For decades, the Libet experiment has been interpreted as evidence that free will is an illusion: the brain’s “readiness potential” — a slow buildup of neural activity in motor areas — begins roughly a full second before a spontaneous movement, while subjects report conscious awareness of the urge to move only about 200 milliseconds before the action. This has been taken to mean the brain decides before “you” do, and conscious intention is a post-hoc confabulation.
Aaron Schurger, a professor of neuroscience, challenges this interpretation. He argues the readiness potential does not reflect an unconscious decision or commitment to move. Instead, it reflects autocorrelated stochastic neural noise that happens to drift toward a threshold, and the threshold crossing — the actual decision — occurs only about 150 milliseconds before movement, closely aligning with when subjects report conscious awareness.
This reframing dissolves what Schurger calls “Libet’s paradox” (how can the brain be deciding a full second earlier than the conscious urge?) and undermines the claim that neuroscience has disproved conscious free will — while also not proving free will exists.
The Readiness Potential and Why the Standard Interpretation Is Flawed
The readiness potential is a slow, ramp-like increase in electrical activity over motor cortical areas (especially supplementary motor area) that precedes spontaneous, self-initiated movements by up to a second or more. It is reliably observed in EEG and in single-neuron recordings across species, including humans, monkeys, rats, and even crayfish.
The standard interpretation holds that this ramp reflects the brain unconsciously committing to a movement and then preparing to execute it — meaning the conscious feeling of “deciding” comes too late to be causally relevant.
Schurger’s core objection is methodological: in Libet-style experiments, researchers analyze brain data only from trials that ended in a movement. They never examine what the brain was doing during moments when no movement occurred. This is a form of selection bias.
Flu analogy: If you give people health monitors during flu season and only analyze data from those who got sick, you might find their health was declining before they even encountered the virus. This doesn’t mean the immune system “knew” the virus was coming — it means getting the flu necessarily selects for people whose health was already somewhat compromised. Similarly, selecting only movement trials guarantees you’ll see a buildup beforehand, even if that buildup is just noise drifting toward a threshold.
Barometric pressure analogy: Rain is reliably preceded by dropping barometric pressure, but pressure drops all the time without rain. A reliable antecedent is not the same as a cause or predictor — you need to look at all the data, not just the cases where the event occurred.
Autocorrelated Neural Noise and Threshold Crossing
Neural noise — random fluctuations in neural activity — is ubiquitous in the brain, from single neurons to large-scale EEG signals. Critically, this noise is autocorrelated: its value at any moment is statistically dependent on its value at the prior moment, meaning it drifts slowly rather than jumping randomly.
White noise vs. autocorrelated noise: White noise is completely independent from one instant to the next — it can jump anywhere. Autocorrelated noise has the property that large deviations tend to be slow, and small fluctuations tend to be fast. Visually, it looks like smooth, wandering drifts rather than jagged spikes.
This is not a subtle statistical detail — it is a fundamental property of brain signals. If you see pure white noise in EEG, something is wrong with the recording.
Schurger’s model proposes that the motor system has a threshold for triggering movement. When neural activity crosses this threshold, the movement is initiated. Because the underlying noise is autocorrelated, the approach to threshold is necessarily slow and ramp-like — exactly the shape of the readiness potential.
He demonstrated this with simulations: generate autocorrelated noise, set a threshold, identify every moment the noise crosses it, align all those crossings, and average. The result closely matches the empirically observed readiness potential waveform.
This means the readiness potential is not a purposeful “preparation” signal — it is the statistical artifact of looking backward at noisy activity that happened to reach a threshold.
The Role of the “Imperative” and Why Actions Aren’t Purely Random
A common objection: if movements are triggered by noise, doesn’t that make all behavior random?
Schurger’s model includes two components: (1) a gentle imperative to move sometime soon (reflecting the “demand characteristics” of the experiment — subjects know they’re supposed to move, and typically do so within about 7 seconds), and (2) stochastic noise that determines the precise moment of threshold crossing.
The imperative is weak enough that noise dominates the exact timing. This is why movements feel spontaneous and unplanned — because the precise moment is not determined by a deliberate decision.
In everyday life, most actions are driven by external stimuli or well-learned routines where noise plays little role. Noise is most relevant in unconstrained, spontaneous contexts — like Libet’s task, or a child deciding exactly when to swing a bat in T-ball.
Baseball analogy: When a ball is pitched at high speed, the decision to swing must be driven by sensory input — noise would be detrimental. But in T-ball, where the ball is stationary and no one dictates when to swing, internal noise could plausibly determine the precise timing. Everyday life spans a continuum between these extremes.
Resolving the Timing: The 150-Millisecond Decision
Schurger places the actual decision — the threshold crossing that triggers movement — at roughly 150 milliseconds before movement onset. This is close to when Libet’s subjects reported the conscious urge to move (around 150–200 ms before movement).
In this view, subjects were not mistaken or confabulating. Their report of when they felt the urge was approximately correct. Libet’s error was dismissing his subjects’ reports because the readiness potential seemed to start much earlier — but that earlier buildup was noise, not decision.
Schurger also challenges Libet’s “veto” hypothesis — the idea that while we don’t initiate actions consciously, we can consciously veto them in the last 200 milliseconds.
Newer evidence shows that roughly 200 milliseconds before movement is actually the point of no return: after this point, the movement cannot be inhibited, even if the subject tries. Experiments where subjects are interrupted with a “stop” signal at random times show that interruptions within the last 200 ms fail to prevent the movement.
This directly contradicts Libet’s claim that the last 200 ms is a window for conscious veto — it is in fact the opposite.
The Linear Ballistic Accumulator and Why It Doesn’t Solve the Problem
Some have proposed the linear ballistic accumulator (LBA) model as an alternative explanation for the readiness potential that doesn’t involve stochastic noise. Schurger disagrees.
The LBA pushes all randomness to the starting point of each trial — where you begin relative to the threshold is random, but after that, the accumulation is deterministic and linear.
Schurger considers this a “toy model” that is functionally equivalent to his account but less biologically plausible. It doesn’t eliminate randomness; it just relocates it. And it requires the brain to randomly set a starting point at the beginning of each trial, which is an odd mechanism.
What Would It Take to Disprove Free Will?
Schurger is careful to say his model does not prove free will exists — it merely removes what he considers the strongest neuroscientific argument against it (the Libet interpretation).
He has written about what evidence would genuinely threaten free will: the ability to predict a person’s actions several seconds in advance with near-100% accuracy, not just for routine behaviors (eating food when hungry) but for non-trivial, unconstrained choices.
No current neuroscience comes close to this. Until it does, the question of free will remains open.
Consciousness and Action Initiation
Schurger sees a deep connection between consciousness and free will, rooted in the role of consciousness in initiating new actions.
Working with a blindsight patient (GY), he observed that the patient could guess correctly about stimuli in their blind visual field — but only when cued (e.g., by a beep) about when to guess. Without a conscious cue, the patient would not respond at all, even though the visual information was being processed unconsciously.
This led Schurger to propose that consciousness is logically necessary for initiating a new movement. You can guide an ongoing action with unconscious information (adjusting your grip on a cup), but you cannot trigger a new action without being conscious of the stimulus that prompted it.
This is a claim about logical necessity, not temporal precedence — consciousness doesn’t have to come “before” in time, but when you examine the full episode, the reason for the action must be something the person was conscious of. You never find cases where a new movement is triggered by a stimulus that was processed entirely unconsciously.
This is a falsifiable claim: finding a single clear counterexample would disprove it.
Consciousness Theories: A Hybrid Approach and Attention Schema Theory
For the mechanics of how consciousness works in the brain, Schurger favors a hybrid of three frameworks:
Global workspace theory: Consciousness arises when information is broadcast widely across brain networks.
Higher-order thought theories: Consciousness involves meta-representations — the brain representing its own states.
Integrated information theory (IIT): Consciousness relates to the degree of integrated information in a system.
Each captures something necessary, but none is sufficient alone.
For the hard problem of consciousness — why there is “something it is like” to have subjective experience — Schurger favors attention schema theory by Michael Graziano.
The theory proposes that consciousness is a simplified internal model the brain constructs of its own attention. Just as the brain constructs a body schema (a model of the body’s position and state), it constructs an attention schema (a model of what it is attending to).
This schema is not an accurate description of attention’s neural mechanisms — it is a simplified, schematic representation. And it is this schematic model that we experience as “consciousness.”
The theory’s key insight: it explains why we are so convinced there is something more in the brain than mere matter — why the hard problem feels so intractable. The brain’s model of its own attention is constructed in such a way that it seems to have an ineffable, non-physical quality, even though it is entirely a product of neural processes.
Schurger likens this to a neurological patient who insists there is a squirrel inside their skull. The rational response is not to search for the squirrel, but to explain why the patient is so convinced one is there. Similarly, the hard problem may be dissolved by explaining why we are so convinced subjective experience is mysterious — not by solving the mystery directly.
The theory predicts that people will find it unsatisfying — and that this dissatisfaction is itself a consequence of how the brain models its own attention.
Current Research: Consciousness Thresholds and Surprise
Schurger is currently testing whether the threshold for consciousness is higher and steeper than the threshold for performing a task — a long-debated question.
In visual discrimination tasks, subjects can perform above chance (e.g., 75% correct) while reporting they felt they were guessing and were not conscious of the stimuli. Are they genuinely unconscious of the stimuli, or just reluctant to report?
Skin conductance experiment: Subjects perform blocks of trials at varying difficulty levels. At the end of each block, they see their score. Skin conductance (a measure of autonomic surprise) is recorded.
Prediction: If subjects in the middle range (65–85% correct) were genuinely unconscious of the stimuli, they should be surprised by their own scores — because getting that many right by pure guessing is very unlikely. If they were conscious, they should not be surprised.
This tests whether there is a dissociation between the threshold for performance and the threshold for conscious awareness — with potential implications for understanding blindsight and the neural basis of consciousness.
Neural Stability as a Hallmark of Conscious Processing
Schurger’s earlier work (published 2015) proposed that transient stability in neural population activity — neurons firing in stable patterns for 100–200 milliseconds — is a necessary condition for conscious experience.
Unstable, rapidly changing activity does not support conscious perception; stable activity does.
A 2024 paper by Eisen in Earl Miller’s lab provided supporting evidence: neural activity was stable during conscious states and unstable under anesthesia.
Schurger is returning to this line of research, exploring how network stability relates to conscious processing and perceptual decision-making.
