TL;DR
Two empirical claims carry most of the popular free-will argument, and both are weaker than advertised. The first is Libet’s readiness potential — brain activity ramping up before a reported decision. Schurger and colleagues modelled that buildup as a leaky stochastic accumulator, in which the moment of movement is largely set by spontaneous subthreshold fluctuations, and the apparent ramp is an artefact of averaging time-locked to movement onset; they then confirmed a specific prediction of the model in a second EEG experiment.1 The second is that undermining belief in free will makes people behave worse. The original result was d = 0.88; a registered replication across five countries (N = 621) found d = 0.076, confidence interval spanning zero.2 A meta-analysis of 145 experiments — 95 of them unpublished — found the manipulations reliably shift belief but produce no detectable downstream consequences.3 Determinism is still hard to argue against. Neither of these is why.
A finger, a domino, a chain. The hand is not standing outside the causal order looking in; it is part of it. Photo: Bradyn Trollip on Unsplash.4
The Argument As It Circulates
The popular version runs in two steps, and it is worth stating cleanly because it is genuinely coherent.
Step one: neuroscience has caught the brain deciding before “you” do. Electrodes pick up a slow rise in activity — the readiness potential — hundreds of milliseconds before the moment a person reports having decided to move. If the brain is already committed before the experience of choosing arrives, then that experience is a narration, not a cause.
Step two: this is dangerous knowledge. If people stop believing they are the authors of their actions, they will stop behaving as though it matters — cheat more, help less, take less responsibility.
Both steps have famous experiments attached. Both experiments have since been examined closely.
What the Readiness Potential Actually Shows
The interpretive move in step one is that the readiness potential is a preparation signal: the brain assembling a decision, which consciousness later takes credit for.
Schurger, Sitt and Dehaene proposed a different account. They modelled the decision of when to move — in a task with no external cue, only a general instruction to move at some point — as a leaky stochastic accumulator. On this model, when the imperative to act is weak, the precise moment the threshold is crossed is largely determined by spontaneous subthreshold fluctuations in neural activity. Crucially, time-locking the average to movement onset makes those random fluctuations appear, in the average, as a gradual exponential-looking rise.1
In other words: the ramp may be an artefact of how the data are aligned. You are averaging over trials selected by the moment of movement, so you inevitably find an upward drift leading into it, whether or not anything was being “prepared.”
This is not hand-waving — the model made a falsifiable prediction and it was tested. If the account is right, fast responses to unpredictable interruptions should also be preceded by a slow negative-going deflection beginning well before the interruption, even when the subject was not preparing to move at that moment. A second EEG experiment confirmed it.1
The honest summary is that the standard reading — the brain decides, then tells you — is no longer the only account of the data, and arguably no longer the best one. That does not prove free will. It removes what was widely treated as the decisive evidence against it.
The Second Claim, and What Happened To It
Step two produced one of psychology’s most quoted results: participants who read a passage arguing that free will is an illusion subsequently cheated more.
Then it was replicated properly.
Many Labs 5 ran a registered replication across five sites in five countries, N = 621. The original effect size was d = 0.88 — large. The replication found d = 0.076 for cheating behaviour, 95% CI [−0.082, 0.22]. The manipulation check was no better: the anti-free-will passage shifted expressed free-will belief by d = 0.064, 95% CI [−0.087, 0.22].2 Both intervals comfortably contain zero.
And then the whole literature was pooled. Genschow and colleagues meta-analysed 145 experiments, 95 of them unpublished — which matters, because it means the file drawer was actually opened rather than assumed empty. Their finding is precise and two-sided: exposure to anti-free-will manipulations does decrease belief in free will and increase belief in determinism. But they could find no evidence of downstream consequences.3
So the manipulations work. The consequences don’t appear.
What Survives
Strip out the two weak pillars and the philosophical question is untouched, because it never depended on them.
If every decision emerges from genes, history, chemistry and circumstance, then “could have done otherwise, under exactly identical conditions” looks very hard to defend. That was true before Libet and remains true after Schurger. Neuroscience did not create the problem and cannot dissolve it.
What it does clarify is which conception of freedom is worth arguing about. The libertarian one — a self standing outside the causal order, initiating — has no mechanism anyone has been able to describe. The compatibilist one is more modest: you act freely when your action follows from your own reasons, values and intentions rather than from coercion or incapacity, whether or not those reasons themselves have prior causes.
The move that makes compatibilism more than a word game is noticing that deliberation is itself part of the causal chain, not an exception to it. Thinking about a decision is a physical process that changes the outcome. Which is why determinism does not imply fatalism: if reasons, arguments, incentives and institutions are causes, then deploying them changes what happens. Fatalism says nothing you do matters. Determinism says what you do is caused and matters — those are compatible, and conflating them is the most common error in the whole conversation.
What Responsibility Is Actually For
The practical stake is what happens to blame.
You can hold the whole causal picture and still want a functioning system of accountability, provided you are clear about what it is doing. Forward-looking justifications survive intact: protecting other people, changing future behaviour, rehabilitation, deterrence, and distinguishing between coercion, incapacity and deliberate action — a distinction that remains perfectly real and perfectly measurable inside a deterministic world.
What does not survive is desert in the strong sense: the idea that a person is the uncaused author of their own character and therefore intrinsically merits suffering. That is the part the causal picture erodes, and it is worth being clear that it is a large part of how punishment is actually justified in public.
This is close to an argument the site has made from the other direction. Treating a bad event as the centre of your identity can make things worse, and explanation can overreach into excuse. The resolution is the same in both cases: a cause is not an alibi. Knowing why someone did something tells you how to change it. It does not tell you to do nothing.
Where I’d Hold This Loosely
Four limits, and the last is the one that matters most.
Schurger’s account is a model, not a demonstration that nothing is being prepared. It fits the data and made a confirmed prediction, and the debate over what the readiness potential represents is live rather than closed. I have claimed only that the standard interpretation is no longer safe, which is a smaller claim than “it has been refuted.”
Second, Many Labs 5 tested one paradigm with one induction text. A single failed replication of one operationalisation is not proof that belief in free will has no behavioural consequences anywhere — though the 145-experiment meta-analysis makes that a much harder position to hold.
Third, compatibilism is a philosophical position, not an empirical finding, and plenty of careful people reject it as changing the subject rather than answering the question. Nothing measured here settles it. What the evidence does is remove two bad reasons for thinking the question was already settled the other way.
Fourth, and cutting against the usual anxiety about writing this at all: the literature claiming it is harmful to tell people free will might be an illusion is precisely the literature that failed to replicate. The worry was that this argument corrodes behaviour. On the best available evidence, 145 experiments deep, it doesn’t appear to do anything. Which means the discussion can be had on its merits — and that this post, on its own evidence, is not the dangerous object the genre keeps warning about.
