Aaron Schurger, Jacobo D. Sitt and Stanislas Dehaene (2012), Proceedings of the National Academy of Sciences 109(42):E2904ff. — DOI-10.1073/pnas.1210467109
Key takeaways
- The readiness potential that precedes self-initiated movements need not be the signature of planning. If the decision to move is a threshold crossing by a leaky accumulator driven mostly by ongoing noise, then averaging brain activity time-locked to the movement recovers a slow, exponential-looking buildup, because only the epochs that ended in a crossing are averaged. The buildup’s causal role is “incidental”.
- The model was fitted to behaviour only (waiting times in Libet’s task, three parameters: urgency 0.11, leak 0.5, threshold at the 80th percentile), and with those parameters it fitted the readiness potential from −3 to −0.15 s (r² = 0.96). Its new prediction was tested and held: when participants were interrupted by a click at random, their fastest responses were preceded by a slow negative deflection beginning before the click, which the planning account does not predict.
- It relocates the decision. The “neural decision to move now” is the threshold crossing, about 150 ms before the button press, which coincides with the lateralised readiness potential and with participants’ reported time of the urge (−152 ms here). So the conclusion that the brain decides to move well before the person is aware of deciding does not follow from the readiness potential.
Standing in the record
Filed on 2026-10-03 at the owner’s request, for the Libet readiness-potential debate and its use against free will. It extends Usher and McClelland (LIT-565): its model is the leaky stochastic accumulator of that paper (their reference 27), taken as a single unit with no competitor and fed with a constant urgency plus noise, and the paper cannot stand without it. Its evidence-accumulation framing is the one the other filings of the batch state: Ratcliff and McKoon (LIT-576) for the diffusion model whose mean–SD property it checks, and Bogacz et al. (LIT-569) for the regimes of the leaky process.
What it bears on in the record:
- Dennett and Kinsbourne (LIT-431). The record’s only other treatment of Libet. They argue that the timing of the reported intention is an artefact of confusing the time a content is represented with the time it represents. Schurger et al. attack a different premise, the timing of the neural decision, and reach a compatible conclusion: the gap between brain and awareness that Libet reported is not evidence that action is initiated unconsciously. The two critiques are independent, and the record now holds both. Schurger et al. set aside the clock method for the urge as “irrelevant to this experiment” (they note it has been criticised), so they do not depend on Dennett and Kinsbourne, nor answer them.
- Free will (THEORY-029, THEORY-040). The record’s free-will theories are about compatibilist conditions on ownership and manipulation. Neither rests on the Libet argument, so this paper neither supports nor tests them. What it removes is an empirical premise sometimes used against conscious will, that a neural decision precedes awareness by half a second or more. It does not show that the conscious urge causes the movement; the model is “silent with respect to the urge to move”.
- Agency. Its picture of agency is specific: a task goal sets the baseline near threshold, and the exact moment of acting is left to spontaneous fluctuation. “The precise moment is not directly decided by a goal-directed operation.”
No instruction for machine-learning practice; nothing here belongs in the anthology.