The Stochastic Accumulator Model of the Readiness Potential

Core idea of the model

Schurger, Sitt and Dehaene model the decision of when to move — in Libet-style "move whenever you feel like it" tasks — as a threshold-crossing process in a leaky accumulator. Random subthreshold fluctuations in ongoing neural activity are integrated until they cross a decision threshold, at which point a movement is triggered.

Because trials are time-locked to movement onset, averaging across many threshold-crossing trajectories produces an apparent slow buildup — the RP — even though, in single trials, there is no stereotyped slow ramp starting seconds earlier. The classical RP is, on this account, a statistical artifact of averaging stochastic trajectories that happen to have ended in a crossing.

The readiness potential, reinterpreted

Classically, the RP was taken as evidence that the brain begins preparing a voluntary action several hundred milliseconds before the subject becomes consciously aware of deciding. Schurger et al. argue instead that the RP mainly reflects the statistical structure of ongoing neural noise plus time-locking — not a deterministic "unconscious decision" beginning at RP onset.

In the accumulator model, the neural decision to move now is identified with the threshold crossing, which occurs close to movement onset (within roughly 150 ms). Earlier premovement activity reflects non-committal fluctuations biased weakly toward threshold by the instruction to act — not a fixed unconscious decision.

The shift is subtle but consequential. The RP is real; it is observed; the accumulator model fits its empirical shape. What changes is the interpretation of what the RP measures — from a deterministic premovement command to the statistical signature of noise being integrated until it tips over the edge.

The task context: self-initiated movements under weak imperative

The model is tailored to paradigms where subjects are instructed to make spontaneous movements after an unspecified delay, without external cues or strategies like counting. Under these instructions, the "imperative" to move is weak and nonspecific, so the exact timing is largely set by random fluctuations rather than by a precise internal clock or stimulus.

The authors replicate Libet-style behavioral and EEG findings using this framework, fitting RP shape with parameters obtained from behavioral waiting-time distributions — not from the EEG signal itself. That methodological choice matters: the model's predictions about the RP are derived from behavior alone, and the EEG agreement is therefore an independent confirmation rather than a circular fit.

Leaky stochastic accumulator: how it works

The stochastic-decision model uses a leaky accumulator whose input is intrinsic neural noise plus a small constant bias reflecting the instruction to move. The accumulator's state drifts stochastically upward and downward, but because it is biased and leaky, it tends to hover below threshold until a fluctuation pushes it across.

Once the state hits threshold, a commitment to act is made and movement is initiated — analogous to evidence-accumulation models in perceptual decision-making (Ratcliff, Gold & Shadlen) that trigger a choice when a criterion is met. The mathematical machinery is borrowed wholesale from the perceptual-decision literature; what changes is the source of the "evidence" — not external sensory input but internal spontaneous activity.

Unification with perceptual decision-making

A key theoretical claim is unification: the same accumulator-with-threshold mechanism that underlies stimulus-driven perceptual decisions can also account for self-initiated voluntary actions, with "evidence" replaced by intrinsic activity fluctuations. This situates voluntary action within the broader decision-making framework, rather than requiring a special mechanism unique to free actions.

The authors note existing work suggesting common neural mechanisms for voluntary and stimulus-driven actions, supporting the idea that the accumulator architecture is reused across domains. The implication: "self-initiated" and "externally-cued" actions are not two different types of brain activity but one underlying process running on different sources of input.

Experimental tests and predictions

The authors validate the model in two ways.

• First, they match simulated RPs to empirical EEG RPs using parameters fit to waiting-time behavior, not to the EEG itself — an independent prediction that holds.
• Second, they test a stronger prediction by cuing speeded responses at random times and sorting trials by reaction time: shorter reaction times correspond to trials where the ongoing fluctuations were already closer to threshold, consistent with the accumulator account.

The "interruption" logic leverages the assumption that forced and spontaneous responses tap the same underlying decision variable; when you interrupt at a random moment, you just sample the current state of the accumulator. The data fit the prediction.

Philosophical and free-will implications

In the free-will debate, the model undercuts a simple reading of Libet that equates RP onset with the onset of an unconscious decision fixing the movement before conscious intention. Instead, the decision is tied to a late threshold crossing, with earlier activity reflecting stochastic dynamics that only probabilistically influence when the decision happens.

This supports interpretations in which conscious intention may still be closely time-locked to the actual commitment to act, and in which the RP is not straightforward evidence that "the brain decides before you do." The reading-of-record for thirty years — that Libet had falsified free will — has to be retired, or at least seriously qualified, in light of the accumulator model.

None of this proves free will. What it does is restore the question: the gap between RP onset and reported W-time, taken by many as proof that consciousness arrives too late to matter, is on Schurger's account the gap between random fluctuation and threshold-crossing decision — with the decision itself far closer in time to the conscious intention than the classical Libet reading suggested.

Why this matters for the trilogy

The trilogy's engagement with Libet is more honest with Schurger in the picture than without him. The Anima and Numen conversations about Sapolsky, the readiness potential, and the 300-millisecond gap take the empirical phenomenon seriously but refuse the deterministic interpretation that had become orthodox. Schurger explains why that refusal is now methodologically respectable rather than merely metaphysical preference: the empirical evidence allegedly closing the question never actually closed it.

The receiver-model reading the trilogy offers is independently motivated by anomalous neurology, by Bell's nonlocality, and by the Planck-scale renderer framework. Schurger is not required to make the receiver model coherent. But Schurger removes the single most-cited empirical objection to it. Once the RP is no longer evidence that an unconscious neural commitment occurs hundreds of milliseconds before awareness, the architecture Anima describes — the field rendering local agency through biological tissue with a non-zero latency — becomes consistent with the same data the productionist had used against it.

Senna Park's compression thesis in Anima reads more cleanly through Schurger than through Libet alone: the 40-bit filter is the architecture in which intrinsic neural fluctuations integrate into a coherent moment of choice. Lucía Reyes's cymatic 300-millisecond pre-event window in Numen — geometric patterns forming in water before the chord is played — sits in a complementary symmetry: at one end, internal noise integrating toward a crossing that becomes the felt "now"; at the other, the field's effects appearing in the medium before the local causal trigger. Both gaps are now empirical. Neither demands that consciousness be late, illusory, or productionist.

For the original Libet experiment, see the Libet explainer. For the philosophical framing that organizes both papers, see the Chalmers explainer. For the synthesis that weaves all of this together, see What the Evidence Shows So Far. The Schurger paper itself is open-access at PNAS.