Evidence refuting free will is being misinterpreted, scientists argue

Do you believe in free will? Some scholars do not — and they rely on evidence from the brain sciences to make their case. Some people find the dismissal of the idea that we are in control of our decisions and actions to be deeply disturbing. We, as professionals active in the field, know they do because we regularly receive their e-mails asking — often in desperation — about neuroscientific studies that seem to threaten the possibility of free will. Most of these assertions rest on scientists claiming to anticipate or predict choices based on brain activity observed before a person in an experiment is even aware of what their own choice will be. Free will naysayers contend that unconscious brain processes may initiate an action that a person then erroneously believes to be set in motion by their own volition.

But what if the results of that research were misconstrued, with the devil lurking in the fine details that most people do not read or do not understand?

Neuroscience research going back to the early 1980s claimed to demonstrate that conscious free will is an illusion (“conscious free will” refers to our conscious decisions determining our actions). These results accumulated like nails in the coffin of free will, offered up by neuroscientists and hammered in by the mainstream media, until, in 2016, the Atlantic declared, “There’s no such thing as free will.”

Not so fast. More recent studies, combining empirical data and computational modeling, suggest this prior research had been misinterpreted, and none of it bears on conscious free will one way or another. Neuroscience, we conclude, has not disproven conscious free will.

Many cognitive neuroscientists in the field, including former “no-free-will” proponents, now acknowledge that the supposed neuroscientific evidence against it is dubious. Unfortunately, the public still hasn’t heard the news, and the idea that neuroscience has disproven conscious free will, or even free will in general, still hangs in the air.

Once the sole purview of philosophers, free will and consciousness have been increasingly studied by neuroscientists. These topics differ from other areas of study in neuroscience in that they matter deeply to most, if not all, of humanity. In contrast, few would lose sleep over the relative importance of other human attributes, such as whether people can directly sense magnetic fields (magnetoreception).

Science often moves forward by posing hypotheses that are later modified or rejected. Given the deep existential nature of research on volition, however, we face two very important questions: Where should we set the bar for evidence claiming to bear on free will? And how should we evaluate and interpret such evidence to know if or when it has been met?

Recognizing what philosophers of science call “inductive risk,” or the costs of potential errors, we should set the bar high. The cost of mistakenly denying free will is considerable, as those troubled letters we received show. And there is good reason to doubt the evidence often cited. The neuroscience of volition typically focuses on immediate (or proximal) and meaningless decisions, (like “press the button from time to time, whenever you feel like it, for no reason at all”). The decisions we care about with respect to free will and responsibility, however, are ones that are meaningful and often have longer time horizons. Perhaps many, or even most, of our day-to-day decisions — choosing when to take the next sip from your water cup or which foot to put forward — are not acts of conscious free will. But maybe some decisions are. Fortunately, or unfortunately, those consequential ones are the most difficult ones to study.

What would it take for neuroscience to disprove conscious free will? The evidence must clearly show that people settle on a decision unawares. Here the devil is indeed in the details of predicting behavior and inferring consciousness from brain activity. For example, using machine learning to “predict” behavior in advance of the conscious decision will not necessarily tell us much. Consider a simple free choice of pressing a button with your right hand or your left hand, where predictions that are about 60 percent correct might be statistically significant (compared with a coin toss of roughly 50 percent); such predictive power would not undermine conscious free will.

Why not? Because a 60 percent accurate prediction might just pick up on a tendency toward one alternative or the other rather than a firm decision. Moreover, many of us have enduring preferences and character traits that affect some decisions, and it would be surprising if such choices were not at least somewhat predictable in advance based on brain activity. In addition, because consciousness and decision-making play out over time and rely on past experiences, prediction need not indicate determination. Thus, in such cases, the details of performance of the machine-learning classifier do matter, not just whether it is “significantly above chance.” In fact, anything less than close-to-perfect predictive accuracy may be equivocal.

In addition, neuroscience results depend on their data-analysis method, which can mislead. For example, some digital data filters can, in effect, “leak” future information into the past, and analyses involving a sliding window can inadvertently allow the system’s data analysis to “peek” into the very future that it is trying to predict. The devil, again, is in the details.

These considerations matter because new scientific data on free will are on the horizon, mainly because of the proliferation of invasive recordings from surgically implanted brain electrodes in humans. An informed reader needs to know what evidence would truly falsify conscious free will and what would not.

To be clear, we are not arguing for or against the existence of conscious free will; we are talking about the data here and the way to know whether those data constitute evidence that undermines conscious free will. We must ensure that the paradigms that we investigate in neuroscience allow us to draw conclusions about the actions that pertain to conscious free will. For many behaviors, being predictable to some degree should not surprise us: Does it undermine your free will if we predict that you will brush your teeth before going to bed tonight?

The neuroscientist Robert Sapolsky has taken a different approach. He discounts the brain data and instead focuses on statistical regularities — for example, that early-childhood adversity can negatively impact the kind of choices we make and outcomes we experience later in life. He argues in his book Determined that we are part of a deterministic world over which we have no influence and that statistics like the childhood adversity findings bear this out. We do not deny the reality of regularities; our actions today may indeed be constrained (or partly determined) by our past environment or experiences. But just how much constraint is enough to rob us of free will? The lack of very high predictability in those statistics leaves plenty of room for acts of conscious free will (again, it would be strange if your early life experiences had no effect whatsoever on your later life).

Finally, we note that a single human brain is arguably far more complex than the entire Earth’s atmosphere, and we can’t even predict the weather more than a few days into the future. So throwing sophisticated AI at brain data is unlikely to enable us to predict future brain states based on past ones, at least any time soon. We leave open the possibility that we will get there one day (though you are free to disagree). But one thing is clear: we are not there yet.

This is an opinion and analysis article, and the views expressed by the authors are not necessarily those of Scientific American.