With real decisions the signal is missing altogether

In 2019 Uri Maoz and colleagues put their participants in front of a decision that actually cost something. Two non-profit organisations were on offer; the one chosen received a donation of $1,000. In a second condition nothing was at stake — an arbitrary, inconsequential choice of the kind that has been standard in this field for decades.

For the arbitrary decisions, the EEG showed the expected signal: the readiness potential, that slow build-up in the brain trace on which the entire free will debate rests. For the deliberate decisions it was absent:

The results section puts it more sharply still: for the deliberate decisions the amplitude of the signal did not differ significantly from zero.

That is an uncomfortable result, because the field’s famous conclusion tacitly presupposes that a signal measured in the laboratory says something about decisions in general. Precisely that, it seems, had never been tested.

What Libet measured in 1983

The setup around which everything turns is disarmingly simple. A participant is to move a wrist or a finger spontaneously at some point — for no reason and with no consequences. On a rapidly running clock she notes the point at which she first felt the urge to move; Benjamin Libet called this moment W. Electrodes meanwhile register the readiness potential. Hans Helmut Kornhuber and Lüder Deecke had described it in 1965, when they compared changes in brain potentials during voluntary and during passive human movements.

Libet, Curtis Gleason, Elwood Wright and Dennis Pearl reported in 1983 that cerebral activity had preceded the reported conscious intention by at least several hundred milliseconds. The numbers at issue: for movements the participants classed as spontaneous and unplanned, the readiness potential set in on average around 550 ms before the movement; where the movement had been premeditated, onset lay about a second before it. W itself lay at around −200 ms, that is, shortly before the movement. The famous gap is the difference between those two figures: a good third of a second in which the brain is already charging and the person claims to have noticed nothing. The cultural translation was spectacular and is still in circulation: the brain had decided before “we” decide; free will is an illusion.

Between the measurement and that statement, however, lie several assumptions, each of which has since been tested. Three of them have not held.

The clock takes part in the measurement

The first concerns W. The moment of inner intention is reported in retrospect, and this form of introspection is not a direct view into the decision. Attention, memory, and the question of what counts as an “intention” in the first place can all shift the value.

Worse still: the clock is not a neutral instrument. Jeff Miller, Peter Shepherdson and Judy Trevena recorded EEG while participants made spontaneous key presses — once with the usual clock monitoring, once with no clock and no timing judgement at all. The averaged brain activity before the key press differed markedly between the two conditions. Smaller clock-related differences appeared even before mere tones in a task involving no movement whatsoever. The apparatus meant to date consciousness thus alters the very signal that is to be dated.

The second assumption is that the readiness potential belongs to movement preparation. Trevena and Miller tested this directly by comparing the electrophysiological signs before a decision for a movement with those before a decision against one. No difference appeared. Whatever rises before the decision is therefore not movement-specific — and makes poor evidence that the movement had already been settled.

The third assumption is the hardest: if the signal were the cause of the conscious urge, it would have to shift along with the urge’s timing. Patrick Haggard and Martin Eimer tested this. First they compared a fixed condition, in which the same finger was always moved, with a free one, in which participants chose between left and right index finger on each trial. Neither the movement-related potentials nor the reported times differed — the added freedom of choice left no trace in the signal. They then compared trials with early and late awareness. The readiness potential occurred later, if anything, when awareness came early — which rules it out as the cause. Only the lateralised readiness potential, which reflects the choice of side, behaved as expected. The authors concluded that at most the processes underlying it can be candidates for the experience of initiating a movement.

Where the ramp comes from

That leaves the question of why the readiness potential looks the way it does at all. Aaron Schurger, Jacobo Sitt and Stanislas Dehaene proposed an accumulator model in 2012. In it, neural activity fluctuates continuously in any case; the moment at which a movement threshold is crossed depends substantially on where those subthreshold fluctuations happened to stand.

The decisive step is a statistical one. Because the data are averaged backwards from movement onset, the trials that systematically stack on top of one another are those in which the fluctuation happened to be rising. On average this produces a gradually ascending ramp — even when nothing ramp-shaped occurred in any single trial. The supposed harbinger of the decision would on this account be in good part an artefact of the analysis.

What is remarkable about this criticism is what it does not claim. None of the work cited disputes Libet’s measurements. The dispute is about interpretation throughout: whether an averaged signal denotes an event that took place in a single trial at all, and whether that event is the decision. A finding can be correct and still show something other than what four decades of reception ascribe to it.

In 2021 Schurger, together with Pengbo Hu, Joanna Pak and Adina Roskies, drew up a balance sheet. Their verdict: the advances in understanding the readiness potential, especially the computational models, call for a reassessment of its significance for free will — not a confirmation of it.

The field’s second great headline fares similarly. Chun Siong Soon, Marcel Brass, Hans-Jochen Heinze and John-Dylan Haynes worked with functional imaging in 2008. They found that the outcome of a decision is discernible in prefrontal and parietal areas up to 10 seconds before it enters awareness. In the reporting, ten seconds became predetermination. What was passed over was the accuracy rate: it stood at around 60 %, above chance, therefore, but far removed from a prediction that fixes the outcome.

Freedom is not a timestamp

Philosophically, everything hangs on what freedom is supposed to mean. Anyone who takes it to be an uncaused decision outside nature will not find it with an EEG — nor with any other instrument. Compatibilist positions ask a different question: whether a person can act for reasons of their own, weigh alternatives, and control their conduct. These are capacities that unfold over seconds, years and a life history, not within a window of 200 ms.

Libet himself held on to a remainder: after the intention becomes conscious, there is still time for a veto, a “free won’t” — by his reckoning about 200 ms before the movement begins. That too is not experimentally secured. Another point matters more. That preparatory activity precedes a movement shows that actions have prehistories. It does not show that reasons, self-control or responsibility are meaningless.

What remains

The 1983 result stands — and it says less than four decades of reception claim. What was measured was the dating of an urge to make an inconsequential finger movement, obtained with a procedure that itself influences the measured signal. What became of it was a statement about free will in general. The leap between the two is a problem of ecological validity, and Maoz’s finding makes it visible: where a decision has reasons and carries consequences, the very signal on which the argument rests disappears.

The pattern can be recognised by a recurring shape. A measurement is obtained under laboratory conditions that leave out precisely the everyday version of the thing — here: reasons, consequences, time — and the conclusion is then carried back to everyday life. The experiment remains great all the same. It was what first made a philosophical claim measurable, and so created the conditions under which its own interpretation could be corrected. Its result is not a death sentence for free will but a warning: anyone who wants to answer a question of millennia in milliseconds must first say which kind of freedom their clock can measure at all.