determine our behavior and thus make us less than fully free. The more general term is “agency,” the capacity for purposeful action. Terminology notwithstanding, the conviction that we act with some degree of freedom features not just in our private thoughts but in our public life. Legal institutions, theories of government, and economic systems are built on the assumption that humans make choices and strive to influence the choices of others. Without some kind of free will, politics has no point. Nor does sports. Or anything, really.
Nonetheless, Sam Harris, a neuroscientist and philosopher who wrote the popular book Free Will (2012), insisted not only that free will is an illusion but that the concept “cannot be made conceptually coherent.” Consider it a challenge: “No one has ever described a way in which mental and physical processes could arise that would attest to the existence of such freedom.”
Kevin J. Mitchell answers exactly this challenge in Free Agents: How Evolution Gave Us Free Will. A neuroscientist and geneticist at Trinity College Dublin, Mitchell sets out to rescue our intuitive sense of agency from a cloud of obfuscation. Yes, he says, free will exists. It is neither an illusion nor merely a figure of speech. It is our essential, defining quality and as such demands explanation. “We make decisions, we choose, we act,” he declares.
These are the fundamental truths of our existence and absolutely the most basic phenomenology of our lives. If science seems to be suggesting otherwise, the correct response is not to throw our hands up and say, “Well, I guess everything we thought about our own existence is a laughable delusion.” It is to accept instead that there is a deep mystery to be solved and to realize that we may need to question the philosophical bedrock of our scientific approach if we are to reconcile the clear existence of choice with the apparent determinism of the physical universe.
Agency distinguishes even bacteria from the otherwise lifeless universe. Living things are “imbued with purpose and able to act on their own terms,” Mitchell says. He makes a powerful case that the history of life, in all its complex grandeur, cannot be appreciated until we understand the evolution of agency—and then, in creatures of sufficient complexity, the evolution of conscious free will.
Mitchell is one of a new breed of biologists who espouse a complex-systems perspective as an antidote to reductionism. He aims to reclaim from the philosophers words like purpose, reason, and meaning, which scientists often avoid as being unquantifiable. He mostly eschews jargon. This is a plainspoken book. It gets mildly technical in matters of biology and neuroscience, but it builds an argument that is methodical and crisp, and it cuts through years of disputation like a knife through cotton candy. This is what you are, Mitchell asserts: “You are the type of thing that can take action, that can make decisions, that can be a causal force in the world: you are an agent.”
If the denial of free will has been an error, it has not been a harmless one. Its message is grim and etiolating. It drains purpose and dignity from our sense of ourselves and, for that matter, of our fellow living creatures. It releases us from responsibility and treats us as passive objects, like billiard balls or falling leaves.
What Is Life? was an influential little book by the quantum pioneer Erwin Schrödinger, assembled from lectures he delivered in Dublin in 1943. When can we say that a thing is alive? He gave a surprising answer:
When it goes on “doing something,” moving, exchanging material with its environment, and so forth, and that for a much longer period than we would expect an inanimate piece of matter to “keep going” under similar circumstances.
Notice the emphasis on time. First a living organism has to persist. It does this in defiance of the second law of thermodynamics, which says that the universe and its contents tend inexorably toward disorder. Left alone, a sandcastle degrades to a pile of sand. Cream disperses into the coffee. Everything in a closed system arrives at the same temperature, because entropy’s disorder also implies equilibrium. Against this universal tendency, the organism fights back. It sucks order out of disorder.
“The organism is not a pattern of stuff,” Mitchell says; “it is a pattern of interacting processes, and the self is that pattern persisting.” Within the first single-celled organisms the raw materials were already in place for the mechanism of replication famously described by James Watson and Francis Crick in 1953. Nucleic acids—RNA and DNA—act as templates, complex macromolecules storing information in a coded sequence. The code in DNA remains chemically stable, while the RNA molecules read its information and replicate it. Cells could divide, making copies of themselves. Then they could evolve.
This part of the story is well known. Populations of microorganisms compete for resources. Random errors in the transcription process create mutations. Some organisms compete more effectively than others, and thus nature selects the fittest to survive. It’s worth noting that all the elementary particles engaged in this activity are obeying the laws of motion, but the processes that interest us—metabolism, reproduction, mutation—take place on a different scale of complexity and abstraction. It may seem paradoxical, but just because laws of physics apply to everything doesn’t mean they explain everything. Sometimes they just aren’t the right tool. The equations of particle physicists don’t explain evolution any more than they explain genes or epidemics.
Biological entities develop across time, and as they do, they store and exchange information. “That extension through time generates a new kind of causation that is not seen in most physical processes,” Mitchell says, “one based on a record of history in which information about past events continues to play a causal role in the present.” Within even a single-celled organism, proteins in the cell wall respond chemically to changing conditions outside and thus act as sensors. Inside, proteins are activated and deactivated by biochemical reactions, and the organism effectively reconfigures its own metabolic pathways in order to survive. Those pathways can act as logic gates in a computer: if the conditions are X, then do A.
“They’re not thinking about it, of course,” Mitchell says, “but that is the effect, and it’s built right into the design of the molecule.” As organisms grow more complex, so do these logical pathways. They create feedback mechanisms, positive and negative. They make molecular clocks, responding to and then mimicking the solar cycle. Increasingly, they embody knowledge of the world in which they live.
The tiniest microorganisms also developed means of propulsion by changing their shape or deploying cilia and flagella, tiny vibrating hairs. The ability to move, combined with the ability to sense surroundings, created new possibilities—seeking food, escaping danger—continually amplified by natural selection. We begin to see organisms extracting information from their environment, acting on it in the present, and reproducing it for the future. “Information thus has causal power in the system,” Mitchell says, “and gives the agent causal power in the world.”
We can begin to talk about purpose. First of all, organisms struggle to maintain themselves. They strive to persist and then to reproduce. Natural selection ensures it. “The universe doesn’t have purpose, but life does,” Mitchell says.
And unlike the designed machines and gadgets that surround us in our daily lives, which also have a purpose or at least serve a purpose, living organisms are adapted for the sake of only one thing—their selves. This brings something new to the universe: a frame of reference, a subject. The existence of a goal imbues things with properties that previously never existed relative to that goal: function, meaning, and value.
And yet—impressed though we may be when we contemplate the paramecium, its oblong single cell covered in motile hairs, spiraling through the water in response to electrical signals sent by ion receptors, gathering food and avoiding obstacles and even forming symbiotic relationships with other organisms—no one would say that it has will, free or otherwise.
A determinist believes that whatever happens had to happen. The laws of nature carry the present into the future like gears in an unyielding machine or like the sequential states of a computer. It’s no mystery why physicists are drawn to determinism: the laws of nature are their bread and butter. The canonical expression of scientific determinism came from Pierre-Simon Laplace, an enthusiastic disciple of Newton:
An intelligence knowing all the forces acting in nature at a given instant, as well as the momentary positions of all things in the universe, would be able to comprehend in one single formula the motions of the largest bodies as well as the lightest atoms in the world, provided that its intellect were sufficiently powerful to subject all data to analysis; to it nothing would be uncertain, the future as well as the past would be present to its eyes.
As far as the equations of motion are concerned, the future and the past look the same. Einstein formalized this picture two centuries later when he envisioned the universe as a four-dimensional space-time continuum. “Everything is determined,” he said,
the beginning as well as the end, by forces over which we have no control. It is determined for the insect as well as for the star. Human being, vegetables or cosmic dust, we all dance to an invisible tune, intoned in the distance by a mysterious player.
Is there no room for slippage in the gears? As it happens, there is. When physicists attempt to carry out Laplace’s program, they discover that they cannot perfectly specify the state of even the smallest, simplest particle. This is the famous uncertainty principle: Werner Heisenberg established in the 1920s that the more precisely one determines a particle’s position, the less one can specify its momentum. This is sometimes discussed as a matter of epistemology, of what an observer can know, but the problem is more fundamental. No human needs to be part of the picture. The uncertainty is a feature of the universe.
Theorists handled this troublesome discovery by replacing Newtonian mechanics with a new mathematical system. Quantum mechanics treats particles as waves of probability via a wave function, using an equation that Schrödinger devised. The Schrödinger equation enables physicists to calculate—with astonishing success—how a quantum system evolves over time. Like Newton’s laws of motion, the Schrödinger equation is deterministic in form. When physicists rerun the calculation, they always get the same answer, perforce. Once again, in quantum mechanics the current state of the universe seems to determine the next state.
This is why so many modern physicists continue to embrace philosophical determinism. But their theories are deterministic because they’ve written them that way. We say that the laws govern the universe, but that is a metaphor; it is better to say that the laws describe what is known. In a way the mistake begins with the word “laws.” The laws aren’t instructions for nature to follow. Saying that the world is “controlled” by physics—that everything is “dictated” by mathematics—is putting the cart before the horse. Nature comes first. The laws are a model, a simplified description of a complex reality. No matter how successful, they necessarily remain incomplete and provisional.
Furthermore, quantum calculations of the wave function produce not a specific result but rather a probability distribution—hence the predicament of Schrödinger’s hapless cat, neither dead nor alive until the wave function “collapses.” Some physicists do agree that indeterminacy cannot be swept away but remains inherent at every level. “The upshot of these views is that the future is open: indeed, that is what makes it the future,” Mitchell writes.
Because we only inhabit the present we don’t experience this indeterminacy first-hand…. If we could really glimpse the future, we would see a world out of focus. Not separate paths already neatly laid out, waiting to be chosen—just a fuzzy, jittery picture that gets fuzzier and jitterier the further into the future you look.
Between the strict regime of physical determinism and the naive feeling of free will is an uneasy territory where we find many leading philosophers. What if, they ask, we could describe some version of free will, or something like free will, that is compatible with determinism? This approach is called compatibilism. Compatibilists argue that even if we accept that the future is already fixed, we can still talk about psychological freedom in a way that preserves concepts like moral responsibility.
Daniel Dennett made the modern case for compatibilism in his 1984 book Elbow Room: The Varieties of Free Will Worth Wanting—as he put it, he was “saving everything that mattered about the everyday concept of free will, while jettisoning the impediments.” (Updating Elbow Room in 2014, he expressed some frustration with the never-ending debate: “It is fair to say that I underestimated the persistence of some of the ideas I sought to dismantle and discredit in the 1980s.”) Compatibilism comes in many flavors, but the essential point is to set physics aside, leave it quietly in the corner, and study the ways in which people speak of free will, the deep questions on which it touches, the consequences for behavior and ethics—in short, to continue the investigation of free will as if the universe had room for such a thing.
Mitchell could have made his argument a compatibilist one. Dennett did something like that, exploring many of the same themes, in his 2017 book From Bacteria to Bach and Back: The Evolution of Minds. But Mitchell finds it unsatisfying to act as if “a perspectival shift is all that is needed to get us out of the metaphysical hole we seem to find ourselves in.” He wants to say, yes, we live in a materialistic universe; yes, the laws of physics apply; yet the future is not yet written, and living things have the power to change it.
Rejecting the reductionist view does not mean resorting to mind–body dualism—positing some extra, nonphysical entity, like a soul or a spirit. There is no ghost in this machine. “Our minds are not an extra layer sitting above our physical brains,” Mitchell says. They are the holistic sum of that continuous, dynamic, distributed activity.
The brain is material, and its parts are increasingly well understood. Where the earliest organisms had protein sensors and ion pathways to perform the most basic kind of information processing, we have networks of neurons firing electrical signals that excite or inhibit others, organized by the millions in columns and sheets—“levels and levels of internal processing in which information is being processed, parsed, and transformed from each cortical area to the next,” as Mitchell says. On top of information about smell and touch come increasingly complex signals from the visual and auditory cortices. These inputs are combined and layered to reveal higher-order relationships, and in this way organisms build up internal models of the external world.
It’s still just chemistry and electricity, but the state of the brain at one instant does not lead inexorably to the next. Mitchell emphasizes the inherent noisiness of the system: more or less random fluctuations that occur in an assemblage of “wet, jiggly, incomprehensibly tiny components that jitter about constantly.” He believes that the noise is not just inevitable; it’s useful. It has adaptive value for organisms that live, after all, in an environment subject to change and surprise. “The challenges facing organisms vary from moment to moment,” he notes, “and the nervous system has to cope with that volatility: that is precisely what it is specialized to do.” But merely adding randomness to a deterministic machine still doesn’t produce anything we would call free will.
Free will, as distinct from agency, implies consciousness and self-reflection. Yet so much of what we do is involuntary. Many neurologists see involuntary behavior as the norm and the sensation of willing as a sometime adjunct. They have a litany of examples of action disconnected from will. We breathe, we blink, we daydream, we scratch, we blush, we reach for a glass, we drift to sleep (easier than willing ourselves to sleep), we walk the same familiar route, all without a moment’s thought. Memories appear unbidden. Daniel M. Wegner illustrated his influential 2002 text The Illusion of Conscious Will with a picture of Dr. Strangelove (played by Peter Sellers in Stanley Kubrick’s film), whose black-gloved right hand kept shooting up in an involuntary Nazi salute. “Alien hand syndrome,” Wegner explains, is a genuine disorder “in which a person experiences one hand as operating with a mind of its own.” The hand acts contrary to the patient’s conscious intent, at least as the patient perceives it.
We may think it’s normal for action to coincide with will, but Wegner argues otherwise:
They come apart often enough to make one wonder whether they may be produced by separate systems in the mind…. As soon as we accept the idea that the will should be understood as an experience of the person who acts, we realize that conscious will is not inherent in action—there are actions that have it and actions that don’t.
People have multiple “personalities.” They say, “I am torn.” They have an angel on one shoulder and a devil on the other.
Give the free-will deniers their due: in all these ways and more, the idea of our conscious self as a trustworthy and competent master of our destiny—a pilot in the cockpit—has badly frayed. Even on our best days we’re subject to delusion and confusion. We act without thinking, from habit or reflex or instinct. We behave impulsively, for no reasons we can discern. Yet unconscious decision-making is still decision-making. And sometimes we do think. We reflect, ponder, dither, weigh alternatives for some time before choosing to act.
A touchstone for neuroscientists who doubt free will is a series of controversial experiments conducted by Benjamin Libet in the 1980s. Libet, a neuroscientist at the University of California, San Francisco, attached electrodes to participants’ scalps and asked them to move a finger whenever they chose and to report the instant of making the decision. He found that brain activity relating to the finger began many milliseconds before the awareness of any decision. If the conscious decision came after the action, how could it be the cause? “The position of conscious will in the time line suggests perhaps that the experience of will is a link in a causal chain leading to action, but in fact it might not even be that,” Wegner wrote. “It might just be a loose end—one of those things, like the action, that is caused by prior brain and mental events.”
But somebody moved those fingers. No one suggests they had marionette strings attached. Sam Harris says that “I” don’t choose; choices are made “by events in my brain that I, as the conscious witness of my thoughts and actions, could not inspect or influence.” But where is the line that separates events in my brain from the conscious witness? Let’s say instead: We make choices. We make decisions. Some of our decision-making is prolonged and thoughtful, while some is spontaneous and practically random. We may understand our choices, we may rationalize them, or they may remain mysterious and obscure.
Mitchell proposes what he calls a “more naturalized concept of the self.” We are not just our consciousness; we’re the organism, taken as a whole. We do things for reasons based on our histories, and “those reasons inhere at the level of the whole organism.” Much of the time, perhaps most of the time, our conscious self is not in control. Still, when the occasion requires, we can gather our wits, as the expression goes. We have so many expressions like that—get a grip; pull yourself together; focus your thoughts—metaphors for the indistinct things we see when we look inward. We don’t ask who is gathering whose wits.
Mitchell points out that the Libet experiment was designed to encourage randomness: subjects were told to “let the urge to act appear on its own at any time without any preplanning or concentration on when to act.” But some decisions may well be unconsidered, spontaneous, or even random, while others involve careful deliberation:
Overall then, Libet’s experiments have very little relevance for the question of free will. They do not relate to deliberative decisions at all…. Instead, they confirm, first, that neural activity in the brain is not completely deterministic and, second, that organisms can choose to harness the inherent randomness to make arbitrary decisions in a timely fashion. It is likely that we do this all the time, without being aware of it.
Indeed, some degree of randomness is essential to Mitchell’s neural model for agency and decision-making. He lays out a two-stage model: the gathering of options—possible actions for the organism to take—followed by a process of selection. For us, organisms capable of conscious free will, the options arise as patterns of activity in the cerebral cortex, always subject to fluctuations and noise. We may experience this as “ideas just ‘occurring to you.’” Then the brain evaluates these options, with “up-voting” and “down-voting,” by means of “interlocking circuit loops among the cortex, basal ganglia, thalamus, and midbrain.” In that way, selection employs goals and beliefs built from experience, stored in memory, and still more or less malleable.
The primitive single-celled organism moves and eats without thinking. We humans also move and eat, and we think about it. On the way from there to here, nature created organisms of increasing ability. Multicellular creatures developed specialized morphology, including linked neurons and grouped muscle cells, communicating with one another through chemical signals. Organs for vision and hearing became valuable means of gathering information about the environment. Complex networks of neurons gained the ability to store symbolic representations of the world and its parts, including, eventually, representations of the self, distinguished from everything else. All this happened; there’s nothing controversial about it.
We can compare different organisms by looking at their cognitive depth. Humans are admirably deep. “If a nematode worm could be said to be thinking, it’s certainly not thinking about much,” Mitchell says.
It may integrate a few signals at a time and can do simple forms of learning, but it doesn’t create much of a map of the world or its own self and doesn’t do any kind of long-term cognition. It inhabits the here and now.
We’re not like that, and we know it. Our mental processes are seldom algorithmic, changing their states one step at a time. Thought involves continual feedback and self-correction, and the individual components cannot be teased apart. Mitchell writes:
The various subsystems involved are in constant dialogue with each other, each attempting to satisfy its own constraints in the context of the dynamically changing information it receives from all the interconnected areas.
He draws from the computer scientist Douglas Hofstadter the concept of cognitive loops—recursive representations of other representations, with feedback—from which arise the ability to think about thinking, to reason about reasons.
These capabilities required more than just increased size or computing power. As brains evolved, a convoluted processing hierarchy arose: “As the cortical sheet expands, there is a tendency for existing areas to split into two, creating new areas that can act as new levels of the processing hierarchy.” These new levels are “capable of abstracting information and thinking about new things.” We look out at the world, and we look in upon ourselves with our mind’s eye. Not only do we develop goals and desires, but we recognize them; we develop language for them; we talk about them with our fellow creatures. We exercise free will, and we say so—because we are social organisms, and culture, too, evolves.
In the present moment, it’s natural to ask whether an artificial intelligence might develop any degree of free will or agency. Indeed, the question of agency in AI systems may be more critical than the question of consciousness. In an epilogue, Mitchell takes stock of the latest developments in neural networks and large language models, noting that their ability to generate text and respond to conversational prompts creates an impression of knowledge, if not sentience. At the same time, the models’ limitations are well known. They are designed for specialized tasks, as distinct from artificial general intelligence. They simulate human language with astonishing skill, based on statistical pattern-finding in enormous volumes of training data, but the text they generate is seldom tethered to real-world meaning, and they often fail in novel situations. Understanding of causal relations appears to be a weakness. Most starkly, however, they are passive.
Agency is what distinguishes us from machines. For biological creatures, reason and purpose come from acting in the world and experiencing the consequences. Artificial intelligences—disembodied, strangers to blood, sweat, and tears—have no occasion for that. If they have goals, the goals are imposed by their creators. They don’t plan. They don’t strive. At least so far.