Free Will

You are reading this, and it feels as though you could stop at any moment, which is either the most obvious fact about you or the most persistent illusion anyone has.

That feeling is not the question, though, and a great deal of confused argument comes from treating it as if it were. This lesson does the unglamorous work of saying what is actually being asked, because the free will problem is one of those disputes where most of the disagreement dissolves under a careful statement of the question and the rest gets much harder.

What is not in dispute

Nobody denies that you can usually do what you want. You want to turn the page, your arm moves, nothing prevents it. That capacity has a name, and the name is not free will: it is liberty of action, and it is the sort of thing prisons remove and locks restore. Every party to this dispute agrees that prisoners have less of it than free citizens, and no interesting philosophical position has ever denied it.

Nor is the dispute about political freedom, which is a relation between a person and a state, or about spontaneity, or about whether your behaviour is predictable by your friends. Your friends predict you fairly well, and this alarms nobody.

The question begins one step further in. Grant that you did what you wanted. Could you have wanted otherwise? And whatever it was that produced the wanting, was it you? Those two questions look like one question asked twice, and the entire subject depends on their being different.

The two demands

Philosophers separate them as follows.

The first is the ability to do otherwise, sometimes called the leeway condition. Take the world exactly as it was one second before you decided, every atom in place, every memory and mood and inclination as it actually was. Was more than one future open from there? If only one was, then in the sense that matters you could not have done otherwise, whatever it felt like from the inside.

The second is sourcehood. Is the action traceable back to you as its origin, or are you a place where a chain of causes happens to pass through on its way from somewhere else? A river runs through a valley and the valley shapes its course, but nobody thinks the valley is the source of the water. The worry is that a person is a valley.

John Locke pulled the two apart in 1690 with a case worth remembering, in the Essay Concerning Human Understanding. A man is carried while asleep into a room where a friend he has been longing to see is waiting. The door is locked behind him. He wakes, delighted, and stays of his own accord, never touching the door. He cannot leave: the leeway condition fails completely. Yet his staying is entirely his: it comes from his own desire, and he would have stayed had every door in the building been open. Locke concluded that his staying was voluntary although not free, and later writers have taken the same case to show that the two demands can come apart, so a theory must say which one responsibility actually needs.

Example. A driver has a stroke at the wheel, loses consciousness, and the car mounts the pavement and injures someone. Which of the two demands fails, and does it matter which?

Both fail, and that is exactly why the case is useless as a test. The driver could not have done otherwise, since the stroke fixed the outcome, and he is not the source of the swerve, since it came from a burst vessel and not from anything he wanted or decided. Cases where both conditions fail together are easy, and everyone gets them right. The interesting cases are the ones Locke constructed: one condition failing while the other holds. A test that never separates the two demands cannot tell you which one your judgement is tracking.

Now you. A man with a phobia of spiders sees one on the floor and runs from the room, screaming, in a way he later describes as not being up to him at all. He wanted to run, in some sense, since running is what he was moved to do. Which demand is failing here?

The three positions

Now the two demands can be turned into a map. Write the central claim of determinism as the thesis that the state of the world at any one time, together with the laws of nature, fixes its state at every other time. The next lesson does nothing but make that precise. For now, take it as a claim that could be true or false.

Incompatibilism is the thesis that if determinism is true, nobody has free will. It is a conditional, and note that it says nothing about whether anyone actually has free will, because it says nothing about whether determinism is true. Incompatibilists then split by what they add. Hard determinism adds that determinism is true, and concludes that we have no free will. Libertarianism, which has no connection whatever to the political doctrine of the same name, adds instead that we do have free will, and concludes that determinism is false. The two share a premise and differ over which further claim to keep.

Compatibilism denies the conditional. It holds that free will and determinism could both be true together, because what freedom requires was never the absence of prior causes but something else, usually something about acting on your own reasons without constraint. A compatibilist can then be perfectly relaxed about physics.

A fourth position has become prominent since the 1990s. Hard incompatibilism, associated with Derk Pereboom, agrees with the incompatibilist conditional and adds that indeterminism would not help either, since undetermined events are not thereby under anyone's control. On that view we lack free will whichever way physics turns out, which is a stronger and, its defenders argue, a more honest conclusion than hard determinism.

Two features of this map matter. First, the positions are not points on a spectrum from more freedom to less: they are answers to different questions, and the deepest division is between those who accept the incompatibilist conditional and those who reject it. Second, everything hangs on one conditional claim, which means the whole subject can be organised as a single argument with a few deniable premises. That is exactly what the final lesson will do.

What philosophers actually think

It is worth knowing where the expert opinion sits, if only because the popular presentation of this subject implies a consensus that does not exist.

The PhilPapers Survey, run by David Bourget and David Chalmers, asked professional philosophers about their views in 2009 and again in 2020, with 1785 respondents from the target faculty group in the later round. On free will, about 59 percent accepted or leaned towards compatibilism, about 19 percent towards libertarianism, and about 11 percent towards no free will, with the remainder undecided or holding some other view. The 2009 survey gave almost the same compatibilist share.

So the majority view among people who work on this professionally is that the problem, as usually posed, rests on a mistake. That is not an argument, and a reader should not treat it as one: majorities of experts have been wrong before, and this particular majority has been accused of defining the problem away for two hundred years. But it does mean that anyone presenting the debate as physicists and neuroscientists on one side against sentimental defenders of the soul on the other has not described the actual landscape.

Why it matters

The reason this is not a parlour game is that a large part of ordinary life is built on the assumption that people are answerable for what they do.

We praise and blame, and blame in particular seems to say that the person should have done otherwise. We feel resentment when wronged and guilt when we wrong others, and neither attitude makes sense towards a falling rock. We punish, and one whole family of justifications of punishment says that the offender deserves it, quite apart from any good the punishment does. That notion of desert is the load-bearing element: if it goes, punishment needs a different foundation, which is the subject of the second-to-last lesson.

The law has already worked out much of the structure, without settling the metaphysics. English law from the M'Naghten Rules of 1843 excuses a defendant who, through disease of the mind, did not know the nature and quality of his act or did not know it was wrong. A sleepwalker is not convicted. A person acting under duress has a defence in most systems for most crimes. Automatism, infancy and insanity are exemptions from the practice rather than excuses within it, and the difference is not a technicality: an excuse says this act was not culpable, while an exemption says this person is not the sort of thing we hold answerable at all.

Each of those doctrines is an admission that responsibility has conditions. The free will debate is the argument about what the conditions are, and whether anybody ever meets them.

Example. Classify these three claims. First: "The world is fully determined, so nobody is ever really to blame for anything." Second: "Freedom is doing what you want without coercion, and physics has nothing to do with it." Third: "We are genuinely responsible, so the future cannot be fixed by the past."

The first is hard determinism: it accepts the incompatibilist conditional and adds that determinism holds. The second is compatibilism: it rejects the conditional outright by giving freedom a content that determinism does not touch. The third is libertarianism: it accepts the conditional and runs it backwards, from the reality of responsibility to the falsity of determinism. Notice that the first and third speakers agree entirely about the philosophy and disagree only about the physics, while the second disagrees with both about what the word means.

Now you. A fourth speaker says: "Whether or not physics is deterministic makes no difference. If my choice was fixed, it was not up to me, and if it came out of quantum noise, that was not up to me either." Which position is this, and what is the argument's shape?

What could settle it

One last piece of orientation, about method, because readers arriving from the sciences often expect the wrong kind of resolution.

No experiment can show that determinism is true, since determinism is a claim about all events at all times and no finite body of evidence entails it. Physics can make it more or less plausible, and the next lesson but one reports what physics currently says, which is less decisive than either side tends to admit. Nor can any experiment show that compatibilism is true, since compatibilism is a claim about what the word "free" requires, and no measurement of a brain reports on the content of a concept.

What experiments can do is remove the props from under specific claims. If a decision can be predicted from brain activity long before the person feels themselves deciding, that is evidence against one popular picture of how deciding works, whatever it does or does not show about freedom. If the traits that drive behaviour turn out to be heavily heritable, the claim that you made yourself gets harder to hold. Two later lessons take exactly those two bodies of evidence seriously and try to say precisely what they establish.

And the argument does have a standard of success. It is the ordinary one from logic: an argument is valid when its premises cannot be true with its conclusion false, and it is sound when it is valid and its premises are true. Everything that follows is an attempt to construct a sound argument for one of the four positions above, or to show that the best argument for a rival is unsound by identifying the false premise. That is a checkable standard, and it is the reason this subject is not merely a clash of intuitions.

Example. Someone offers this argument: "Every event has a cause. Your decision was an event. So your decision was caused. So it was not free." Where would you attack it?

At the last step, and this is the single most common error in the whole subject. The first three lines may all be granted: causation is not the issue. The move from "caused" to "not free" is where the entire dispute lives, and here it is smuggled in as though obvious. A compatibilist grants every premise and denies the conclusion, holding that a decision caused by your own deliberation is the paradigm of a free one, not a counterexample to it. Notice also that the first premise is doing no work: an uncaused decision would be worse for the agent, not better, which is the hard incompatibilist's point.

Now you. Someone else argues: "Scientists have shown that brain activity precedes conscious choice. So the conscious self is not in charge. So free will is an illusion." Identify the gap between the second and third steps.

Everything so far has taken determinism as a placeholder: the thesis that the past and the laws fix the future. That formulation is loose enough to hide several different claims, at least one of which is obviously false and gets attributed to determinists constantly. Making it precise is the next lesson, and it turns out to matter enormously which version is on the table.

Determinism is the thesis that the state of the world at any one time, together with the laws of nature, fixes its state at every other time, and almost every popular objection to it is aimed at something else.

The previous lesson left the thesis as a placeholder. This one makes it precise, because four different claims travel under the name, at least two of them are false, and an argument that beats one of the false ones has beaten nothing. A reader who can state determinism exactly is already ahead of most published commentary on the subject.

Laplace's version

The canonical statement comes from Pierre-Simon Laplace, in the Essai philosophique sur les probabilités of 1814. Imagine, he wrote, an intelligence that at a given instant knew all the forces animating nature and the position of every item that composes it, and that was vast enough to analyse this data. For such an intelligence nothing would be uncertain, and the future, like the past, would be present to its eyes.

Three features of that sentence are worth extracting, because they are usually lost in the retelling.

The first is that the intelligence is a rhetorical device. Laplace's claim is about the world, not about what any knower can achieve. Whether such a being could exist is irrelevant to whether the world has the property he describes, and the fact that it plainly could not exist is not an objection to the thesis.

The second is that the entailment runs both ways. "The future, like the past" is not decoration: if a complete state plus the laws fixes every later state, the same machinery run backwards fixes every earlier one. Determinism is time-symmetric, so the past is exactly as fixed by the present as the future is. Anyone who finds the fixed future oppressive and the fixed past unremarkable should notice that the thesis treats them identically, and ask what work the asymmetry in their reaction is doing.

The third is that Laplace was writing an introduction to a book about probability. He did not think probability was a rival to determinism; he thought it was the science of our ignorance of a determined world, which is why the two subjects sat happily in one volume.

The modern statement

Stripped of the demon, determinism is a claim about entailment. Let $L$ be the complete set of laws of nature and let $S(t)$ be a complete description of the world's physical state at time $t$. Then determinism says that for any two times $t_0$ and $t_1$,

where $\models$ is entailment in the sense of the logic course: there is no possible world in which the laws hold and the state at $t_0$ is as described and the state at $t_1$ differs. An equivalent and often more useful formulation is due to Peter van Inwagen: if two possible worlds obey the same laws and agree completely at any one instant, they agree at every instant.

Notice what is packed into "complete". The state must fix everything relevant, not just the large objects: a description that omits one photon is not $S(t_0)$, and determinism claims nothing about it. Notice also that this is a claim about a total state of the whole world at an instant, not about isolated systems. Nearly every real subsystem is open to influence from outside, so a determinist expects local behaviour to be unpredictable in practice.

And notice what the formula does not contain: any mention of agents, of choices, of compulsion or of causation. It is a relation between descriptions and laws. Every argument from determinism to the absence of free will therefore needs further premises, and the whole of the incompatibilist case consists of supplying them.

It is not fatalism

Fatalism is the doctrine that some outcome will occur no matter what you do. It has a famous ancient form, the Idle Argument reported by Cicero and by Origen: if you are fated to recover from this illness, you will recover whether or not you call the doctor, and if you are fated not to recover, you will not recover whether or not you call, so either way calling is pointless.

Determinism says the opposite of this. On a deterministic picture the doctor matters enormously: whether you recover depends on whether antibiotics reach the infection, which depends on whether the doctor is called, which depends on your deliberating and reaching for the telephone. Determinism says that your deliberating was itself fixed by earlier conditions. It does not say that the deliberating is idle. It says it is one link in a chain, and removing a link changes the outcome.

The distinction is easy to state and surprisingly hard to hold on to, because determinism produces the same feeling as fatalism. But the feeling misdescribes the thesis. Under determinism your choices are causes of what happens; under fatalism they are irrelevant to it. The Chrysippan reply to the Idle Argument, from the third century BC, is exactly this point: recovering may be co-fated with calling the doctor, so the fated outcome comes about through your action rather than around it.

Example. During an epidemic someone reasons: "If I am going to catch it, I will catch it whatever I do; if I am not, I will not. So there is no point wearing a mask." Diagnose the error, then say what a determinist should say instead.

The error is fatalism, and specifically the assumption that the outcome is fixed independently of the intermediate steps. It is not: catching the disease depends on the viral dose reaching the airway, which depends on the mask. A determinist says that whether you wear a mask is itself fixed by prior conditions, including this argument if you find it convincing, and that the outcome is fixed partly through the mask. Both branches of the fatalist's dilemma are false, since neither antecedent is settled independently of what you do. Notice that the argument would be equally bad in an indeterministic world, which is a sign that determinism was never the problem with it.

Now you. A soldier says there is no point taking cover, because either the bullet has your name on it or it does not. What is the closest true claim in the neighbourhood of what he is saying?

It is not predictability

The second confusion is between a property of the world and a property of us. Chaos theory made the difference vivid.

In 1961 Edward Lorenz was rerunning a numerical weather model and typed a value from a printout, 0.506, instead of the stored 0.506127. The difference is about one part in four thousand. The new run tracked the old one for a while and then diverged completely, and the paper he wrote in 1963, "Deterministic Nonperiodic Flow", named the phenomenon: a deterministic system with sensitive dependence on initial conditions.

The point for our purposes is in the title. Lorenz's system of three equations is perfectly deterministic. It is also unpredictable beyond a horizon, because any error in the starting state grows exponentially. For the real atmosphere, errors at the scales that matter double in something like a day and a half, so an initial uncertainty of one part in ten thousand grows to order one in about twenty days, and both Lorenz's own estimates and modern reruns of the calculation put the intrinsic limit on weather forecasting at roughly two weeks. No better computer or denser network of instruments removes that limit, since it comes from the dynamics rather than from the instruments.

So determinism does not imply predictability. The implication also fails in the other direction: many indeterministic systems are extremely predictable in aggregate, which is why insurers can price a policy for a population without knowing anything about an individual.

This matters for the free will argument, because a great deal of loose writing treats "your behaviour cannot be predicted" as though it established freedom. It establishes nothing of the sort. A hurricane's path cannot be predicted either.

Example. A coin toss is the standard picture of a chance event. Is it one?

Not in any sense that involves indeterminism. A tossed coin is a rigid body obeying Newtonian mechanics, and its outcome is fixed by the initial spin, upward velocity and orientation of the hand. Persi Diaconis and colleagues built a machine that tosses a coin to the same outcome every time, and their 2007 analysis of the physics predicted that an ordinary human toss should land with the same face upward as it started slightly more often than half the time, at about 51 percent, because the coin precesses about an axis that keeps one face favoured. A team led by Frantisek Bartos tested it in 2023 with 350,757 recorded flips and found the same-side proportion to be 0.508, an excess of roughly 2,800 flips over what a fair process would give. The randomness of a coin is entirely our ignorance of the initial conditions, exactly as Laplace said, and calling an outcome random is a statement about the tosser rather than about the coin.

Now you. Someone concludes from the coin result that human decisions are likewise determined, since brains are physical systems. What is right and what is missing in that inference?

It is not the claim that everything has a cause

The third confusion is with universal causation, the thesis that every event has a cause. These come apart in both directions.

An indeterministic world can be full of causes. If a radioactive nucleus has a fifty percent chance of decaying in the next hour and does decay, its decay was undetermined, and yet there is nothing wrong with saying that the instability of the nucleus caused it. Probabilistic causation is a perfectly respectable notion, and most of medicine works in it: smoking causes lung cancer without determining it.

A deterministic world need not be full of causes in the ordinary sense either. Fundamental physics is written in laws of evolution, not in cause-and-effect pairs, and Bertrand Russell argued in 1913 that the notion of cause survives in science only as a relic. Whether or not that is right, the point is that determinism as stated says nothing about causes, and an argument that leans on "everything is caused" is leaning on a different thesis that will need its own defence.

It is not compulsion

The fourth confusion is the most rhetorically powerful and the least defensible. Determinism is regularly presented as though the laws of nature force us to act, so that a determined agent is a puppet on strings.

Laws of nature are not agents. They do not push, command or constrain: on the standard reading they are descriptions of regularities the world in fact exhibits, and on the strongest realist reading available they are relations between universals, which still do not exert pressure on anyone. A puppet is unfree because a puppeteer, an agent with its own aims, moves it against or without its own. There is nothing playing that role in a determined universe. The image imports an oppressor into a picture that contains only the world doing what worlds do.

This is not a knockdown reply, and it should not be taken for one. The incompatibilist argument in a later lesson concedes the whole point and then reconstructs the problem without any hint of compulsion, using nothing but the fixity of the past and the laws. That is why the argument is good. But the puppet is the version most people carry around, and it should be put down before going further.

Example. Sort these four claims: (a) the state of the world in 1900 plus the laws entails the state of the world today; (b) whatever will happen will happen, so effort is pointless; (c) a sufficiently powerful computer could predict your next sentence; (d) you are compelled by physics to do what you do.

Only (a) is determinism. (b) is fatalism, which determinism denies, since under determinism effort is one of the things that fixes outcomes. (c) is a claim about predictability, which does not follow from determinism because of sensitive dependence, and which in any case is impossible for a predictor inside the system it is predicting. (d) is the compulsion picture, which adds a coercer to a thesis that mentions none. Someone attacking free will properly needs (a) and nothing else, and someone defending it against (b), (c) or (d) has not yet engaged.

Now you. A physicist objects that quantum mechanics has refuted determinism, so the whole debate is over. What has gone wrong with the objection, before we even get to the physics?

Near-determinism is enough

That last point deserves stating on its own, because it decides how much the physics can matter.

Suppose the world is deterministic except for genuinely chancy events at the smallest scale, which is roughly what a standard reading of quantum mechanics gives. Ted Honderich called this near-determinism, and it changes nothing about the argument. Rewrite the thesis with chances: the state at $t_0$ plus the laws fixes the objective probability of every later state. Now ask whether it is up to you what those chances are. It is not, since they are fixed by conditions in place before you were born, and adding a random element to a process you do not control does not hand you control of it.

An analogy makes the shape clear. If a decision were settled by a coin flip inside your head, the outcome would be undetermined and would not be up to you in any sense worth having. The libertarian's task, in a later lesson, is to explain how the indeterminism could be located and structured so that it yields control rather than noise. It is a genuinely difficult task, and the best attempts at it are serious philosophy rather than hand-waving.

So the thesis is now clean. Determinism is an entailment claim about complete states and laws, distinct from fate, from predictability, from causation and from compulsion, and it has a weaker relative that supports the same arguments. The obvious next question is whether it is true, and the answer physics gives is more interesting and much less decisive than either side of this debate usually reports.

The previous lesson stated determinism precisely enough to ask whether it is true, and the honest answer from physics is that nobody knows, for reasons more interesting than a shrug.

This lesson is the one place in the course where the subject matter is empirical. It is worth doing carefully, because the two most common moves in popular discussion, that classical physics proved determinism and that quantum physics refuted it, are both wrong, and a reader who knows why is protected against a great deal of nonsense in both directions.

Classical mechanics, and the cracks in it

Newton's laws are second-order differential equations: give the positions and velocities of every particle at one instant, and the equations return the trajectory forwards and backwards in time. That is determinism in the sense of the previous lesson, and it is why Laplace could write his sentence in 1814 with confidence.

The confidence was slightly too high, and the exceptions are worth knowing because they show how delicate the property is. In 2003 John Norton described a dome, shaped so that a particle resting exactly at its apex is in equilibrium, and yet Newton's laws permit the particle to remain there for any length of time and then spontaneously slide off in any direction, with no force applied and no violation of the equations. The dome exploits the fact that the standard existence-and-uniqueness theorem for differential equations requires a smoothness condition that the surface fails at exactly one point.

Worse, in 1992 Zhihong Xia proved that a system of five point masses under Newtonian gravity can send a particle to spatial infinity in finite time, which by time-reversal means that a particle can arrive from infinity at an unpredictable moment. Nothing in the initial state announces it. Classical mechanics, taken literally, is therefore not quite deterministic, and it is saved only by conditions on the shape of surfaces and the finiteness of forces that are physically reasonable and not part of the laws themselves.

None of this matters for human beings and none of it delivers free will. It matters as a warning: determinism is not something a theory wears on its face, and reading it off a set of equations takes more care than it looks.

The two dynamics of quantum mechanics

Quantum mechanics is famously said to have destroyed determinism, and the truth is more specific. The theory has two rules for how states change, and they are not alike.

The first is the Schrödinger equation, which governs an isolated system's evolution. It is deterministic, linear, and time-reversible: the state now plus the equation gives the state at any other time, exactly as Newton's laws do. Nothing in this half of the theory threatens Laplace.

The second is the measurement rule. When a measurement occurs, the outcome is one of the eigenvalues of the measured quantity, with probabilities given by Born's rule of 1926, and the state jumps to match. This half is irreducibly probabilistic. A single silver atom in a Stern-Gerlach apparatus goes up or down, the theory gives only the probabilities, and nothing in the prior state distinguishes the atoms that will go up from those that will go down.

The awkwardness is that the theory does not say what a measurement is. Measuring devices are made of atoms and should obey the first rule, so the second rule looks like an extra postulate applied by hand when a system gets large or complicated. That is the measurement problem, and it is not a puzzle at the fringes: whether the world is deterministic turns entirely on how it is solved.

What Bell's theorem does and does not rule out

The obvious repair is to say that the probabilities reflect our ignorance, exactly as Laplace said of probability generally. If each silver atom carried a hidden property fixing which way it goes, quantum mechanics would be an incomplete statistical description of a determined world. Einstein, Podolsky and Rosen argued in 1935 that the theory is incomplete in something like this way.

In 1964 John Bell showed that the repair has a testable consequence. If the outcomes of measurements on two separated particles are fixed by properties carried locally by each particle, then correlations between the results must satisfy an inequality. Quantum mechanics predicts violations of it. This is the rare case where a metaphysical question about the completeness of a theory has an experimental answer.

The experiments have been done, and done to exhaustion. Alain Aspect's group produced the first convincing violations in 1982. Loophole-free versions arrived in 2015: a group at Delft, led by Ronald Hanson, ran the test with entangled electron spins in diamond separated by 1.3 km, far enough that no signal at light speed could connect the measurement choices to the distant outcomes, and reported a violation with a p-value of 0.039 on 245 trials, while photonic experiments at NIST and in Vienna reported violations with vastly smaller p-values the same year. In 2018 a group used light from quasars billions of years old to choose the measurement settings, pushing any conspiracy between the settings and the particles back into the deep past.

Now the crucial point, routinely misreported. Bell's theorem rules out local hidden variables. It does not rule out hidden variables, and it does not establish indeterminism. David Bohm's theory of 1952 is a fully deterministic hidden-variable theory in which particles have definite positions at all times and are guided by a wave that acts nonlocally; it reproduces every prediction of standard quantum mechanics. It is unpopular for other reasons, chiefly its awkward relationship with relativity, but it is not refuted by any experiment, and it is deterministic in Laplace's exact sense.

The interpretations disagree, the predictions do not

Line up the serious options and the situation becomes clear.

The Copenhagen approach and its descendants take the measurement rule at face value: real chance, and determinism is false. The spontaneous collapse theory of Ghirardi, Rimini and Weber, published in 1986, replaces the vague measurement rule with a precise stochastic one in which each particle's wavefunction localises at random at a fixed low rate, so that a macroscopic object collapses almost instantly; it is genuinely indeterministic and, unlike Copenhagen, makes predictions that differ slightly from standard quantum mechanics and are being tested. Bohmian mechanics is deterministic. The Everett or many-worlds view keeps only the Schrödinger equation, deletes the measurement rule entirely, and is therefore the most deterministic theory on the list: the universal state evolves unitarily, and the appearance of chance comes from an observer's ignorance about which branch they are in.

These make the same predictions for every experiment yet performed. So the answer to "is the world deterministic?" is currently a function of an interpretive choice that the evidence does not force, and anyone who tells you physics has settled it is reporting their favourite interpretation as a result.

Example. A commentator writes that Bell's theorem proves the universe is not deterministic, so Laplace was refuted in the laboratory. What is the mistake, and what would a Bohmian say?

The mistake is dropping the word "local". Bell's inequality is derived from local hidden variables, meaning that each particle's outcome depends on properties it carries and on the setting of the nearby apparatus, and it is that conjunction the experiments refute. Give up locality and determinism survives untouched: in Bohmian mechanics every particle has a definite position at every moment, the trajectories are fixed by the guiding equation, and the statistical predictions match quantum mechanics exactly because the initial positions are distributed according to the Born rule. The Bohmian's reply is therefore that the experiments refuted locality, which is what Bell himself said they showed, and that determinism was never on the scaffold.

Now you. Bell tests assume that the experimenters' choices of measurement setting are statistically independent of the hidden state of the particles. What position denies this, and why is it uncomfortable for everyone?

Does any of it reach a brain?

Suppose the world is indeterministic at the smallest scale. The libertarian needs that indeterminism to be present where decisions are made, which means inside a warm, wet organ at 310 kelvin. That is a physical question with a fairly hard answer.

The relevant comparison is between quantum length scales and the sizes of the structures that carry a neural signal. A particle in thermal equilibrium has a typical momentum $p=\sqrt{3mkT}$, and its de Broglie wavelength is $\lambda =h/p$, which measures how spread out its quantum state is. For a sodium ion, mass about 23 atomic mass units, at body temperature, that wavelength is about 0.03 nm. The narrowest part of a potassium channel, the selectivity filter mapped by Roderick MacKinnon's group in 1998, is about 0.3 nm across, and a neuron is roughly ten micrometres wide. An ion whose quantum spread is ten times smaller than the hole it passes through behaves, for every purpose that matters to a cell, like a small classical ball.

Max Tegmark made the argument sharply in 2000. He estimated decoherence times for the candidate quantum degrees of freedom in the brain, including the microtubules proposed by Penrose and Hameroff, and got figures in the region of ${10}^{-13}$ seconds and below, against neural dynamics operating on timescales of ${10}^{-3}$ seconds and slower, a gap of ten orders of magnitude or more. Hagan, Hameroff and Tuszynski replied in 2002 with corrected estimates around ${10}^{-4}$ seconds, which narrows the gap without closing it, and the dispute is not resolved. What is not in dispute is that no quantum effect in the brain has been measured, whereas quantum effects in biology have been measured elsewhere, in photosynthetic energy transfer and in enzymatic tunnelling.

Note the honest qualification. The same calculation for an electron gives a wavelength of about 6 nm, larger than the 5 nm thickness of a cell membrane, which is why electron tunnelling in proteins is real, routine and essential to respiration. The claim is not that brains are classical objects. It is that the specific structures carrying decisions, ions crossing channels and vesicles fusing with membranes, are far too heavy and far too warm for coherence to survive.

Example. Repeat the wavelength calculation for a potassium ion, mass 39 atomic mass units, at 310 K, and say what it shows.

Thermal energy first: $kT=1.381\times {10}^{-23}\times 310=4.28\times {10}^{-21}$ J. The mass is $39\times 1.661\times {10}^{-27}=6.48\times {10}^{-26}$ kg. Then $p=\sqrt{3mkT}=\sqrt{3\times 6.48\times {10}^{-26}\times 4.28\times {10}^{-21}}=2.88\times {10}^{-23}$ kg m/s, and $\lambda =h/p=6.626\times {10}^{-34}/2.88\times {10}^{-23}=2.3\times {10}^{-11}$ m, that is 0.023 nm. It is smaller than the sodium figure, as it must be, since a heavier particle at the same temperature has more momentum and a shorter wavelength. Against a 0.3 nm filter, the ion is localised to about a thirteenth of the aperture, so its passage is a matter of electrostatics and geometry rather than of interference.

Now you. Neurons are noisy: the opening of an ion channel is a random event, and a vesicle often fails to release when an action potential arrives. Does that noise give the libertarian what they need?

Example. Grant for the sake of argument that each ion channel's opening is genuinely undetermined. Estimate how much of that randomness survives at the level of a neuron's decision to fire.

Very little, and the estimate is a one-line calculation. Independent random events average out with relative fluctuations of order $1/\sqrt{N}$, where $N$ is the number of contributing events. A patch of membrane bringing a neuron to threshold involves on the order of ${10}^4$ channel openings, giving a relative fluctuation of $1/\sqrt{{10}^4}=0.01$, or one percent. A decision that recruits populations of ${10}^8$ neurons is averaging over vastly more, and the corresponding figure is ${10}^{-4}$. So microscopic indeterminacy does not vanish, but it is suppressed by the square root of a very large number every time the signal is passed up a level. A libertarian needs not merely noise but a mechanism that amplifies it, which is why Kane's account in a later lesson appeals to chaotic sensitivity: chaos is the only known way to run this calculation backwards.

Now you. Does that calculation refute libertarianism?

What the physics settles

Three conclusions, and they are narrower than either camp usually wants.

Determinism is not established. Classical physics is deterministic only under conditions it does not itself guarantee, and quantum physics divides on the question along interpretive lines that no experiment yet distinguishes.

Determinism is not refuted either. Bohmian mechanics and the Everett interpretation are deterministic, empirically adequate and taken seriously by working physicists, and Bell's theorem refutes locality rather than determinism.

And, most importantly for what follows, the debate does not turn on the answer. The incompatibilist argument of the next two lessons works just as well against near-determinism, where the past and the laws fix the chances, since chances fixed before your birth are no more up to you than outcomes fixed before your birth. Meanwhile the compatibilist never needed the question answered at all. It is a curious feature of this subject that the empirical question everyone reaches for first is close to irrelevant to it.

There is one further reason not to lean on physics, which the next lesson develops. Arguments that freedom is impossible do not need determinism in the first place. Some of the oldest of them use nothing but the assumption that statements about the future are already true, or that somebody already knows what you will do, and they reach the same conclusion with no laws of nature anywhere in sight. They fail, but seeing exactly how they fail sets the standard that a good argument against free will has to meet.

If it is true today that you will read this sentence tomorrow, then tomorrow you will read it, and it can seem to follow that you have no choice about it.

That argument uses no laws of nature, no initial conditions and no neuroscience. It is older than all of them, and versions of it were being answered in Athens in the fourth century BC. Two of the arguments in this lesson fail, and they fail on a mistake that the logic course diagnoses exactly. The third does not fail so easily, and understanding why is the best possible preparation for the serious argument against free will in the next lesson.

The sea battle

Aristotle raises the problem in chapter nine of De Interpretatione, around 350 BC, with an example about a naval engagement. Take the statement "there will be a sea battle tomorrow". By the principle of bivalence, every statement is either true or false. So this one is either true or false today, before the battle. If it is true today, then the battle will occur, and no admiral's decision can prevent it. If it is false today, then it will not occur, and no admiral can bring it about. Either way, tomorrow's outcome is settled today, and the deliberations of the admirals are theatre.

The argument generalises immediately to everything anyone will ever do, which is why it matters. Note that it appeals to nothing but the truth values of statements. A world with no laws of nature at all, in which events simply happened at random, would face the same argument, provided that statements about the future have truth values.

Aristotle's own response is disputed. The traditional reading, and the one that dominated for centuries, has him restricting bivalence: statements about future contingents are neither true nor false until the matter is settled. That is a serious cost, since it means abandoning a principle that is otherwise as safe as anything in logic, and it also fails to touch the argument's real weakness. A modern treatment can do better.

The scope error

Here is the argument written out, with $p$ standing for "there will be a sea battle tomorrow" and $\mathrm{Box}$ for necessity.

First premise: $p$ is either true or false, and suppose it is true. Second premise: necessarily, if $p$ is true then there will be a sea battle. Conclusion: so if $p$ is true, then necessarily there will be a sea battle, which is to say the battle cannot be prevented.

The second premise is a truth of logic and cannot be denied: it just says that if a statement is true then things are as it says, which is what truth means. The conclusion, however, does not follow from it, and the failure is a scope error of exactly the shape the logic course flags. The premise has the form

and the conclusion has the form

and the second does not follow from the first. Medieval logicians had a name for the distinction and used it for precisely this argument: necessitas consequentiae, the necessity of the inference, against necessitas consequentis, the necessity of the thing inferred. To see the gap with a countermodel, take $p$ as "the cup is on the table". Necessarily, if the cup is on the table then the cup is on the table. It does not follow that if the cup is on the table then it is on the table necessarily, since somebody may lift it.

So the sea battle argument is invalid, and bivalence survives. What was true today was true because of what the admirals would do; the truth did not reach back and make them do it. Truth tracks the world rather than constraining it, and that phrase is the second lesson of this lesson, worth as much as the first.

Example. Diagnose this: "If I know that you are sitting down, then you cannot stand up, since if you stood up I would not have known it."

The scope error again, dressed in knowledge instead of truth. What is true is that necessarily, if I know you are sitting, you are sitting: knowledge entails truth. What does not follow is that if I know you are sitting, you are necessarily sitting, in the sense of being unable to stand. The final clause of the objection actually gives the game away: if you stood up, then I would not have known, which is exactly right and shows that my knowledge depends on your behaviour rather than the other way round. My knowing tracks what you do. It does not hold you in the chair.

Now you. A friend says that since a photograph taken yesterday shows what you were wearing yesterday, and the photograph cannot now be altered, your having worn those clothes was unavoidable. Is the argument the same one, and where does it differ?

The cost of Aristotle's escape

Since the scope reply works, it is worth seeing what the traditional reply would have cost, because the comparison shows how much cheaper a logical diagnosis is than a semantic surgery.

Denying bivalence for future contingents means saying that "there will be a sea battle tomorrow" is neither true nor false today. That immediately threatens the law of excluded middle, since the disjunction of a statement with its negation should be a logical truth and here both disjuncts lack a value. It also plays badly with ordinary talk: if I say today that it will rain tomorrow and it rains, most people think what I said was true, and that it was true when I said it, not that it became true overnight.

Twentieth-century logicians built machinery to soften the damage. Richmond Thomason's supervaluational semantics of 1970 treats the future as a branching tree of possible continuations and counts a statement as true now if it is true on every branch, false if false on every branch, and neither otherwise. The disjunction "either there will be a sea battle or there will not" then comes out true on every branch and so true now, which rescues excluded middle while leaving each disjunct without a value.

That is elegant, and it is much more machinery than the problem needs. The scope diagnosis leaves classical logic untouched, keeps bivalence, and explains the illusion. When a small logical error and a large semantic revision both dissolve a paradox, the error is nearly always the right diagnosis.

Example. Under supervaluation, evaluate these three statements as spoken today: "there will be a sea battle tomorrow", "either there will be a sea battle tomorrow or there will not", and "if there is a sea battle tomorrow, ships will be lost".

The first is neither true nor false, since some branches contain a battle and others do not. The second is true, because on every branch one disjunct or the other holds, which is exactly the feature the semantics was designed to deliver. The third is true if on every branch where a battle occurs ships are lost, which is a substantive claim about naval warfare rather than about logic, and the point of including it is that conditionals about the future are not automatically valueless: only the contingent parts go silent, and the structure connecting them survives.

Now you. Does denying bivalence actually save free will?

The idle argument, briefly

The other ancient argument was mentioned in the previous lesson and can be dealt with quickly. If you are fated to recover, you will recover whether or not you call the doctor; if you are fated not to, you will not, whether or not you call; so calling is pointless.

The reply, attributed to Chrysippus in the third century BC, is that outcomes and actions are co-fated. If it is fated that you recover, it may equally be fated that you call the doctor, and the recovery comes about through the call rather than in spite of it. The fatalist's dilemma assumes that each branch is settled independently of what you do, and neither is.

The reason the idle argument is worth a paragraph is that its error is the one people actually make when they first meet determinism. They hear "the future is fixed" and conclude "so nothing I do matters", which reverses the actual implication: on a deterministic picture what you do matters completely, and is itself fixed.

Foreknowledge, which is harder

Now the argument that does not dissolve. It comes from theology, and it was stated in its modern form by Nelson Pike in 1965, but its interest is entirely independent of whether anyone believes in the foreknower.

Suppose there is a being who is essentially omniscient, so that it not only knows everything but could not be mistaken. Eighty years ago, that being believed that you would read this sentence today. Now assemble three claims. The past is fixed: nothing anyone does now can change what was the case eighty years ago. An essentially omniscient being cannot hold a false belief. And if you were to do otherwise now, then that past belief would have been false.

Put together, the argument concludes that you cannot do otherwise. And note what has happened: the scope reply no longer works. The argument does not move from "necessarily, if it was believed then it will occur" to "it will necessarily occur". It moves from a fixed fact about the past, together with an entailment, to a fixed fact about the present. That is a different and much better argument.

Look at its skeleton, because you will meet it again in the next lesson. It has a fixity premise, that the past is not up to us; a connection premise, that the past item entails what you do now; and a transfer principle, that what is entailed by something not up to us is itself not up to us. Change "God's belief eighty years ago" to "the state of the world eighty years ago and the laws of nature" and you have the Consequence Argument almost word for word. The theological version is the dress rehearsal, and it has been rehearsed for sixteen centuries.

Example. Which premise of the foreknowledge argument does the following reply attack: "God is outside time altogether, so there is no moment at which God's belief lies in the past"?

The fixity premise, by denying that the belief has a temporal location at all. This is the reply of Boethius in the sixth century and of Aquinas in the thirteenth: an eternal being sees all times at once, as a person on a hill sees a whole road that a traveller on it experiences in sequence, so God's knowledge is not earlier than your action and the fixity of the past does not apply to it. The standard objection is that the argument can be rebuilt without the temporal ordering: if a timeless fact entails what you do, and that fact is not up to you, the transfer principle still delivers the conclusion. Moving the knower out of time removes one premise and leaves the shape intact, which is a good sign that the shape rather than the theology is what matters.

Now you. The Ockhamist reply distinguishes hard facts about the past from soft ones. Which of these two is soft, and why does the distinction matter here: "Caesar crossed the Rubicon in 49 BC", or "Caesar crossed the Rubicon 2074 years before this lesson was read"?

What these arguments leave behind

Three results carry forward, and they are the tools for reading the next lesson.

The first is the scope distinction. Any argument that concludes from a necessary connection that one of its terms is necessary should be symbolised before it is believed, and most such arguments will not survive the symbolising. This is the most common single defect in bad arguments against free will, and spotting it takes ten seconds once the habit is there.

The second is the difference between tracking and constraining. A true statement, a photograph, a diary entry and an infallible belief all match the world without pushing it. That someone knows what you will do is a fact about them, not a chain on you. When we come to the empirical evidence, the same point returns in scientific dress: a brain signal that predicts your choice is tracking your decision-making, and only a further argument makes it a constraint on you.

The third is the shape that does survive. Fixity, connection, transfer. Nothing in that skeleton commits a scope error, and nothing in it confuses tracking with constraining. It is the best argument against free will that anyone has produced, and in its secular form, using nothing but the past and the laws of nature, it has dominated the field since 1983. That is the next lesson.

If determinism is true, then what you do now is a consequence of the laws of nature and the state of the world before you were born, and neither of those is up to you.

That is the whole argument in one sentence, and it is the reason incompatibilism is not a superstition. The previous lesson dismantled two arguments that reach the same conclusion by a scope error, and isolated the shape that survives: fixity, connection, transfer. Peter van Inwagen assembled exactly that shape from purely secular materials in a 1975 paper and at length in An Essay on Free Will in 1983, and named it the Consequence Argument. Almost everything written since has been a response to it.

The argument in words

Van Inwagen's own statement is worth having in front of you. If determinism is true, our acts are the consequences of the laws of nature and events in the remote past. It is not up to us what went on before we were born, and it is not up to us what the laws of nature are. Therefore the consequences of these things, including our present acts, are not up to us.

Read it slowly and notice what it does not do. It does not say the laws compel anybody. It does not say the future is fixed regardless of what we do. It does not confuse a necessary connection with a necessary conclusion. It grants that your deliberation is a real cause of your action, and simply observes that your deliberation is itself a consequence of things you had no say in.

The plausibility comes from the two fixity claims, each of which is close to undeniable on its face. Nobody can now make it the case that the universe was other than it was a billion years ago. Nobody can now make it the case that the laws of nature are other than they are. Whatever freedom is, it does not include those two abilities.

The argument in symbols

To see whether it is valid, write it out. Following van Inwagen, introduce an operator: $Np$ means that $p$ is true and no one has, or ever had, any choice about whether $p$.

Two rules govern it. The first, usually called alpha, says that if $p$ is a logical or metaphysical necessity then $Np$: nobody has a choice about the truths of arithmetic. The second, beta, is the transfer principle, and it is where all the trouble lives:

In words: if nobody has a choice about $p$, and nobody has a choice about the fact that $p$ leads to $q$, then nobody has a choice about $q$.

Now let $P_0$ be a complete description of the state of the world at some moment in the remote past, $L$ the conjunction of all the laws of nature, and $A$ the proposition that you raised your hand at nine o'clock this morning. Determinism, as stated two lessons ago, gives $\mathrm{Box}((P_0\wedge L)\to A)$: the past state together with the laws entails the act.

The derivation runs in four steps. From determinism and rule alpha, $N((P_0\wedge L)\to A)$, since a logical consequence of the deterministic thesis is not something anyone has a choice about. From the fixity of the past and of the laws, $N(P_0\wedge L)$. Apply beta to those two lines and $NA$ follows: nobody has, or ever had, any choice about whether you raised your hand. Since the argument used nothing specific about the act, the conclusion generalises to every act anyone has ever performed.

The argument is valid if beta is. Its premises are the determinist thesis and the two fixity claims. So a compatibilist has exactly three options: deny the fixity of the past, deny the fixity of the laws, or deny the transfer principle. Every serious compatibilist response since 1983 takes one of those three routes, which is a rare and welcome tidiness in philosophy.

Example. Someone objects that the argument proves too much, since by the same reasoning nobody has a choice about anything at all, including trivial matters like whether a kettle boils. Is that an objection?

No, it is a correct reading of the conclusion. Van Inwagen's argument is meant to show that if determinism is true, then nothing that happens is up to anyone, and the kettle is not a special case. An objection would have to show that the conclusion is false in some instance, not merely that it is sweeping. What the observation does usefully highlight is that the argument targets one specific thing, the ability to do otherwise, and says nothing directly about whether people are the sources of their actions or whether they can be responsible for them. A compatibilist who has given up the leeway condition, in the manner of a later lesson, can accept the entire argument and deny that it touches moral responsibility.

Now you. Rewrite the argument's target so that it concerns a near-deterministic world, in which the past and the laws fix only the chances of your raising your hand. Does the argument survive?

Attacking the laws: Lewis on breaking them

David Lewis published a four-page reply in 1981 titled "Are We Free to Break the Laws?", and it is the most discussed compatibilist response to the argument.

Lewis distinguishes two things the compatibilist might be committed to. The strong thesis: I am able to break a law of nature, that is, able to perform an act which would itself be, or cause, a law-breaking event. The weak thesis: I am able to do something such that, if I did it, a law would have been broken at some point beforehand.

He accepts the weak thesis and rejects the strong one. The point turns on how we evaluate the counterfactual "if I had raised my hand". On Lewis's own account of counterfactuals, we go to the nearest possible world in which I raise it. Since the actual laws plus the actual past entail that I do not, that world must differ either in its laws or in its past, and the nearest such world contains a tiny divergence miracle shortly before the moment of choice, after which the ordinary laws resume. In that world I raise my hand, and a law is broken. But I do not break it: the miracle happens before my act and is no part of it. My raising my hand is a perfectly law-abiding event in a world whose history diverged slightly from ours.

The compatibilist claim is therefore only that had I acted otherwise, the past or the laws would have been slightly different, and that is not an outrageous ability to attribute to anyone. What is outrageous is the strong thesis, and nobody needs it.

The reply has not gone unchallenged. Critics, including Helen Beebee, argue that Lewis's account smuggles in a picture of laws as mere summaries of what happens, so that a divergence costs nothing metaphysically, and that on a stronger view of laws as genuine governors of events the weak thesis is no more available than the strong. Which side you find convincing depends on your view of laws of nature, which is a question in metaphysics with no connection to agency, and that dependence is itself interesting: the free will problem keeps turning out to be entangled with other unfinished business.

Attacking the fixity of the laws

Lewis's reply concedes that the laws are fixed and argues about what an agent's abilities amount to given that. A bolder line denies the fixity itself, and it turns on what a law of nature is.

On a Humean view, associated with Lewis's own metaphysics, laws are not governors of events but summaries of them: the laws are whatever axioms give the best combination of simplicity and strength in describing the total history of the world. Notice what follows. If the laws are a summary of everything that ever happens, and your raising your hand is part of everything that ever happens, then what the laws are depends in part on what you do. Helen Beebee and Alfred Mele called the resulting position Humean compatibilism in 2002: the fixity premise is false, not because anyone can break a law, but because the laws were never independent of the total pattern that includes our actions.

The reply is unsettling in a specific way. It does not give you a power to do otherwise in any dramatic sense; it says that in the nearest world where you act otherwise, the best summary of history is slightly different, and no fact about that world is being violated by anybody. Whether that counts as denying the premise or as a very careful restatement of Lewis's position is itself argued about.

The cost is that the whole reply is hostage to a view about laws. If laws are genuine necessitation relations, as many philosophers of science hold, the move is unavailable, and the fixity of the laws is as secure as the fixity of the past.

Example. Which premise does each of these deny: (a) a Humean who says the laws supervene on the total history; (b) someone who says that had you acted otherwise, a small miracle would have occurred before you acted; (c) someone who says that having no choice about $p$ and no choice about $p\to q$ leaves it open whether you have a choice about $q$?

(a) denies the fixity of the laws, by making what the laws are partly depend on what happens, including what agents do. (b) denies the fixity of the past, in the specific and modest form Lewis defends: the divergence lies before the act, so nothing is broken by the agent, but the past of the nearest world in which she acts otherwise is not quite this one. (c) denies the transfer principle. These are the three doors out of the argument, and there is no fourth, which is what makes the Consequence Argument such a good organiser of the debate.

Now you. A determinist objects that Humean compatibilism gets things backwards, since the laws explain the events rather than being constituted by them. Is this an objection to the position or a disagreement about something else?

Attacking the transfer principle

The other line of attack goes for beta directly, and it succeeded.

In 1996 Thomas McKay and David Johnson showed that beta entails a principle of agglomeration: from $Np$ and $Nq$, infer $N(p\wedge q)$. And agglomeration has a counterexample. Suppose I did not toss a coin this morning, though I could have. Let $p$ be that the coin does not land heads and $q$ be that the coin does not land tails. Both are true. I had no choice about either taken singly, since nothing I could have done would have guaranteed heads, and nothing would have guaranteed tails. But their conjunction says the coin lands neither way, which is true only because it was never tossed, and I certainly had a choice about that. So $Np$ and $Nq$ hold while $N(p\wedge q)$ fails, agglomeration is invalid, and beta with it.

This is a genuine result and it killed the original formulation. It did not kill the argument. Alicia Finch and Ted Warfield pointed out in 1998 that the derivation never needs the full strength of beta, because determinism supplies a strict entailment rather than a mere conditional. So replace the rule with

which says that whatever is entailed by something nobody has a choice about is itself something nobody has a choice about. That version is not touched by the coin, since it never licenses agglomeration, and the four steps go through as before.

Whether the repaired rule is true is now the question, and the honest answer is that it is very plausible and not provable. It has the same status as the fixity premises: something almost everyone accepts on reflection, which a determined compatibilist can reject at a cost.

Example. Where does each of these replies attack the argument? (a) "Had I acted otherwise, the past would have been very slightly different." (b) "It is only in an irrelevant sense that I could not have done otherwise, since I certainly could have if I had chosen to." (c) "The coin case shows that the rule licensing the argument is invalid."

Reply (a) attacks the fixity of the past, in Lewis's manner, by allowing that my acting otherwise is compatible with the past having been different, and it must then explain why this does not amount to an absurd power over history. Reply (b) attacks neither premise directly but the interpretation of "up to us", proposing that the ability the argument denies is not the ability freedom requires; that is the conditional analysis of the next lesson, and it has its own problems. Reply (c) attacks the transfer principle, and it is correct against the original version and answered by the repaired one. Only (c) is a technical objection; (a) and (b) are substantive philosophical positions.

Now you. A compatibilist says the argument equivocates, because "up to us" in the premises means "within our power to alter", whereas in the conclusion it must mean "within our power to determine". Is this a good objection?

Where this leaves the dispute

The Consequence Argument does not prove that we lack free will, and van Inwagen never claimed it did. He is a libertarian: he accepts the argument and rejects determinism, while candidly writing that free will remains a mystery, since he cannot explain how an undetermined choice would be his.

What the argument does is fix the burden. Before 1983 a compatibilist could treat incompatibilism as a confusion, a hangover from the puppet picture that a little clear thinking would remove. That is no longer available. There is now a valid argument, from premises most people accept on reflection, to the conclusion that determinism removes the ability to do otherwise. Anyone who wants to keep both determinism and free will must say precisely which premise is false, and pay for it.

The next lesson is the answer that has convinced the majority of philosophers: that the ability the argument targets is the wrong ability, because freedom was never about how many futures were open. It has a distinguished history, a devastating counterexample in its simplest form, and two modern successors that survive.

The Consequence Argument shows that determinism removes one thing, the ability to have done otherwise given the actual past, and the compatibilist reply is that this was never what freedom meant.

That reply is old, unpopular with the public, and held by roughly three philosophers in five. This lesson takes it seriously, which means watching its simplest version fail to a clean counterexample and then following the two repairs that survive. A reader who leaves thinking compatibilism is the claim that "free just means uncoerced" has met the 1748 version and none of the work done since.

The classical statement

Thomas Hobbes put the position in Leviathan in 1651 and in his exchange with Bishop Bramhall soon after. Liberty, he says, is the absence of external impediments to motion. Water is free to flow downhill when the channel is open and unfree when a dam blocks it, and it is no less free for flowing necessarily. A man is free when nothing stops him doing what he wills, and the question of what made him will it is a different question with a different answer.

David Hume made the argument sharper in the eighth section of the Enquiry Concerning Human Understanding of 1748, which he called a reconciling project. Liberty, on his account, is a power of acting or not acting according to the determinations of the will: if we choose to stay, we may stay; if we choose to move, we may move. This is consistent with every action being necessitated by prior causes, since necessity in Hume's sense is nothing more than the regular succession of like causes and like effects.

Hume then adds the move that makes compatibilism more than a definition, and it is the strongest card the position holds. Not only is responsibility compatible with necessity, it requires it. We blame a person for an action because it flows from their character, and it is evidence about them. If actions were disconnected from character, if they issued from nothing in the agent at all, then a person could not be praised or blamed for them any more than for a lightning strike. On this view the libertarian has the relationship backwards: causal connection to the agent's own dispositions is not the enemy of responsibility but its precondition.

That argument does real work, and every later lesson has to answer it, including the ones sympathetic to the other side.

The conditional analysis, and why it fails

The classical position needs an account of "could have done otherwise", since ordinary people plainly use the phrase and mean something by it. The classical answer is a conditional. To say that Jones could have stayed at home is to say that he would have stayed at home if he had chosen to. G. E. Moore gave the analysis its standard form in Ethics in 1912.

The attraction is that it sorts the ordinary cases correctly, and it sorts them without any reference to physics. A prisoner could not have left, because he would not have left even had he chosen to: the door was locked. A paralysed man could not have raised his arm, on the same test. A man who stayed at home reading could have gone out, because had he chosen to go out he would have gone. Determinism does not disturb any of these verdicts, because each conditional is about what would follow a different choice, not about whether that choice was open.

The analysis is nonetheless false, and the counterexample is due to Roderick Chisholm in 1964 and to Keith Lehrer. Consider a man with a severe phobia of snakes, so severe that he is psychologically incapable of picking one up. Offer him a large sum to do it. It is perfectly true that if he chose to pick up the snake, he would pick it up: his arms work, nothing external prevents him. So the conditional is satisfied, and the analysis says he can. But he cannot, and the reason is that the very choosing is beyond him. The analysis tests the wrong link in the chain.

The obvious repair, adding that he could have chosen otherwise, either reintroduces the categorical ability the analysis was meant to replace or launches a regress of conditionals. Classical compatibilism, in its simple form, is dead, and this is worth stating plainly because it is often assumed that compatibilists have never faced a serious objection. They faced this one and rebuilt.

Example. Apply the conditional analysis to two cases and say where it goes right and where it goes wrong. First, a woman does not donate to a charity because she has no money in her account. Second, a heroin addict does not refuse a dose.

For the first, the analysis works. Had she chosen to donate, the transfer would still have failed, so the conditional is false and the analysis correctly says she could not have donated. It also gets the moral verdict right: nobody blames her. For the second, the analysis gives the wrong answer. Had the addict chosen to refuse, he would have refused, since nobody was holding him down, so the conditional is true and the analysis says he could have refused. Yet the whole point of addiction is that the choosing is what has been compromised. This is the phobia case in more familiar clothes, and it shows that the defect is not an exotic one: the cases we most want a theory of freedom to handle are exactly the cases where the impairment lies inside the will.

Now you. A compatibilist proposes to repair the analysis: Jones could have done otherwise if and only if he would have done otherwise had he chosen to, and he could have chosen otherwise. What is wrong with the repair?

Frankfurt's hierarchy

The first successful repair came from Harry Frankfurt in 1971, and it relocates the question from what an agent can do to what an agent wants.

Human beings, unlike other animals, have desires about their desires. A smoker wants a cigarette and also wants not to want one. Frankfurt calls the first level first-order desires and the second level second-order desires, and when a second-order desire concerns which first-order desire should actually move you to act, he calls it a second-order volition.

Now compare three agents, all of whom take the drug. The unwilling addict wants the drug, wants not to want it, and is defeated by his craving: the desire that moves him is one he repudiates. The willing addict wants the drug and wants that desire to be the one that moves him: if the craving were to fade he would seek it out again. The wanton simply has cravings and no view whatever about which of them should move him; he is not fighting his desires or endorsing them, because he does not have the reflective structure to do either.

All three are determined, all three take the drug, and our reactions to them differ sharply. Frankfurt's proposal is that this is what freedom of the will consists in: your will is free when the desire that moves you is the one you want to be moved by. The unwilling addict acts on a desire that is in him but not of him, and that, rather than any fact about alternative futures, is why we treat him as unfree.

The account explains a great deal, including why we regard someone in the grip of a compulsion as unfree while regarding a person acting on carefully considered reasons as free, even though both are equally caused. It has two well-known problems. The first is regress: what makes a second-order volition authoritative, given that a third-order desire might repudiate it? Frankfurt answers with the notion of decisive identification, which critics find stipulative. The second is that the hierarchy can be installed from outside, and a manipulated agent might endorse exactly the desires their manipulator wanted them to endorse. That objection is powerful enough to have its own lesson.

The deep self, and whether it must be sane

Between the hierarchy and reasons-responsiveness lies a family of views usually called deep-self accounts, and they are worth a section because they contain the objection that leads to the rest of this course.

Gary Watson argued in 1975 that Frankfurt's levels are the wrong currency. What matters is not a desire about a desire, which is just another desire, but the distinction between what you happen to want and what you actually value: between a motivational system and an evaluative one. An action is free when it flows from your values, and the addict's tragedy is that his motivations have come apart from his judgement of what is worth pursuing.

Susan Wolf then pressed the case that no such account can be enough on its own, in a 1987 paper and in Freedom Within Reason in 1990, using an example that has stayed in the literature. JoJo is the son of a brutal dictator, raised in his father's company, given every advantage, and encouraged to admire everything his father does. He grows into an enthusiastic tyrant. He tortures people, and he wholeheartedly endorses torturing people: his values, his second-order volitions and his actions are in perfect agreement, and his mechanism responds to reasons of the sort he recognises.

Every structural account says JoJo is free and responsible. Most readers say he is not, or at least much less so than an ordinary tyrant who chose his cruelty from an ordinary childhood. Wolf's diagnosis is that the accounts have left out a condition she calls sanity: the ability to know what one is doing and to recognise that it is wrong, understood as a normative competence rather than a clinical category. JoJo's evaluative system is itself deformed, so tracing the act back to it settles nothing.

Adding the condition has a price that Wolf accepts openly. It makes responsibility depend on getting the moral facts roughly right, and so produces an asymmetry: a person who does the right thing for the right reasons is responsible even if they could not have done otherwise, while a person who does wrong because they cannot see it as wrong is not. Many find that asymmetry attractive and many find it unprincipled, and either way it is the first point in this course where a theory of freedom has to lean on a theory of value.

Example. Apply the hierarchy, the deep self and the sanity condition to JoJo, and say what each verdict depends on.

The hierarchy says JoJo is free: he wants to torture, wants that desire to move him, and is not alienated from it in the way the unwilling addict is. The deep-self account agrees, since his action flows from what he genuinely values rather than from a craving he repudiates. The sanity condition disagrees, because his capacity to recognise the wrongness of what he does was destroyed in childhood by his upbringing, and that capacity is a precondition of the whole practice rather than one more preference. So the first two verdicts depend only on the internal structure of his psychology at the moment of acting, and the third depends on a fact about how that psychology was formed. Watch that division carefully: it is the same division the manipulation argument exploits, and JoJo is a manipulation case with the manipulator replaced by an ordinary upbringing.

Now you. Does the sanity condition let a compatibilist answer every case of a deformed upbringing?

Reasons-responsiveness

The second repair is the dominant one in the current literature, developed by John Martin Fischer and Mark Ravizza in Responsibility and Control in 1998, and it starts from a concession.

They concede the Consequence Argument entirely. If determinism is true, nobody can do otherwise; they call the ability to do otherwise regulative control and give it up without a fight. What they claim instead is that moral responsibility never required it. What responsibility requires is guidance control: that the action issue from the agent's own, moderately reasons-responsive mechanism. Because this position defends responsibility while conceding the leeway argument, it is called semi-compatibilism.

The content is in the two conditions. A mechanism is moderately reasons-responsive when, holding the mechanism fixed, the agent would recognise a range of reasons for acting otherwise across a range of possible situations, and would act on at least one sufficient reason in at least one of them. The test is a counterfactual one, but note carefully what it does not require: it does not require that the agent could actually do otherwise in this world, since the mechanism is being tested across worlds where the reasons differ.

That distinction does the work. Take an ordinary shoplifter. Would he have refrained if a police officer had been standing beside him, or if the item had cost him his job, or if his mother had been watching? Plainly yes, in many such cases. His larcenous mechanism tracks reasons. Now take a genuine kleptomaniac, who steals objects he does not want, cannot use and often discards, and who steals with the store detective visibly watching. His mechanism is not tracking reasons at all, and the difference is measurable in his behaviour rather than postulated in his metaphysics.

The second condition, ownership, requires that the agent has taken responsibility for the mechanism: that they see themselves as an agent whose actions have effects, view themselves as a fair target of the reactive attitudes, and hold these views on the basis of their own experience. It exists chiefly to handle manipulated and brainwashed agents, whose mechanisms may be perfectly reasons-responsive while nobody thinks them responsible, and it is the part of the theory that critics press hardest.

Example. Two people fail to keep a promise. One forgot because he was drunk; the other has advanced dementia and has no recollection of ever making the promise. Apply guidance control.

The drunk's mechanism is reasons-responsive in the relevant sense, though not at the moment of failure: he is responsible via the earlier decision to drink, which was itself taken by a mechanism that responded perfectly well to the reason that he had an obligation the next day. This is the standard tracing analysis, and it is how the theory handles all cases of culpable incapacity. The person with dementia is different: no mechanism available to him now would register the reason across the relevant range of situations, because the capacity to hold and retrieve the commitment has gone. He is exempted rather than excused, and the theory says so without any appeal to what physics permits.

Now you. A hypnotised subject is given a post-hypnotic suggestion to open a window at three o'clock, and does so, inventing a reason afterwards about the room being stuffy. Which condition of guidance control fails, and what does that show about the theory's structure?

The charge of redefinition

The standing objection to all of this is that it changes the subject. Kant called compatibilist freedom a wretched subterfuge and compared the agent to a turnspit which, once wound up, carries out its motions by itself. William James called the position a quagmire of evasion. The complaint is that being free to act on desires you did not choose, produced by a history you did not select, is the freedom of a clock, and dressing it up in second-order volitions does not change what it is.

Compatibilists have two answers, and it is worth being clear that they are different answers.

The first is deflationary: the concept of freedom that our practices actually use is the compatibilist one, and this can be checked. We excuse the coerced, the compelled, the ignorant and the insane, and we do not excuse the ordinary well-informed adult on the ground that his upbringing shaped his character. If the incompatibilist were describing our concept, ordinary practice would look entirely different from how it looks.

The second is dismissive: the freedom the objector wants is impossible in principle, not merely absent in fact, since it would require an agent to be the uncaused cause of their own character. If a demand cannot be met in any possible world, failing to meet it is not a deprivation. That argument is developed properly in a later lesson, and it belongs as much to the hard incompatibilist as to the compatibilist, which is one of the odder alliances in the subject.

Both answers concede something to Kant. Compatibilist freedom is not the power to have been a different person; it is a property of how an action is produced in the person you are. Whether that is enough for desert is the question the last third of this course is about.

There is one more move available, and it is the most radical thing in the compatibilist arsenal. Everything so far has assumed that responsibility requires alternative possibilities in at least some sense, and has argued about how to interpret them. In 1969 Frankfurt published a two-page counterexample designed to show that responsibility never required alternatives at all, and it reset the entire field. That is the next lesson.

Everything so far has assumed that being responsible for an action requires having been able to do otherwise, and in 1969 Harry Frankfurt published a counterexample to that assumption which changed the shape of the field.

The previous lesson set out compatibilism as a theory of what makes an action free. This lesson takes the more radical route: not arguing about how to interpret "could have done otherwise", but arguing that the phrase was never doing the work anyone thought it was doing.

The principle under attack

Frankfurt named the target the principle of alternate possibilities, and stated it as follows: a person is morally responsible for what he has done only if he could have done otherwise.

The principle is enormously plausible. It underwrites the excuses we all accept. The cashier who hands over money at gunpoint, the driver whose brakes fail, the man pushed into someone else: in each case what removes the blame seems to be that no other course was open. It is also the premise that makes the Consequence Argument matter. If determinism removes alternatives, and alternatives are necessary for responsibility, then determinism removes responsibility, and that two-step is why anyone outside a philosophy department cares about the first step.

Frankfurt's strategy is to construct a case where the alternatives are removed by something that plays no part in what actually happens.

Black and Jones

Jones is deciding whether to shoot a man. Unknown to him, a neurosurgeon called Black wants him to shoot, and has installed a device in Jones's brain. Black is an excellent judge of these things and can tell from some prior indicator, a twitch, a pattern of blood flow, whatever you like, which way Jones is about to decide. If the indicator shows that Jones is about to decide not to shoot, the device fires and produces the decision to shoot instead. If the indicator shows that Jones is about to decide to shoot, the device stays idle.

As it happens, Jones decides to shoot entirely on his own, for his own reasons, and Black never intervenes. He watches, ready, and does nothing.

Now the two verdicts. Jones could not have done otherwise: every path in which he begins to decide against shooting ends with the device producing the decision to shoot. And Jones is responsible: the actual sequence of events ran through his own deliberation, his own reasons and his own decision, exactly as it would have in a world containing no Black at all. Black's presence made no difference to anything that happened, and it is hard to see how a fact that made no difference to what happened can affect the responsibility for it.

If both verdicts stand, the principle of alternate possibilities is false. And if it is false, the Consequence Argument loses its bite: determinism may well remove the ability to do otherwise, and that removal turns out not to be what responsibility was ever tracking.

Frankfurt added a diagnostic point that is easy to miss and does most of the persuasive work. In the ordinary excusing cases, the inability to do otherwise is also the explanation of the act: the cashier hands over the money because of the gun. The excuse works through the explanation. In his case the inability is idle: Jones did not shoot because of Black, and would have shot in exactly the same way had Black never existed. So the principle, stated in its usual form, has been overgeneralising from cases where something else was doing the excusing.

Example. Distinguish these two cases and say why the first excuses and the second does not. First: a bank teller hands over the money because an armed robber threatens her. Second: a bank teller who has decided to embezzle hands over the money to an accomplice, unaware that an armed accomplice in the queue would have forced her to do it had she hesitated.

In the first, the threat explains the act. Remove the gun and she does not hand over the money, so the inability to do otherwise is part of the story of why she acted, and the excuse runs through that. In the second, the armed accomplice is causally idle: she hands over the money because she decided to steal, and the story of the act contains no mention of him. Take him out of the world and every step of what actually happened is unchanged. She had no alternative and she is fully to blame, which is exactly the pattern Frankfurt claims refutes the principle.

Now you. Someone objects that Jones is not responsible, because Black's device made the outcome inevitable and nobody is responsible for the inevitable. What has this objection missed?

The dilemma defence

The counterexample did not go unanswered, and the best reply, developed by Robert Kane and stated sharply by David Widerker and Carl Ginet in the mid-1990s, attacks the prior indicator.

Black needs to know, before Jones decides, which way Jones is going to decide. So ask what the connection between the indicator and the decision is.

Take the first horn. Suppose the connection is deterministic, so that the twitch guarantees the decision. Then the case describes a world where Jones's decisions are determined by prior states, and an incompatibilist will say that Jones is not responsible in that world for exactly the reasons the Consequence Argument gives. The case would then be assuming the falsity of incompatibilism in order to refute a principle incompatibilists hold, which is question-begging.

Take the second horn. Suppose the connection is merely probabilistic, which is what an incompatibilist thinks the world is like at the moment of a free decision. Then the indicator does not guarantee anything. Jones might show the twitch associated with shooting and then decide not to shoot, so Black cannot rely on it, and to be sure of his outcome he would have to intervene before Jones decides, which turns the case into ordinary manipulation where nobody thinks Jones is responsible.

Either way, the case fails to do what it was built for. This is a good objection and it has been the centre of the literature for thirty years.

The main compatibilist counter is the flicker of freedom strategy, and the counter to the counter. Fischer's version concedes that Jones retains some alternative, perhaps the involuntary showing or not showing of the indicator, and argues that such alternatives are too thin to ground responsibility: they are not things Jones does, he has no control over them, and they are not the sort of thing that could make the difference between blame and exculpation. The demand is for a robust alternative, one the agent could have voluntarily taken and which would have been an exercise of control.

Example. Design a Frankfurt case and test it against the dilemma. A student is deciding whether to cheat in an exam. A device implanted by an examiner will force the decision to cheat if she shows signs of deciding otherwise. She cheats on her own. Where does the dilemma bite?

At the phrase "shows signs of deciding otherwise". If the signs are deterministically linked to what she will decide, then her decision was determined by prior states, and an incompatibilist denies responsibility for the same reason they deny it in any determined world, so the case cannot be used against them. If the signs are only probabilistically linked, the examiner cannot wait for them: she might display every sign of honesty and then cheat anyway, or the reverse, so no setting of the device both guarantees the outcome and stays out of the actual sequence. To make the case work, the device must be triggered by something that necessitates the decision without being part of it, and constructing such a trigger is the technical problem the next section is about.

Now you. Suppose the dilemma defence is right and no Frankfurt case can be built. Does the incompatibilist then win?

Buffers and blockage

The technical problem the dilemma poses is precise: build a case in which the agent has no robust alternative, without using a deterministic link between a prior sign and the decision. Three strategies have been tried.

The first is blockage, proposed by David Hunt and developed by Eleonore Stump. Instead of an intervener who watches and reacts, imagine that all the neural pathways leading to any decision other than the actual one are simply blocked, from before the deliberation begins. Nothing is triggered by a prior sign, so no deterministic link is needed, and the agent's actual deliberation proceeds through unblocked channels exactly as it would have anyway. Critics reply that a mind with most of its pathways walled off is not obviously undergoing normal deliberation at all, and that the case may be a disguised form of determination rather than an alternative to it.

The second is simultaneous overdetermination, from Alfred Mele and David Robb in 1998. A deterministic process is set running in Jones's brain which will produce the decision to steal at a set time, while Jones's own indeterministic deliberation runs in parallel. If Jones decides to steal on his own first, his process wins and the implanted one is preempted; if he does not, the implanted process produces the decision. There is no prior sign anywhere, so the dilemma has nothing to grip. The dispute here is about what happens when two processes converge on the same result, which is an old and unresolved problem about causal preemption.

The third and most discussed is Derk Pereboom's buffer case. In his tax evasion example, a device is set so that Plum can decide to evade only if he first attains a specific level of attentiveness to his moral reasons, and the device prevents him from ever reaching that level, while otherwise leaving him entirely alone. Plum decides to evade on his own, without ever coming close to the buffer. The only alternative left to him is failing to reach a state he never reaches anyway, which is not something he does and not a course of action he could have taken.

It is the fairest summary of the area to say that none of the three has been generally accepted and none has been generally refuted. What all three share is the shape of the reply: keep the agent's actual deliberation untouched, and remove the alternatives by a mechanism that is not a reaction to a sign.

Example. Why does the buffer case avoid the dilemma that defeats the original Black and Jones story?

Because it removes the prior sign. Black needed to know in advance which way Jones would go, and the dilemma attacked the link that gave him that knowledge. Pereboom's device needs no forecast: it simply makes one route permanently unavailable, in advance and unconditionally, and then does nothing at all. Plum's deliberation may be as indeterministic as any libertarian wants, so no assumption of determinism is smuggled in, and yet the only thing he could have done other than decide to evade is fail to attain a level of attentiveness, which is not an action, not under his control, and not something a jury could regard as the difference between guilt and innocence. Whether an alternative that thin is genuinely irrelevant is the remaining live question.

Now you. Suppose a critic accepts the buffer case, agrees that alternatives are irrelevant, and remains an incompatibilist. What must their argument now look like?

What responsibility tracks instead

If alternatives are not what responsibility depends on, something else must be, and the answer the literature converged on is the actual sequence.

The idea is that responsibility supervenes on how the action was actually produced. Two agents whose deliberations run identically, step for step, must receive the same verdict, whatever differs in the surrounding possibilities. Black's presence in the room, the buffer that was never approached, the counterfactual robber in the queue: none of these belongs to the actual sequence, and none of them changes anything about how the decision was reached.

This is a genuinely powerful principle, and it explains why Frankfurt cases feel compelling rather than merely clever. It also explains what a theory of responsibility should be looking for: features of the actual causal path, such as whether the mechanism producing the decision responds to reasons, whether the agent's values were engaged, whether they knew what they were doing.

The principle cuts both ways, though, and the incompatibilist can use it too. An actual sequence has a history, and the history is part of how the action was actually produced. Whether "actual sequence" is read narrowly, covering the proximate mechanism, or widely, covering the whole causal ancestry back to the agent's formation, is exactly what separates a compatibilist from an incompatibilist about sourcehood. Frankfurt's counterexample settled that alternatives are irrelevant. It did not settle how far back the relevant sequence extends.

What the field looks like afterwards

Three things changed after 1969, and they explain the shape of the rest of this course.

First, responsibility and the ability to do otherwise came apart as topics. It is now standard to distinguish leeway views, which make responsibility depend on alternatives, from source views, which make it depend on where the action came from. Fischer's semi-compatibilism from the previous lesson is the clearest example: determinism removes leeway, responsibility survives, and no contradiction is involved because responsibility never needed leeway.

Second, the burden on incompatibilists shifted. An incompatibilist who wants to press the case has to show that determinism spoils sourcehood, not merely that it removes alternatives. That is a harder and more interesting claim, and the manipulation argument of the next lesson is the best attempt at it.

Third, and less often noticed, the ordinary intuition that Frankfurt cases exploit is not neutral between the two sides. What makes us judge Jones responsible is that the actual sequence ran through his own reasons and his own deliberation. That is a compatibilist-friendly criterion. An incompatibilist has to explain why an actual sequence that satisfies it is not enough, and the answer will be about the history of the agent's reasons rather than about the moment of decision. Once the argument is about histories, it is about upbringing, genes and manipulation rather than about physics, and the second half of this course lives there.

A designed agent who meets every condition compatibilism asks for still looks like a puppet, and the manipulation argument turns that reaction into the strongest case against compatibilism anyone has made.

The previous lesson moved the debate from alternatives to sourcehood. This lesson is what incompatibilists did with the new ground. The method is unusual and worth noticing in its own right: instead of a formal derivation like the Consequence Argument, it works by constructing a sequence of cases and daring the opponent to find a principled place to stop.

Professor Plum, four times

Derk Pereboom set out the argument in Living Without Free Will in 2001 and refined it in 2014. In each of four cases, Professor Plum kills Ms White for self-interested reasons. In each case he satisfies every condition compatibilists have proposed: his action flows from a moderately reasons-responsive mechanism, he acts on desires he reflectively endorses, he is not coerced, not compelled, not deceived, and his egoistic reasoning is the ordinary sort that many people engage in.

In the first case, a team of neuroscientists manipulates Plum's brain directly, by radio, in the moment: they produce his reasoning process as it happens, and the desire that issues from it is irresistible only in the sense that his own reasoning is, which is to say not irresistible at all in the compatibilist's sense.

In the second, the neuroscientists did their work at the beginning of Plum's life, programming him so that he reasons in a rationally egoistic way, a trait he had no control over acquiring. They then leave him alone, and thirty years later that programming issues, through his own deliberation, in the killing.

In the third, no neuroscientists exist. Plum's character was fixed by rigorous training from infancy, administered by his household and community, which produced the same rationally egoistic outlook by ordinary social means.

In the fourth, nothing unusual happened at all. Plum is an ordinary man in a deterministic world, and the state of the world before his birth, together with the laws, entails his killing Ms White.

The claim is that there is no principled difference between adjacent cases, and that most people judge Plum not responsible in the first. If the judgement is right and the cases really do shade into each other, the judgement carries through to the fourth, which is the situation every determined agent is in.

Pereboom's preferred formulation is not merely "you cannot draw a line". It is an inference to the best explanation: the best explanation of why Plum is not responsible in the first case is that his action traces to factors beyond his control, and that explanation applies word for word to the fourth.

The zygote

Alfred Mele's version, from Free Will and Luck in 2006, compresses the argument into a single case and is harder to wriggle out of.

A goddess, Diana, creates a zygote in Mary. She has a complete knowledge of the deterministic laws and of the state of the world, and she designs the zygote precisely so that, thirty years later, the resulting man Ernie will perform a particular action at a particular time. Ernie grows up normally. He deliberates in the ordinary way, satisfies every compatibilist condition, and does what Diana designed him to do.

Mele's argument then has three lines. Ernie is not responsible for that action. There is no relevant difference between Ernie and any ordinary agent in a deterministic world, since the only difference is that Diana intended the outcome and intentions of a distant third party do not change what happens inside Ernie. Therefore no agent in a deterministic world is responsible.

The second premise is what gives the argument its force. Everything Diana does is done by the initial conditions in an undesigned deterministic world too. She adds a designer's intention and subtracts nothing from Ernie.

Example. A compatibilist replies that Ernie is responsible after all, but that Diana is also responsible, as a puppet-master is. Does that answer the argument?

Not by itself, though it is a step. Shared responsibility is a familiar phenomenon: a person who deceives another into causing harm is culpable without the deceived party being wholly innocent, so the presence of a responsible designer does not by itself exculpate Ernie. But the argument does not need Ernie to be innocent because Diana is guilty. It needs Ernie to be relevantly identical to an undesigned determined agent, and the reply concedes exactly that by locating the difference in Diana rather than in Ernie. If Ernie is responsible, the compatibilist has answered the argument, and the reply about Diana is then a piece of bookkeeping rather than a defence.

Now you. A critic says the four-case argument is a sorites, like the paradox of the heap, and that sorites arguments are known to be fallacious. Is that a good objection?

The soft line: find a difference

The first family of replies accepts that Plum is not responsible in the early cases and looks for a condition he fails there and meets later. This requires historical compatibilism: the claim that whether an action is free depends not only on its structure at the moment of acting but on how the agent came to be that way.

Fischer and Ravizza's ownership condition, from the previous lesson, is one such attempt. Plum in the second case never took responsibility for his own mechanism in the required sense, because the mechanism was installed rather than acquired through his own experience of himself as an agent. John Christman's account of autonomy takes a similar route: a desire is autonomous when the process by which it was acquired is one the agent could endorse on reflection, and covert programming fails that test while an ordinary upbringing passes it.

The difficulty is always the same, and it appears between Pereboom's second and third cases. Community training in infancy is not something a child consents to, endorses at the time, or could have resisted. If it counts as ownership, then it is hard to see why programming that produces exactly the same psychology does not, and if it does not count, then very few actual human beings own their mechanisms, which concedes most of what the incompatibilist wanted. Every soft-line reply lives or dies on that junction, and none has produced a criterion that most philosophers accept.

The hard line: bite the bullet

The second family of replies, defended by Michael McKenna among others, denies the first premise. If Plum genuinely meets every compatibilist condition, then Plum is responsible in all four cases, manipulation included, and our reluctance to say so is a mistake with an explanation.

This sounds outrageous until the dialectic is made explicit. The argument asks us to accept a verdict about a case described in a way that is guaranteed to trigger a reaction: white-coated neuroscientists, radio control, a designer goddess. Nobody's intuitions were formed on such cases. McKenna's point is that the argument is only as strong as the initial judgement, and that the judgement is being made by people who have not yet absorbed the stipulation that Plum's deliberation is entirely normal. Since it is the incompatibilist who is arguing from intuition here, the burden is not obviously on the compatibilist.

What ordinary people say

There is empirical support for that diagnosis, and it is worth reporting with its limits. Eddy Nahmias and colleagues found in 2005 and 2006 that when ordinary people are given a determinist scenario described concretely, without the word "determinism", roughly three quarters judge the agent to have acted of his own free will and to be blameworthy. Shaun Nichols and Joshua Knobe found in 2007 that the framing matters enormously: asked abstractly whether people in a fully deterministic universe are fully morally responsible, 86 percent said no, while asked about a specific man in that universe who murders his family, 72 percent said he is fully morally responsible. Nahmias and Dylan Murray have argued that when people do read determinism as excluding responsibility, they are misreading it as bypassing, the idea that our beliefs and desires make no difference to what we do, which the second lesson of this course showed determinism does not claim.

The limits of that evidence should be stated. Surveys record what people say about vignettes, not what is true, and both sides can point to results favourable to them: Hagop Sarkissian and colleagues found in 2010 that in the United States, India, Hong Kong and Colombia alike, most people judged our own universe to be indeterministic and gave incompatibilist answers to abstract questions. The most that can be concluded is that ordinary judgements are unstable across framings, which weakens any argument that leans on their being obvious, and the manipulation argument leans on exactly that.

Example. Quantify the framing effect in the Nichols and Knobe result and say what it does and does not show.

In the abstract condition, 86 percent denied full responsibility, so about 14 percent affirmed it. In the concrete condition, 72 percent affirmed it. The same question about the same deterministic universe therefore moves by $72-14=58$ percentage points depending on whether it is asked in general terms or about a named man who burned his family. What that shows is that at most one of the two responses can be tracking the philosophical question, since the metaphysics is identical in both vignettes; something else is moving people, and the leading candidates are affective engagement in the concrete case and a bypassing misreading in the abstract one. What it does not show is which response is correct. A survey can demonstrate that an intuition is unreliable, which is a real and useful result, and it cannot supply the truth the intuition was supposed to deliver.

Now you. A compatibilist cites the Nahmias results as evidence that ordinary people are compatibilists. What is the strongest reply available to an incompatibilist?

Example. Take the hard line for a moment. What must a compatibilist say about the first case, and what is the cost?

They must say that Plum, whose reasoning is produced in real time by radio from a team of neuroscientists, is morally responsible for the killing, provided the process running in him is his own reasons-responsive deliberation and he endorses what it produces. The cost is high and should not be minimised: this is a case where blaming Plum feels like blaming the tool rather than the hand. The compatibilist's mitigation is that our reaction is being driven by the vividness of the manipulators and by the natural assumption that Plum is being overridden, and that once the case is stripped to what it actually stipulates, a deliberating agent acting on endorsed reasons, the verdict looks different. Whether that mitigation is enough is a judgement call, and it is exactly the sort of judgement call this course wants you to make explicitly rather than by feel.

Now you. Which reply, soft or hard, is available to a semi-compatibilist who has already given up the ability to do otherwise, and why does the manipulation argument threaten them at all?

Where the argument leaves things

The manipulation argument is the best case for incompatibilism about sourcehood, and it has not been refuted. Nor has it been established. Its critical premise, that Plum is not responsible in the first case, is exactly as strong as an intuition about an artificial case, and the experimental work shows such intuitions to be sensitive to how the case is put.

What it has certainly done is force compatibilism to become historical. Almost nobody now defends the view that the internal structure of an agent at the moment of acting is the whole story, and the interesting question has become which histories are disqualifying. That is a question the earlier lessons could not even ask.

There is also a warning in it for the other side. If the argument works, it works by showing that having your psychology installed by causes outside your control removes responsibility. That premise does not stop with determinism: it applies to indeterministic worlds too, since nobody chooses the undetermined events that shape them either. An incompatibilist who takes the manipulation argument seriously is under pressure to say what a free agent's history could possibly look like. That is the libertarian's problem, and it is the next lesson.

Libertarianism is the view that we have free will and that determinism is therefore false, and it has to explain how an undetermined decision could be more under an agent's control than a determined one rather than less.

That is a real bill, and the position is not disreputable: roughly one philosopher in five holds it, including van Inwagen, who built the argument in the fifth lesson of this course. This lesson sets out what the view has to supply, the objection that all versions face, and the two families of answer. It is the fairest test of the view, because the strongest objection to libertarianism comes from within the incompatibilist camp rather than from compatibilists.

Three things the position must deliver

A libertarian owes three separate things, and the second and third are much harder than the first.

The first is that determinism is false, which the physics lesson said is a live option and not established either way.

The second is that the indeterminism is located where decisions are made. Undetermined decay in a distant star is no help. The indeterminism has to be in the process that produces the choice, at a time when it can make a difference to the outcome.

The third, and the one everything turns on, is that the indeterminism yields control rather than noise. If the undetermined element makes the decision less predictable without making it more the agent's own, the view has purchased randomness at the cost of authorship, and randomness excuses rather than empowers. Nobody blames a person for a seizure.

The rollback argument

Van Inwagen stated the difficulty in its cleanest form in 2000, in a paper whose title is a fair summary of his own position: free will remains a mystery.

Alice is deciding whether to tell a damaging truth. She deliberates, and at time $t$ she decides to tell it. Suppose the decision was undetermined: the state of the world and the laws left both outcomes open right up to the moment.

Now imagine God rewinds the world to a moment shortly before $t$ and lets it run again, with everything exactly as it was, and does this a thousand times. Since the decision is undetermined, we should not expect the same outcome every time. Suppose she tells the truth in 726 of the replays and lies in the other 274. That distribution is all there is to say: nothing about Alice differs across the replays, since the world is in exactly the same state each time, and no fact about her, her reasons, her character or her efforts explains why this replay went one way and that one went the other.

The conclusion is uncomfortable. If in the actual world she told the truth, that was the 0.726 branch coming up rather than the 0.274 branch, and the difference between the two is not attributable to anything about her. It looks exactly like luck. And if instead the replays all came out the same way, the decision was not undetermined after all.

The argument is not a proof, but it isolates the problem precisely. Indeterminism, placed at the moment of decision, appears to insert chance exactly where control was wanted.

Example. A libertarian replies that the rollback picture misdescribes the case, because in every one of the thousand replays Alice acts for reasons she has, and does what she wants. Is that a good reply?

It is a real observation and an incomplete reply. It is true that Alice has reasons for both options, since that is why the decision was hard, and true that whichever she takes she takes for a reason. What the argument asks is a different question: what explains the pattern of 726 to 274? Not her reasons, since they are identical across replays. Not her character, likewise. Nothing about her at all, since everything about her is held fixed by the stipulation. So the reply establishes that neither outcome would be alien to her, which is worth having, and does not yet establish that she controls which one occurs. The best versions of libertarianism start from exactly this observation and try to build the rest.

Now you. Does the rollback argument work equally against a compatibilist account of decision?

Event-causal libertarianism: Kane

Robert Kane's The Significance of Free Will of 1996 is the most developed attempt to make indeterminism yield control without invoking anything exotic.

Kane concedes that most actions are determined by the agent's character, and that this is fine: an honest person acting honestly out of settled habit is free enough, provided her character was formed by acts of the right kind. The freedom-conferring acts are the rare ones, which he calls self-forming actions. They occur when a person is genuinely torn, when two incommensurable sets of reasons pull in different directions, a businesswoman on her way to a meeting who sees an assault in an alley being his standard example.

In such a case, Kane argues, the agent makes two efforts at once. She is trying to make the moral choice and trying to make the prudential one, and the conflict is realised in the brain as competing processes that are indeterministic because they are sensitive to noise amplified through chaotic neural dynamics. Whichever effort succeeds, it succeeds by the agent's own effort of will, for her own reasons, and she endorses the outcome as hers. She therefore has plural voluntary control: whichever way it goes, she wanted it, tried for it, and is responsible for it.

The move is genuinely clever, because it converts the indeterminism from an intruder into a feature of a divided will. The residual objection, pressed hardest by Alfred Mele, is that it does not remove the luck but relocates it. Grant that Alice is responsible either way. Which effort in fact succeeds is still settled by nothing about her, and the difference between the world where she helps and the world where she goes to her meeting is still, in the rollback sense, a matter of chance. Kane's answer is that responsibility does not require an explanation of that difference, only that both outcomes be willed and endorsed. Whether that is enough is where the argument currently stands.

There is also an empirical bill. Kane's picture requires indeterminism to be amplified to the level of competing neural processes, and no such amplification has been demonstrated. The physics lesson set out why the candidate mechanisms are unpromising at body temperature, and Kane has always been candid that his account is a proposal about how the neuroscience might turn out rather than a report of it.

Agent-causal libertarianism

The other family takes the harder metaphysical route: the cause of a free decision is not a prior event at all, but the agent, as a substance.

The idea is old. Thomas Reid stated it in the Essays on the Active Powers of Man of 1788, and Roderick Chisholm revived it in 1964 in a lecture with a famous line: each of us, when we act, is a prime mover unmoved, causing events without being caused to cause them. Timothy O'Connor and Randolph Clarke have developed modern versions.

The attraction is that it answers the rollback argument directly rather than living with it. In the replays, what explains why Alice tells the truth this time is that she, the agent, exercised her causal power that way. The explanation terminates in a substance rather than in a further event, so there is no unexplained difference between events to be worried about.

The costs are three, and they are steep.

The first is that agent causation is unlike any causation elsewhere in nature. Every other causal relation science describes holds between events or states, and a relation whose first term is a persisting thing rather than an event is a new category admitted for one purpose.

The second is the timing problem. If Alice's causal power is not itself triggered by any prior event, what accounts for its being exercised at 3.42 in the afternoon rather than a minute earlier or later? Answering "she exercised it then" restates the fact; answering that a prior event prompted the exercise reintroduces event causation at the crucial point.

The third is fit with the brain. A decision is realised in neural activity, so the agent's causing must show up as neurons firing that would not otherwise have fired, or as chances being altered. Clarke's version accepts this and says agent causation works alongside event causation, adding a causal contribution within the space the indeterminism leaves. That is coherent, and it makes the theory empirically committed: there would be a real difference between a brain in which an agent acts and one in which the same physical antecedents run their course.

A third and smaller family, non-causal libertarianism, defended by Carl Ginet and Hugh McCann, denies that a free action needs any causing of the decision at all: basic actions have an intrinsic character of being an agent's own doing, and asking what caused them is a mistake. The standard objection is that this names the phenomenon rather than explaining the control, and few have been persuaded.

Example. Apply the timing problem to a mundane case: you decide, at some moment, to stand up and make tea. What must the agent-causal theorist say?

They must say that the standing up is caused by you, as a substance, and not by any prior event, while conceding that the moment at which the power is exercised has no event-causal explanation. Reasons can still be cited, since agent causationists usually allow that reasons make an exercise of the power more or less likely without necessitating it, but as soon as reasons raise probabilities we are back to a chancy connection between antecedent states and the decision, and the rollback argument applies to the residue. The honest statement of the position is that it purchases control by adding a primitive, and the question for a reader is whether the phenomenon to be explained is strange enough to justify a primitive. Note that this is exactly the sort of trade philosophers make elsewhere, and not automatically illegitimate.

Now you. What empirical result would embarrass a libertarian, and what would embarrass a compatibilist?

The argument from how it feels

Most people who hold a libertarian view hold it because of how deciding seems from the inside, and that deserves an honest hearing rather than a dismissal.

The claim has a distinguished history. Descartes said the freedom of the will is known without proof by experience of it. Thomas Reid built his philosophy on the deliverances of common sense, of which the sense of active power is one. Samuel Johnson gave the popular version in 1778: we know our will is free, and there's an end on't.

The evidence against treating the feeling as a report is now substantial, and it is not the Libet material. Daniel Wegner assembled it in The Illusion of Conscious Will in 2002. In the experiment he ran with Thalia Wheatley in 1999, two people jointly moved a computer mouse over a screen of objects; when a subject heard the name of an object shortly before the mouse stopped on it, they reported having intentionally stopped it there, even though the movement had in fact been made by a confederate. The sense of having willed an action can be produced by a thought that merely preceded the action and matched it.

A second body of work points the same way. Patrick Haggard and colleagues reported in 2002 that when a voluntary action produces a tone, people perceive the action as occurring later and the tone as occurring earlier than they really do, each by tens of milliseconds, so that cause and effect are pulled together in experience. This intentional binding effect is absent for involuntary movements produced by magnetic stimulation. The sense of agency is therefore something the brain constructs and can construct wrongly, rather than a channel through which the will is directly observed.

None of this refutes libertarianism, and it is important to say why not. The theory is a claim about the causal structure of decisions, and the fallibility of introspection about agency is a claim about how well we detect that structure. What the evidence does remove is the argument from obviousness, which was the position's main popular support. A libertarian who wants the phenomenology as evidence has to explain why a faculty demonstrably prone to false positives should be trusted on the one occasion when metaphysics depends on it.

Example. A libertarian argues: I directly experience the openness of my future when I deliberate, so at least one undetermined decision has been observed. Assess the argument.

The premise is a report about experience, and the conclusion is a claim about the world, so the argument needs a bridge, namely that this kind of experience is reliable about this kind of fact. That bridge is what the evidence undermines. What deliberation actually presents is that you do not yet know what you will do, which is true, and compatible with the outcome being fixed: a chess engine calculating its move does not know its move until it finishes, and nothing follows about indeterminism. The experience of openness is the experience of ignorance about your own future decision, which every party to this dispute agrees is real. Note that this is one of the rare arguments in the subject that both compatibilists and hard incompatibilists reject for the same reason.

Now you. Does the same reply work against the experience of effort in a difficult decision, which Kane appeals to?

The state of the position

Libertarianism is a minority view among philosophers and probably the majority view outside philosophy, which is an unusual combination and tells you something about how the position gets its support. What people report about their own decisions, that they could have done otherwise and that the choice was up to them, is exactly what the view says.

Its serious defenders are not naive about the difficulty. Van Inwagen accepts the Consequence Argument, believes we have free will, concludes that determinism is false, and writes that he has no idea how an undetermined choice can be an agent's own. That combination is intellectually honest and uncomfortable, and it is the position of somebody who is convinced by an argument they cannot complete.

There is one more line of attack, and it is the one that cuts deepest, because it applies to libertarians and compatibilists alike. Whatever a free decision requires, it will issue from the person you are, and you did not make yourself. The causes of your character reach back through your upbringing to your genes to conditions you had no part in. The next lesson but one turns that thought into an argument, with an empirical literature attached, and it is the strongest thing in this course. Before that, there is a body of evidence that is famous for settling the question and does not: what happens in a brain in the half-second before a movement.

In 1983 Benjamin Libet reported that a brain signal preceding a voluntary movement begins about a third of a second before the person feels themselves deciding to move, and the result has been reported ever since as the experimental refutation of free will.

The experiments are real, well designed for their era, and they replicate. What they show is considerably narrower than the headline, and the most interesting development is that the central signal is now widely thought to be something other than what everyone assumed it was. This lesson works through the measurements, because the honest assessment depends on the numbers rather than on the summary.

What was measured

The background is a discovery by Hans Kornhuber and Lüder Deecke in 1965. Averaging electroencephalogram traces backwards from the moment of a self-initiated movement, they found a slow negative drift over the motor areas beginning up to a second or two before the movement itself. They called it the Bereitschaftspotential, the readiness potential, and it became a standard tool.

Libet's addition was to time the subjective side. His subjects sat with a modified oscilloscope in front of them, on which a spot of light revolved like a clock hand, taking 2.56 seconds per revolution, so that each degree of arc corresponds to about 7 ms. They were asked to flex the wrist whenever they felt like it, with no preplanning, and afterwards to report where the spot had been at the moment they first became aware of the wish or urge to move. That reported moment is called W. Muscle onset was recorded by electromyography and used as time zero.

The results, from Brain in 1983, are the numbers everyone quotes. For spontaneous movements reported as unplanned, the readiness potential began about 550 ms before the muscle activity. The reported moment of awareness, W, came about 200 ms before muscle activity. So the brain signal precedes the felt decision by roughly 350 ms.

Libet checked the timing method rather than assuming it. Subjects were also asked to report the moment of a small skin stimulus delivered at a random time, and their reports were biased by a few tens of milliseconds, which lets the W reports be corrected for the same bias. This is often overlooked by critics, and it matters: the method has a known error and the effect is much larger than the error.

Libet himself did not conclude that free will is an illusion. He argued that since W falls about 200 ms before the muscle activity, and the final motor command occupies about the last 50 ms, there remains a window of roughly 150 ms in which the conscious subject can abort the movement. Conscious will, on his picture, is not the initiator but the editor: not free will but, in his phrase, free won't.

Example. Work out the two intervals in Libet's design and say what each one is evidence about.

The first interval is from the readiness potential to W: $550-200=350$ ms. This is the celebrated result, and it is evidence that measurable preparatory activity precedes the moment a subject reports first noticing an urge. The second is from W to muscle onset, 200 ms, and subtracting the roughly 50 ms of final motor command leaves about 150 ms. This is what Libet's veto proposal rests on, and it is the weaker of the two claims: it depends on a subject being able to cancel in a window they cannot report on afterwards, and the direct evidence for a veto is much thinner than the evidence for the timing.

Now you. A critic says the whole design is worthless because people cannot accurately time their own mental events. Take the objection seriously and say what survives it.

Ten seconds, and sixty percent

The most dramatic follow-up came from Chun Siong Soon, John-Dylan Haynes and colleagues in 2008, using functional magnetic resonance imaging. Subjects chose freely between pressing a left or a right button while watching a stream of letters, and reported which letter was on screen when they decided. Applying pattern classification to activity in frontopolar and parietal cortex, the researchers could predict which button would be pressed up to about 10 seconds before the reported decision, some eighteen times further ahead than Libet's readiness potential.

The number that matters is the accuracy, and it is about 60 percent against a chance level of 50. That is a real effect, statistically solid, and small. Betting on the decoder would win 60 times in 100 instead of 50, an edge of 10 percentage points. It is the signature of a weak prior bias, of the kind you would expect if a subject who pressed left three times running is slightly disposed to press right next, and it is nothing like a readout of a decision already taken.

A third strand comes from single neurons. Itzhak Fried and colleagues in 2011 recorded from electrodes implanted in patients with epilepsy for clinical reasons, and found that populations of a few hundred neurons in the supplementary motor area changed their firing rates progressively before the reported moment of decision, allowing the impending choice to be predicted with better than 80 percent accuracy a few hundred milliseconds ahead. This is a much stronger signal than the fMRI result, from a much more direct measurement, and it is the best evidence that the preparation is real neural activity rather than an artefact of averaging.

The accumulator

The deepest challenge to the standard interpretation is not a criticism of the experiments but a rival explanation of the readiness potential, published by Aaron Schurger, Jacobo Sitt and Stanislas Dehaene in 2012.

Start from a fact about the task. The subject is told to move whenever they feel like it, with no reason to prefer any moment. There is nothing to decide, so something has to break the symmetry, and the obvious candidate is ongoing spontaneous fluctuation in motor cortex activity. Model it as a noisy accumulator drifting up and down, with a movement triggered when the accumulated activity crosses a threshold.

Now consider what happens when you average trials backwards from the moment of movement, which is what the readiness potential is. The trials selected are exactly those in which the noise happened to be drifting upward towards the threshold. Averaging them produces a slow rising negativity before the movement in every trial, even though on no individual trial was there any decision at the moment the average appears to start rising. The readiness potential, on this account, is a picture of the noise that got selected, not a picture of an unconscious decision.

The model makes a prediction that distinguishes it. If subjects are interrupted by an occasional cue demanding an immediate movement, their reaction times should depend on where the fluctuation happened to be when the cue arrived: fast if it was near threshold, slow if it was far. Schurger's group tested this and found the predicted pattern. Later work, including a 2021 review by Schurger and colleagues, has continued to support the reinterpretation, and the readiness potential is no longer safely described as the neural signature of a decision.

There is a second constraint from a different direction. Uri Maoz and colleagues reported in 2019 that the readiness potential appears before arbitrary choices, of the pick-one-at-random kind Libet used, and is largely absent before deliberate choices with real stakes, such as which of two charities should receive a donation. If that holds, the entire literature has been measuring the neural correlate of picking rather than of deciding, and the extrapolation to meaningful choices was never licensed.

Example. A newspaper reports that scientists can predict your decisions ten seconds before you make them, so free will is dead. Rewrite the claim so that it is accurate.

Something like this: in a task where subjects press one of two buttons for no reason at all, a pattern classifier applied to fMRI data predicts which button will be pressed with about 60 percent accuracy, against 50 percent by chance, from activity several seconds before the subject reports deciding. That accuracy corresponds to a weak bias rather than a settled decision, the task involves no reasons and therefore no deliberation, and nothing in the result distinguishes a deterministic brain from an indeterministic one. The honest headline is that arbitrary picking is preceded by measurable brain states that partly bias it, which is what anyone on any side of this debate would have expected.

Now you. What would an experiment have to show in order to genuinely threaten the sort of free will philosophers argue about?

The veto, tested

Libet's own positive proposal, that consciousness retains a power of cancellation, went untested for thirty years and then was tested rather well.

Matthias Schultze-Kraft and colleagues reported in 2016 an experiment in which subjects played a game against a computer. Electroencephalogram signals were decoded in real time, and when the system detected the build-up preceding a movement it presented a stop signal. The question was whether a movement already under preparation could still be cancelled.

It could. Subjects successfully aborted movements after the preparatory activity had begun, which is a direct demonstration that the readiness potential does not commit anyone to anything. But the ability had a deadline: cancellation failed if the stop signal arrived later than roughly 200 ms before the movement, a point the authors called the point of no return, corresponding to the stage at which the final motor command is on its way.

Both halves of that result matter. The first vindicates Libet's veto in outline and further weakens the claim that the readiness potential is a decision. The second sets a real limit: there is a last moment after which nothing can be recalled, and it is a fraction of a second wide. What the experiment does not show is that the cancelling is done by something outside the brain's ordinary causal processes, and nobody involved suggested it was.

Example. How does the 2016 result bear on the standard interpretation of Libet's 1983 finding?

It undermines it directly. The standard interpretation is that the readiness potential is the brain deciding, several hundred milliseconds before the person believes they are deciding, so the conscious decision is a report on a decision already taken. If a movement preceded by that same build-up can still be cancelled, the build-up cannot have been the decision: a decision that can be reversed at will by the agent is a preparation, not a verdict. Combined with the accumulator model, which explains the signal as accumulated noise selected by the averaging procedure, very little is left of the original interpretation. The measurements were sound and the story attached to them has been substantially rewritten by the same experimental tradition, which is what a healthy field looks like.

Now you. Does the point of no return, roughly 200 ms before movement, restrict free will in any philosophically interesting way?

What people report about their own reasons

A quieter literature bears on responsibility more directly than any of the timing work, and it concerns not when decisions are made but how badly people know why they made them.

Richard Nisbett and Timothy Wilson reviewed the evidence in 1977 and reported experiments in which shoppers evaluating identical items in a row chose the rightmost far more often than the leftmost, and then denied firmly that position had anything to do with it, offering explanations in terms of quality instead. The reports were not lies. The subjects had no access to the process that produced the preference, and generated a plausible account in its place.

Later work has made the point sharper. Lars Hall and Petter Johansson demonstrated choice blindness in 2005: subjects who chose which of two faces they found more attractive were handed, by sleight of hand, the face they had rejected, and asked to explain their choice. A large majority failed to notice the swap and then explained, in detail, why they preferred the face they had in fact rejected. Split-brain patients studied by Michael Gazzaniga do something structurally identical: the verbal hemisphere, presented with an action initiated by information only the other hemisphere received, produces a confident reason for it.

The implication for this course is not that we never know why we act. It is that the faculty producing our accounts of our own reasons is a constructor rather than a reader, and that a theory resting on the agent's own report, of the kind the hierarchy of desires might seem to invite, is resting on something unreliable. This is one place where empirical work genuinely constrains the philosophy, and it favours accounts, like reasons-responsiveness, that test the mechanism by what it does rather than by what the agent says about it.

What the evidence establishes

Three conclusions, and a warning.

The first conclusion is that the folk picture of a conscious self standing outside the brain and starting the causal chain is not supported and probably false. Preparation for movement is underway before people report noticing an intention, and the report itself is partly reconstructed after the fact. That is a genuine finding about how the mind works, and it is not nothing.

The second is that this refutes almost nobody in the debate. Compatibilists have held since Hobbes that a free action is caused by processes in the agent, and are untroubled by the discovery that those processes begin before the agent notices them. Libertarians need the decision to be undetermined, and no timing experiment addresses determinism at all: a noisy accumulator can be deterministic or stochastic, and the data do not distinguish those cases.

The third is that the specific signal at the centre of the literature is now contested. If the readiness potential is a selection artefact, the most-cited experimental result in the philosophy of action turns out to be measuring the shape of neural noise.

The warning is against the opposite overreaction. It would be wrong to conclude that neuroscience has nothing to say here, and this lesson is not a defence of the will against science. The evidence that does bear on responsibility is quieter, better replicated and rarely reported as a free will story: how much of behaviour is heritable, how strongly circumstances move people who believe themselves to be acting on principle, and what a tumour can do to a character. That is the next lesson, and it is the one that should worry you.

The evidence that should unsettle you about responsibility is not about the half-second before a movement; it is about where your character came from, and none of it was chosen.

The previous lesson concluded that the famous timing experiments refute a picture of the will that few defenders of free will hold. This lesson assembles the material that does bear on the question, then states the argument that turns it into a conclusion. The argument is a priori and does not strictly need the evidence, but the evidence is what makes it impossible to shrug off.

Constitutive luck

Thomas Nagel and Bernard Williams gave the problem its name in a joint symposium in 1976, published soon after as two essays called "Moral Luck". Nagel distinguished four kinds, and the taxonomy is worth having.

Resultant luck concerns how things turn out: two equally reckless drivers, one of whom meets a child in the road. Circumstantial luck concerns the situations you happen to face: an ordinary German in 1935 was tested in a way an ordinary Argentine was not. Constitutive luck concerns the kind of person you are, your temperament, capacities and inclinations. Causal luck is the free will problem itself, the luck of being determined by antecedent circumstances.

Nagel's observation is that we officially believe people are assessable only for what is within their control, and that once the four kinds of luck are subtracted almost nothing is left to assess. The area of genuine agency shrinks to an extensionless point. He does not resolve the tension, and the essay is more valuable for refusing to.

Constitutive luck is the one this lesson is about, and it is the one that has acquired numbers.

Heritability, and how to read it

The largest relevant study is a meta-analysis by Tinca Polderman and colleagues, published in 2015, which pooled essentially every twin study published between 1958 and 2012: 2,748 publications, 14,558,903 twin pairs, and 17,804 measured traits. Across all of them the average heritability was 49 percent, and for most trait domains the data were consistent with a simple model in which the resemblance between relatives is due to genetic effects that add up.

The classical method behind those numbers is straightforward enough to do by hand. Identical twins share all their genes, fraternal twins about half, and if both kinds are reared together the shared environment is common to both. Falconer's formula estimates heritability as twice the difference between the two correlations, $h^2=2(r_{MZ}-r_{DZ})$, and shared environment as $c^2=2r_{DZ}-r_{MZ}$. Take the classic figures for IQ from Thomas Bouchard and Matthew McGue's 1981 review, an identical-twin correlation of 0.86 and a fraternal-twin correlation of 0.60. Then $h^2=2(0.86-0.60)=0.52$ and $c^2=2(0.60)-0.86=0.34$, leaving 0.14 for everything else including measurement error.

Now the qualifications, because heritability is the most misunderstood statistic in the sciences.

Heritability is a property of a population in an environment, not of a person. It says how much of the variation in a trait, in that population, is associated with genetic variation. It does not say that half of your intelligence comes from your genes, which is not a meaningful claim. Change the environment and the number changes: if every child were given identical schooling, the heritability of educational attainment would rise, because the environmental variance that used to contribute would be gone. High heritability does not mean unchangeable, and the standard counterexample is the height gained across the twentieth century by better nutrition in populations where height was always highly heritable.

What the twin literature does support is Eric Turkheimer's three laws, stated in 2000. First, all human behavioural traits are heritable. Second, the effect of being raised in the same family is smaller than the effect of the genes. Third, a substantial portion of the variation is explained by neither, which means by the accumulation of idiosyncratic experiences nobody planned.

Read those three together and the point for our subject is clear. Your dispositions came from your genes, which you did not choose; from your family, which you did not choose; and from a mass of contingent experience you did not select either. There is no fourth category in the data where the self-made part would live.

Example. A trait shows an identical-twin correlation of 0.45 and a fraternal-twin correlation of 0.22. Estimate the heritability and the shared environment component, and say what the second number means.

By Falconer's formula, $h^2=2(0.45-0.22)=0.46$ and $c^2=2(0.22)-0.45=-0.01$, which rounds to zero. Almost half the variation tracks genetic variation, essentially none of it tracks growing up in the same household, and the remaining 54 percent is non-shared environment plus measurement error. That pattern is extremely common for adult personality traits, and it is the one that surprises people: the family that everyone assumes formed them shows up as close to nothing in the variance, once genetic resemblance is accounted for. A negative estimate, incidentally, is a reminder that these are estimates from a model, not measurements of a quantity, and that a small negative value means the model is being pushed slightly past its assumptions.

Now you. Someone objects that if a trait is 50 percent heritable then a person is 50 percent responsible for it at most, and that upbringing accounts for the rest. What has gone wrong?

Situations, and how much they move people

The second body of evidence concerns circumstantial luck. The best known results are old and have been argued over ever since.

Stanley Milgram reported in 1963 that 26 of 40 subjects, 65 percent, continued administering what they believed to be electric shocks up to the maximum 450 volt setting when instructed by an experimenter in a laboratory coat. John Darley and Daniel Batson reported in 1973 that among seminary students walking to give a talk, 63 percent stopped to help a man slumped and groaning in a doorway when they were told they had plenty of time, against 10 percent when told they were already late. The subject of the talk they were about to give, in half the cases the parable of the Good Samaritan, made no significant difference.

These have to be reported with their weaknesses. Gina Perry's archival work published in 2013 showed that Milgram's procedure varied more between subjects than his report suggested, that some subjects disbelieved the setup, and that the headline 65 percent comes from one of many conditions with results ranging from near zero to near total compliance. The Good Samaritan study has a small sample. The broader replication crisis in social psychology has been hard on this literature, and a reader should treat single striking studies as suggestive rather than settled.

What survives is the direction of the effect and its size relative to what people predict. Asked in advance, observers massively underestimate compliance in Milgram-type situations and massively overestimate the effect of the seminarians' beliefs. John Doris's Lack of Character of 2002 drew the philosophical conclusion: the trait-based picture of character, on which a person's honesty or kindness is a stable disposition that predicts behaviour across situations, is not well supported, and behaviour is more situation-dependent than either folk psychology or virtue ethics assumes.

For our purposes the moral is narrow but real. Whatever moves people is often not what they believe is moving them, and the circumstances that do the moving are not chosen.

Example. In the Good Samaritan study, helping fell from 63 percent to 10 percent between the unhurried and hurried conditions. State precisely what that licenses and what it does not.

The ratio is $63/10=6.3$, so a seminarian who was not running late was more than six times as likely to stop. What it licenses is a claim about a difference between groups: an unchosen feature of the situation, being told the time, moved behaviour far more than the content of the talk the subjects were about to give, including for those about to speak on this very parable. What it does not license is any claim about a particular individual. Ten percent of the hurried students did stop, so the situation was not compelling, and the study gives no way to tell whether a given non-stopper would have stopped had he been unhurried. The sample was also small, one condition among several in a single 1973 experiment, so the effect size should be treated as an estimate with wide uncertainty rather than as a constant of human nature.

Now you. Does the situationist evidence excuse the hurried seminarians?

When a cause is visible

The third body of evidence is clinical, and it is the most vivid.

In 2003 Jeffrey Burns and Russell Swerdlow reported the case of a 40-year-old man with no prior history who developed an intense interest in child pornography and made advances to his stepdaughter. He was convicted, and on the eve of sentencing complained of headache and was found to have a large tumour in the right orbitofrontal region. It was removed, and the behaviour resolved. Some months later the urges returned; imaging showed the tumour had regrown; a second resection resolved them again. The temporal coupling is about as close to an experiment as clinical medicine gets.

The pattern is old. Phineas Gage's frontal injury in 1848 produced a documented change in temperament. Charles Whitman, who in 1966 killed his wife and his mother and then fourteen more people, most of them shot from a tower at the University of Texas, had written that he suspected something was wrong with his own mind and asked for an autopsy; a glioblastoma was found, and the commission convened by the governor concluded that it might have contributed to his actions. The syndrome of acquired sociopathy after orbitofrontal damage is well described in the neurological literature.

Now the question that makes this philosophy rather than medicine. Nearly everyone's reaction to the tumour case is that responsibility is reduced or removed. Ask why. If the answer is that his behaviour was produced by a physical process he did not choose and could not control, then the same description fits a man whose sadism was produced by a childhood he did not choose and genes he did not select. If the tumour excuses, what exactly is the feature it has that the childhood lacks?

There are answers available, and the best of them is reasons-responsiveness from the compatibilist lesson: the tumour patient's behaviour did not respond to reasons, punishment, or the prospect of losing his family, while an ordinary offender's does. That answer is respectable, it is testable in principle, and it draws the line somewhere other than where causal history lies. Whether it holds is exactly what the manipulation argument disputes.

The Basic Argument

Galen Strawson published the sharpest version of the underlying argument in 1994, and called it the Basic Argument. It runs in five steps and uses no empirical premises at all.

You do what you do, in any situation, because of the way you are. To be truly responsible for what you do, you must therefore be truly responsible for the way you are, at least in certain crucial mental respects. To be truly responsible for the way you are, you must have chosen to be that way. But any such choice must itself have been made for reasons, in the light of principles and preferences you already had, so you must already have been a certain way in order to make it. And to be responsible for that way, you would have had to choose it earlier, which requires an earlier self with its own character, and so on without end.

The regress cannot be completed, so true self-determination is impossible, so true moral responsibility, of the kind that would make punishment deserved in the deepest sense, is impossible. Note the reach of the conclusion. It does not depend on determinism: adding undetermined events to the story gives you a character partly formed by chance rather than by you, which is not an improvement. It is impossible in every possible world, which is why Strawson calls the demand incoherent rather than merely unsatisfied.

Example. Which premise does a compatibilist deny, and how?

The second: that being responsible for what you do requires being responsible for the way you are. Compatibilists hold that responsibility is a matter of how an action is produced now, by a reasons-responsive mechanism the agent owns, and not a matter of the history by which that mechanism came to exist. On this view the demand for self-creation is imported from nowhere and answers to nothing in our practices, which never ask an offender whether he chose his own character. The reply that the Basic Argument's defenders give is that the practices in question involve blame and punishment, that these are supposed to be deserved rather than merely useful, and that desert is precisely what an unchosen character cannot support. That exchange is the whole dispute in two sentences, and it is where the last three lessons of this course go.

Now you. A libertarian says agent causation blocks the regress, because the agent originates the choice rather than making it out of a prior character. Does it work?

What follows

Nothing in this lesson proves that nobody is responsible for anything. The empirical results establish that our characters have specific, measurable, unchosen causes, which almost everybody already believed; the Basic Argument establishes that a particular demanding conception of responsibility cannot be satisfied, if its second premise is granted.

What the two together do is shift the burden decisively. Anyone who wants to keep robust desert has to say what makes an unchosen but reasons-responsive character sufficient for it. Anyone who wants to abandon desert has to say what happens to blame, resentment, gratitude, punishment and the ordinary business of holding each other to account.

That second question turns out to be the more interesting one, and the most influential paper in the modern literature answers it by refusing the framing of the whole debate. It was written by a different Strawson, in 1962, and it is the next lesson.

Resentment, gratitude and indignation are not conclusions anyone reached from a theory, and Peter Strawson's 1962 lecture argued that this fact settles more of the free will problem than any argument about determinism.

The previous lesson left a threat: our characters have causes we did not choose, and a regress argument says no character could have been chosen. If that undermines responsibility, everything built on responsibility goes with it. This lesson takes the most influential response in the modern literature, which is to ask what our responsibility practices actually consist of before asking whether determinism undermines them.

Strawson's framing

"Freedom and Resentment", delivered to the British Academy in 1962, opens by dividing the field into two camps that Strawson thinks are both mistaken.

The pessimist holds that moral responsibility requires a kind of freedom that determinism would rule out, and worries that we may not have it. The optimist is the old-fashioned compatibilist who says that responsibility is justified because blame and punishment are useful: they modify behaviour, and that is what entitles us to them.

Strawson thinks the optimist has left something out, and that the pessimist is right to feel the omission. Nobody resents an injury because resentment is socially efficacious. The optimist's story describes a regulation system, not the thing that actually goes on between people, and its coldness is what keeps the pessimist unsatisfied. But the pessimist's remedy, a metaphysical freedom that would license desert, is a demand for something the optimist rightly cannot supply.

His alternative is to look at the phenomenon itself.

The reactive attitudes

What we actually do, constantly and without theorising, is respond to the quality of other people's wills towards us. Someone treads on your hand: whether they did it carelessly, deliberately, or while being pushed changes your response completely, and the change is immediate and not a calculation.

Strawson calls these responses the reactive attitudes. Resentment and gratitude are the personal ones, felt from the standpoint of someone affected. Moral indignation and approval are the vicarious ones, felt on behalf of others. Guilt, shame and remorse are the self-directed ones. Love and hurt feelings belong on the list too, which is a mark of how wide the family is.

Holding someone responsible, on this account, is not a judgement that gets expressed in an attitude. It is the attitude. To regard someone as answerable simply is to be liable to these responses towards them, and this stance, which Strawson calls the participant stance, is the ordinary condition of being in a relationship with another person.

The alternative is the objective stance. We can look at a person as an object of policy, something to be managed, treated, avoided or trained, rather than as a party to a relationship. We do this with small children, with people in the grip of psychosis, and occasionally with anyone when a relationship becomes too much to bear. Strawson's key observation is that the objective stance is available, is sometimes appropriate, and cannot be adopted wholesale for everyone all the time, because doing so would end interpersonal life rather than reform it.

Example. A colleague snaps at you in a meeting. Describe the same episode from the participant stance and from the objective stance, and say what changes.

From the participant stance you are hurt or annoyed, you take the remark as expressing something about his attitude towards you, and the natural next moves are to protest, to demand an explanation, or to wait for an apology. From the objective stance you note that he has been sleeping badly since his father's illness, that people under strain snap, and that the remark carries no information about his regard for you; the natural next moves are patience, a quiet word with someone else, or simply avoiding him this week. What changes is not the facts but whether he is being treated as a party to a relationship or as a system with known failure modes. Notice that the objective stance here is not cold, and it is not a denial of his agency in general: it is local, temporary, and often the kindest available response, which is exactly Strawson's point about its role in ordinary life.

Now you. If the objective stance is always available, why does Strawson think determinism could not move us to adopt it universally?

Excuses and exemptions

The best part of the lecture is the taxonomy, because it shows what our practices are actually sensitive to.

Some pleas work by showing that the quality of will was not what it appeared. "He didn't realise", "he was pushed", "he had no way of knowing", "it was an accident": these are excuses. They do not remove the person from the community of the responsible. They say that in this instance the injury did not express ill will, and the natural response is to withdraw the resentment while continuing to regard the person as a full participant.

Other pleas work differently. "He is only three", "he is schizophrenic", "he was under hypnosis": these are exemptions. They invite us to see the agent as not a fit subject for the reactive attitudes at all, at least for a period or in a domain, and the appropriate response is the objective stance.

Now the argument. Look at the list of things that actually excuse and exempt, and notice what is not on it. No ordinary excuse takes the form "his action was determined by prior causes". The excusing conditions are ignorance, accident, coercion, compulsion and incapacity, each of which is a specific local fact about how this act came about or about what this agent can do. Determinism is not one of them, and if it were true it would be true of every action, including the ones we excuse and the ones we do not, so it could not draw the distinctions our practice draws.

The pessimist replies that determinism is not one excuse among others; it is a global reason to move everyone to the objective stance. Strawson's answer to that is the one people find either liberating or infuriating. The commitment to ordinary interpersonal relationships is too thoroughgoing and deeply rooted for a general theoretical conviction to overturn it. We could not do it, and the question whether it would be rational to do it does not arise in the way the pessimist thinks, because the rationality of the whole framework is not something assessable from outside it: what could be produced as an external justification that would be more secure than the practice it was justifying?

Example. Sort these pleas into excuses and exemptions: (a) "I didn't see you standing there." (b) "She has advanced Alzheimer's disease." (c) "He was acting under a credible threat to his family." (d) "He is six years old."

(a) and (c) are excuses. In each the agent remains a full member of the moral community, and the plea works by showing that this particular act did not express the ill will it appeared to: the first denies knowledge, the second denies that the act expressed the agent's own attitude towards you rather than a response to a threat. (b) and (d) are exemptions. Neither says anything about the particular act; both say that this agent is not, for now or at all, a fit target of resentment and indignation, and both invite the objective stance. The test that separates them is whether the plea generalises across the agent's whole conduct or applies to one act.

Now you. A defence lawyer argues that her client's violent upbringing means he should not be blamed. Is she offering an excuse, an exemption, or something the Strawsonian framework has trouble accommodating?

What the argument does and does not establish

Strawson's lecture is the most cited paper in this literature, and it repays scepticism as well as admiration.

Its strongest achievement is descriptive. It shows that responsibility practices are not built on a metaphysical premise about causation, and that the theoretical dispute has been misdescribing what it is about. Anyone who says that discovering determinism would require us to stop blaming has to explain why no actual excuse in any actual practice has that form.

The first serious objection is that unrevisability is not truth. Suppose the reactive attitudes really cannot be given up. It does not follow that they are appropriate, any more than the unavoidability of an optical illusion makes the illusion accurate. Galen Strawson and Derk Pereboom both press this: the attitudes have a content, they present the target as deserving the response, and that content can be false even if the attitude is inescapable.

The second objection is that the practice is more revisable than the lecture allows. We have in fact revised it, repeatedly and in the direction the pessimist predicts. Children, the insane, the addicted, the coerced and the neurologically damaged have been progressively moved out of the scope of full blame over two centuries, usually as understanding of their condition improved. That is precisely the pattern you would expect if learning the causes of behaviour did tend to dissolve blame, and it suggests the boundary between participant and objective stances is negotiated rather than fixed.

The third is that Strawson's optimist and pessimist do not exhaust the field. A modern hard incompatibilist does not propose to view everyone objectively. Pereboom's position keeps love, gratitude, and moral protest while giving up resentment and indignation, on the ground that only the latter presuppose desert, and argues that a life so organised is not colder but somewhat better. Whether the attitudes can be separated in that way is an empirical question about human psychology as much as a philosophical one.

Three things "responsible" means

The literature after Strawson has done something useful with his framework: it has separated the senses of the word, which had been running together for centuries.

Attributability is the weakest. An act is attributable to you when it expresses your evaluative judgements, your cares and commitments, so that it tells us who you are. A cruel remark is attributable to the person who makes it, and that is why it damages our view of them, quite apart from any sanction.

Answerability is stronger. You are answerable for an act when it is appropriate to ask you to justify it, and when your reasons for doing it are a fit topic of demand. Asking a colleague why they did something, and expecting an account, is holding them answerable.

Accountability is strongest, and it is the one that matters most here. You are accountable when it is appropriate to impose sanctions and to direct the harsher reactive attitudes at you: resentment, indignation, blame with teeth. Gary Watson introduced the distinction between the first and the third in 1996, and David Shoemaker separated all three.

The distinction matters because the sceptical arguments do not bite equally. Attributability seems safe: a determined agent's actions still express their evaluative outlook, and nothing in the manipulation argument suggests otherwise. Answerability is largely safe too, since a reasons-responsive agent can give an account of their reasons whatever their history. What the arguments threaten is accountability in its strongest form, and specifically what Pereboom calls basic desert: the claim that an agent deserves blame or punishment just because of what they have done, and not because of any good the blaming or punishing will do.

That is the real currency of the dispute. Everything in this course has been circling it, and the practical question of what follows if it goes is the next lesson.

Example. A colleague makes a cutting remark in a meeting. Apply the three notions.

The remark is attributable to him: it expresses his judgement of the person, and it tells you something true about him even if he later apologises. He is answerable for it: it is entirely in order to ask him afterwards why he said it, and to expect a justification or an admission that there is none. Whether he is accountable, in the sense of deserving a sanction or a hostile response, is a further question that depends on more: whether he knew what he was doing, what his circumstances were, and, according to some, on the history that made him the sort of person who says such things. The three come apart cleanly here, and noticing that is the practical benefit of the distinction: a sceptic about desert who says "nobody is ever responsible for anything" is denying much more than they need to.

Now you. Pereboom holds that we should give up resentment and indignation but keep moral protest, and that this is possible without ending human relationships. What is the strongest objection?

Where this leaves the practice

Strawson's contribution is best taken as a relocation rather than a refutation. He moved the question from "does determinism make blame unjustified?" to "what are our practices sensitive to, and could that sensitivity survive the discovery?" That is a better question, and it can be pursued with evidence.

He did not show that basic desert is safe. He showed that a great deal of what we call holding people responsible does not require it: attribution of character, demands for justification, the whole texture of relationships that make ill will matter. What remains in dispute is the hardest and most consequential part of the practice, the part where we impose suffering on people because they deserve it.

That is not an abstract remainder. It is the criminal law of every country, and the next lesson takes the argument there, where the claims are testable and the stakes are measured in years.

Of the four standard justifications for punishing people, only one needs the metaphysics this course has been arguing about, and it is the one most legal systems say they are using.

The previous lesson isolated basic desert as the contested item: blame and punishment deserved just because of what someone has done, apart from any good it produces. Here that abstraction meets the institution that spends it, and the argument becomes partly empirical. Where numbers exist they are given, because a claim about deterrence is a claim about the world and not a matter of taste.

Four justifications, and which need what

Retribution holds that a wrongdoer deserves to suffer in proportion to the wrong, and that inflicting that suffering is intrinsically justified. Kant put the position at its starkest in the Metaphysics of Morals of 1797: if a civil society were to dissolve by common agreement, the last murderer in prison must first be executed, so that everyone receives what their deeds deserve. Nothing about consequences appears in that argument, and it is entirely dependent on basic desert.

Deterrence holds that punishment is justified by the offences it prevents, whether by dissuading the offender or others. Incapacitation holds that it is justified by preventing an offender from reoffending while confined. Rehabilitation holds that it is justified by changing the offender.

The last three are forward-looking and need no desert at all. They need punishment to work, which is an empirical question, and they need some constraint to stop them licensing outrages, which is a moral question. So a sceptic about free will does not have to abolish the criminal law. They have to give up one justification of four, and then explain how the remaining three are kept within limits that used to be supplied by desert.

That is the real practical disagreement, and it is narrower than the rhetoric on either side.

The classic objection to going forward-looking

The reason desert kept its place is that consequentialist justifications alone permit things nobody will accept.

The standard case, put by H. J. McCloskey in 1957, is the sheriff in a town on the edge of a riot. He knows that framing and hanging one innocent man will prevent a lynching in which several people would die. If punishment is justified by consequences, and the consequences of framing him are better, then he should frame him. Nearly everyone thinks he must not, and desert is the usual explanation: punishment must fall on the guilty because only the guilty deserve it.

Deterrence has a second problem of scale. If the point is to prevent offences, and severe punishments prevent more, the theory has no natural stopping point: execution for shoplifting would deter shoplifting. Desert supplies the ceiling, in the form of proportionality.

Any sceptic about basic desert has to answer both, and the leading attempt is worth taking seriously.

The quarantine model

Derk Pereboom, developed further with Gregg Caruso, argues that the right analogy for dealing with dangerous offenders is public health rather than deserved suffering.

We already detain people who have done nothing wrong when they carry a dangerous infection. The justification is the right of self-defence and the defence of others, which does not require the carrier to deserve anything. Four constraints come with the analogy, and they are not decorative. Detention is permitted only for genuine danger. It must use the least restrictive means adequate to the danger. It carries an obligation to rehabilitate or cure where possible. And it ends when the danger ends.

Applied to crime, the model gives a system that incapacitates the dangerous, works hard on rehabilitation, and inflicts no suffering beyond what the containment requires. It answers the sheriff case cleanly: the innocent man is not dangerous, so quarantine gives no ground whatever to detain him, and the analogy blocks the very move that embarrassed pure deterrence. It also implies a serious reduction in punitive severity, since nothing in the model justifies harsh conditions that do not reduce danger.

The strongest objection is about proportionality in the other direction. A persistent but minor offender, say a burglar with a high probability of reoffending, may be more dangerous than a man who killed his wife's lover in circumstances that will never recur. Desert says the killer should serve much longer. Quarantine says the opposite, and is committed to detaining the burglar for as long as the danger lasts, which could be indefinitely. Critics regard indefinite detention for minor offences as a decisive objection; Caruso replies that the least-restrictive-means requirement and the obligation to address the causes of the danger together prevent the worst outcomes, and notes that existing systems already do detain minor offenders repeatedly and for long total periods without admitting it.

A second objection is that the model still uses people: detaining a person for what they might do treats them as a hazard. The reply is that self-defence has always permitted this, and that the alternative, deliberate infliction of suffering on the ground that they deserve it, is not obviously more respectful.

Example. A man drives drunk, kills a cyclist, and is genuinely remorseful. Assess him under each of the four justifications.

Retribution asks what he deserves for a serious wrong committed through culpable recklessness, and answers with a sentence proportional to the gravity of the harm and the fault, typically years. Deterrence asks what sentence will most reduce drink-driving by him and by others, and the empirical literature suggests the answer depends far more on the perceived probability of being caught than on the length of the sentence, so it favours breath testing over long terms. Incapacitation asks how dangerous he is now, and a remorseful first offender who has lost his licence is not very dangerous, which points to a short sentence or none. Rehabilitation asks what will change him, and answers with treatment for alcohol dependence if he has it. Only the first justification gives a reason for a long sentence, and noticing that is the point of the exercise: much of what our systems do is retributive in fact whatever they say in their sentencing guidelines.

Now you. Under the quarantine model, what would justify keeping this man in prison at all, and for how long?

What the numbers say

Three empirical claims bear on the argument, and all three are better established than most of what is said about them.

The first concerns deterrence. Reviews of the evidence, notably by Daniel Nagin, converge on the finding that the certainty of apprehension deters, while the severity of the sentence has little detectable marginal effect. Increases in police presence and clearance rates reduce crime; increases in sentence length, at the margins tested, mostly do not. That result damages the deterrence justification for long sentences specifically, which is the part of the system that consumes most of the money.

The second concerns scale. The World Prison Brief puts the United States at roughly 530 prisoners per 100,000 residents and Norway at roughly 55, a factor of about ten. Norway's maximum determinate sentence is 21 years, with a separate preventive detention regime, forvaring, extendable in five-year increments for offenders judged still dangerous, which is close to a quarantine model operating inside an otherwise ordinary system. Anders Breivik, who killed 77 people in 2011, received exactly that: 21 years of preventive detention, extendable indefinitely.

The third concerns reoffending, and it is where care is needed. The United States Bureau of Justice Statistics followed 404,638 prisoners released in 30 states in 2005 and found that 67.8 percent were rearrested within three years and 76.6 percent within five. Norwegian figures are usually quoted at around 20 percent within two years. Those numbers are not comparable: the American figure counts rearrest, the Norwegian counts reconviction, the follow-up periods differ, and the underlying populations and policing differ enormously. The comparison is suggestive and it is not a controlled experiment, and anyone who cites it as one is overreaching. What can be said is that a system with a tenth of the incarceration and a much lower ceiling on severity does not display the catastrophic outcomes that a purely deterrent theory would predict.

Example. The United States imprisons about 530 people per 100,000 and Norway about 55. What can and cannot be inferred from the comparison?

The ratio is $530/55=9.6$, so the American rate is nearly ten times the Norwegian one, and that much is a measurement rather than an inference. What cannot be read off it is a causal claim in either direction. The two countries differ in the prevalence of violent crime, in firearm availability, in inequality, in drug policy, in policing, in what counts as a prison sentence, and in a hundred other respects, and any of these could account for part of the gap. The honest use of the comparison is as an existence proof: a wealthy society can run a criminal justice system at a tenth of the American incarceration rate, with a 21 year ceiling on determinate sentences, without collapsing. That rules out the strongest version of the claim that severe punishment is necessary for social order, and it establishes nothing about what would happen if one country adopted the other's system.

Now you. What kind of evidence would settle whether increasing sentence length reduces crime?

The law's own compromise

Legal systems have never waited for philosophers, and what they have built is a set of exemptions that track capacity rather than causal history.

The M'Naghten Rules, formulated by the English judges in 1843 after Daniel M'Naghten was acquitted of killing the Prime Minister's secretary, excuse a defendant who, through a defect of reason from disease of the mind, did not know the nature and quality of his act, or did not know that it was wrong. The American Law Institute's Model Penal Code of 1962, 119 years later, broadened this to a defendant who lacks substantial capacity either to appreciate the criminality of his conduct or to conform his conduct to the law, which adds a volitional limb.

Notice that both tests are about capacity now, not about how the defendant came to have his character. That is a compatibilist structure, and it is what almost every criminal code in the world uses in practice.

The defence is also far rarer and far less successful than its cultural prominence suggests. An eight-state study by Callahan and colleagues published in 1991 found the insanity defence raised in about 0.9 percent of felony cases and successful in about 26 percent of those, which is roughly two acquittals per thousand felony cases. The political pressure has run in one direction: after John Hinckley's acquittal for the attempted assassination of President Reagan, the federal Insanity Defense Reform Act of 1984 narrowed the test and shifted the burden to the defendant, several states abolished the defence outright, and in Kahler v. Kansas in 2020 the Supreme Court held by six votes to three that the Constitution does not require a state to offer a moral-incapacity test at all.

Does believing this change behaviour?

One practical worry deserves an honest report, because it is often used as a reason not to discuss the subject: that telling people they lack free will makes them behave worse.

Kathleen Vohs and Jonathan Schooler reported in 2008 that subjects who read a passage denying free will subsequently cheated more on a task than controls. The finding was widely repeated and became a standard warning.

It has not held up well. Larger replication attempts, including work by Andrew Monroe, Garrett Brady and Bertram Malle in 2017 and by Damien Crone and Neil Levy in 2019 using substantially larger samples, failed to find the effect, and a 2023 meta-analysis by Oliver Genschow and colleagues put the average effect close to zero. The current state of the evidence is that reading a paragraph about determinism does not measurably make people cheat, which is what one might expect of a manipulation that mild.

That is a limited result, and it should not be inflated either. It says nothing about what a society organised around scepticism about desert would be like, since no such society has existed, and the honest position is that this is unknown. But the specific claim that the belief is dangerous, which has been used to close the discussion, is not currently supported.

Example. A retributivist argues that abandoning desert would make punishment unlimited, since a system aimed only at prevention has no ceiling. How should a sceptic reply?

By pointing out that the ceiling has to come from somewhere else, and then supplying it. The quarantine model's least-restrictive-means requirement is such a ceiling, derived from the same self-defence principle that licenses the detention in the first place: you may use the minimum force needed to avert a threat and no more, which is exactly why a person who has become harmless must be released. That is a genuine constraint with a familiar structure in law. Whether it is as protective as proportionality is arguable, and the honest sceptic concedes that it protects differently: it is more protective of the reformed serious offender and less protective of the persistent minor one. The reply is not that nothing changes. It is that the change is specific and defensible.

Now you. Which of the four justifications does a compatibilist of the reasons-responsive kind need, and does the argument of this course threaten it?

Where the practical argument stands

The gap between a retributivist and a sceptic about desert is smaller in institutional terms than in rhetorical ones. Both detain dangerous people. Both prefer prevention to cure. Both accept excuses and exemptions that track capacity. They differ on whether suffering beyond what containment requires is ever warranted, and on whether the length of a sentence should track the gravity of the past act or the size of the future risk.

That is a real difference, and it is measurable in years of human life. It is also, unusually for this subject, a disagreement where evidence helps: how much deterrence severity actually buys, what rehabilitation achieves, how accurately danger can be predicted. Those questions have answers, and the answers have been arriving steadily for thirty years.

Which leaves the reader with everything needed to take a position. The last lesson assembles the whole argument into one place, shows that every view is a choice about which premise to deny, and sets out how to defend the choice you make.

Every position in this course is a choice about which premise of a single argument to deny, and holding a view responsibly means naming the premise, paying its price and knowing what would change your mind.

This lesson assembles the material. It contains no new arguments, which is deliberate: the work now is to see the structure whole, and to practise the discipline of defending a view rather than merely holding one. If you have arrived here from a search engine, everything you need is stated below, though the reasons behind each premise are in the lessons named beside it.

The master argument

Here is the case against moral responsibility, in five premises, each of which has been examined in this course.

One. If determinism is true, then nobody is able to do otherwise than they actually do. This is the conclusion of the Consequence Argument, and it rests on the fixity of the past, the fixity of the laws, and a transfer principle.

Two. If nobody is able to do otherwise, nobody is morally responsible for what they do. This is the principle of alternate possibilities, the target of Frankfurt's counterexample.

Three. An agent whose character is wholly the product of factors outside their control is not the ultimate source of what they do. This is what the manipulation cases and Galen Strawson's regress are designed to establish.

Four. Moral responsibility in the basic desert sense requires being the ultimate source of what one does. This is the premise that fixes how demanding the notion of responsibility is.

Five. Determinism, or something near enough to it, is true of us.

From one, two and five, nobody is responsible. From three, four and five, nobody is responsible. The argument is over-engineered on purpose: it runs by two independent routes, one through leeway and one through sourcehood, which is why blocking a single premise is rarely enough.

Where each position stands

Now place the views on it.

A classical compatibilist denies premise one, holding that the ability the argument removes is not the ability the phrase "could have done otherwise" picks out in ordinary use. This is the Humean line, and it must survive the collapse of the conditional analysis, so a modern version has to deny premise one via something like Lewis's local-miracle account rather than by a simple conditional.

A semi-compatibilist grants premise one and denies premise two. Determinism does remove leeway, and responsibility never needed leeway; what it needs is guidance control, a reasons-responsive mechanism the agent owns. This is Fischer and Ravizza's position, and it must answer the manipulation argument, which attacks it through premises three and four rather than through the leeway route it has conceded.

A libertarian grants premises one to four and denies premise five. The cost is the whole of the ninth lesson: locating indeterminism where it can do the work, and answering the rollback argument's charge that undetermined events are lucky rather than controlled.

A hard incompatibilist grants all five and accepts the conclusion, then argues that what remains is enough to live on: attributability, answerability, moral protest, and a public-health approach to dangerous conduct in place of deserved suffering.

A revisionist, in Manuel Vargas's sense, splits the question. Our inherited concept of responsibility may well include the demanding requirement in premise four, in which case the sceptic is right about the concept we have; but a successor concept, justified by what our practices actually accomplish in cultivating agency, is compatibilist, and adopting it is a reform rather than a discovery. This is the most self-aware option and it has the corresponding drawback of being a proposal rather than a claim about how things are.

Example. Someone says: "Determinism is probably true, people never could have done otherwise, and none of that stops a manipulative liar from being blameworthy, because his lying flows from a mind that tracks reasons perfectly well." Which premise are they denying, and what must they defend?

Premise two. They have conceded premise one, and their claim is that responsibility survives its loss, which makes them a semi-compatibilist. What they must defend is the sufficiency of the reasons-responsiveness criterion, and the attack will come from the manipulation cases: a liar whose reasons-responsive psychology was installed by a neuroscientist, or by a designed zygote, satisfies their criterion exactly. So their real work is at premises three and four, either by finding a historical condition the manipulated agent fails, which is the soft line, or by accepting that the manipulated liar is blameworthy too, which is the hard line. Note that Frankfurt's counterexample supports them on premise two but does nothing for them here, which is a common way of mistaking the state of one's own position.

Now you. Diagnose this position: "The brain experiments show that decisions are made before we are aware of them, so free will is an illusion." Which premise is it attacking, and is it succeeding?

Four questions that fix a position

If you want to locate yourself, answer these in order and write the answers down.

First: does responsibility require the ability to do otherwise? If yes, you are a leeway theorist and the Consequence Argument is your central problem. If no, you are a source theorist and the manipulation argument is.

Second: does being the source of an action require being the source of your own character? If yes, you have accepted premise four in its strong form and you are heading for scepticism or for libertarianism. If no, you owe an account of why a reasons-responsive mechanism you did not build is enough.

Third: is determinism true? Answer honestly, which for almost everyone means "unknown, and the interpretations of quantum mechanics disagree". Then check whether your position depends on the answer. If it does, you have taken an empirical risk, and you should say so.

Fourth: what happens to punishment if you are right? A position that leaves the criminal law exactly as it is, or that abolishes it entirely, is probably not being thought through, and the thirteenth lesson gives the middle ground both ways.

Example. Run the four questions on a reader who answers: no, no, unknown, and "sentences should be shorter and based on risk".

The first two answers make them a source theorist who denies that sourcehood requires self-creation, which places them as a compatibilist of the reasons-responsive family, denying premise two and premise four. The third answer is the right one and is consistent with their position, since nothing in it depends on how physics turns out, and that independence is a genuine strength worth claiming out loud. The fourth answer is where the tension lies: a compatibilist who keeps desert has grounds for sentences that track the gravity of the past act, so a purely risk-based policy sits oddly with the metaphysics they have just endorsed. They should either supply a separate argument for the policy, which is easy enough since desert sets a ceiling rather than a target, or notice that their practical convictions are running ahead of their theory. Finding that kind of mismatch is the main use of the exercise.

Now you. A friend says: "I'm a compatibilist. Free will obviously exists, because otherwise nothing would make sense." What is missing?

Stating a position properly

A defensible position has four parts, and stating fewer is what makes philosophical argument circular.

The thesis, in one sentence, in the vocabulary of the master argument rather than in slogans. "Free will is real" is not a thesis; "responsibility requires guidance control and not leeway, and guidance control is compatible with determinism" is.

The premise denied, named explicitly. This is what stops two people arguing past each other for an hour before discovering they disagree about premise two.

The cost accepted. Every position has one. The semi-compatibilist accepts that nobody could ever have done otherwise. The libertarian accepts that some decisions have no complete explanation. The hard incompatibilist accepts that no murderer has ever deserved anything. The compatibilist accepts that a person's whole character may have been installed by their upbringing without that touching their responsibility. Stating your cost out loud is the single best test of whether you understand your own view.

The defeater: what would make you change your mind. For a libertarian, a demonstration that decision-making is effectively deterministic at the neural scale, or a convincing account of why the rollback distribution is not luck, would do it in opposite directions. For a compatibilist, a demonstration that our excusing practices do in fact track causal history rather than capacity. For a sceptic, a successful principled distinction between the manipulated agent and the ordinary one. If you cannot name a defeater, you are not holding a position, you are expressing a temperament.

Answering the best objection

The final skill is answering the strongest objection rather than the most convenient one. Each view has a standard best objection, and the standard best reply.

Against compatibilism: the manipulation argument, that an agent designed by another to have exactly this psychology meets every condition you name and is plainly not responsible. The best reply is the soft line with a historical condition, and it must survive the junction between programming and ordinary upbringing, which is where it is weakest.

Against libertarianism: the rollback argument, that if the world were replayed the outcome would vary with nothing about the agent to explain the variation. The best reply is Kane's, that both branches are willed and endorsed by the agent, so responsibility does not need an explanation of the difference.

Against hard incompatibilism: that its conclusion cannot be lived, since resentment and gratitude are not optional for creatures like us, and a theory that no one can act on is idle. The best reply is Pereboom's, that only the desert-presupposing attitudes need to go, that relationships survive on the rest, and that partial versions of this transition happen whenever anyone forgives someone by understanding them.

Against revisionism: that it changes the subject, answering a question about what responsibility is with a proposal about what it would be useful to call responsibility. The best reply is that conceptual reform is ordinary and often correct, and that the question of what we should do with our practices is the one that matters.

Example. Steel-man the objection to your own view, using a compatibilist as the subject, and then answer it in one paragraph.

The objection: the compatibilist's conditions are all satisfiable by an agent whose entire psychology was built to specification by someone else, and in such a case blame is obviously misplaced; since the ordinary determined agent differs only in lacking a designer, and a designer's intentions are facts about the designer rather than about the agent, blame is misplaced in the ordinary case too. The answer: the argument's force comes from a verdict about an artificial case, and the experimental literature shows such verdicts moving from 14 percent agreement to 72 percent on the same question with framing alone; the ownership condition marks a real difference, since an agent who has come to see themselves as an agent through their own history of acting and being held to account has a relation to their mechanism that a freshly programmed agent lacks; and if that reply fails, the hard line remains available, at the cost of a verdict that offends intuition in a case nobody will ever face. That paragraph is not a refutation, and it should not pretend to be. It is a defence with its costs visible, which is what a defence in this subject looks like.

Now you. Take the position you actually hold and write its four parts: thesis, premise denied, cost accepted, defeater. Then write the strongest objection to it and your best reply.

What this subject leaves you with

Three things worth keeping, beyond a position.

The first is the habit of asking what a claim requires. Almost every popular argument in this area, in both directions, fails because it never states its premises: the puppet picture, the neuroscience headline, the appeal to how choosing feels from the inside, the claim that scepticism would destroy society. Each dissolves on being written out.

The second is a calibrated sense of what is known. Determinism is unsettled by physics and probably will remain so. The neuroscience shows something real and narrow. The behavioural genetics is solid and widely misread. The philosophical arguments are unresolved, and their being unresolved after sixty years of concentrated attention by very able people is itself evidence about how hard the problem is, rather than a sign that everyone has missed something obvious.

The third is that the practical stakes do not wait for the metaphysics. Sentencing policy, the insanity defence, how we treat addiction, how much we blame people for the character they were handed: these are being decided continuously, mostly by people who have never heard of the Consequence Argument, and often on the basis of intuitions this course has shown to be unstable. Arguing carefully about them is the applied form of everything here, and it is available to you now whichever premise you decided to deny.