Last month, a paper was published in Nature, in which Kay et al(1) were able to guess which of their stimuli a person was seeing by looking at their fMRI scans. The model looked something like this (from Kay et al’s Figure 1, p. 352):
The image the participant is seeing is on the left, the numbers in the middle represent receptive fields, and the predicted brain activity is on the right. Just compare the predicted brain activity for each image to actual brain activity, and whichever matches the best is the image the person was viewing when they produced that brain activity. Simple, right? Well, not really. To be honest, I understand about .01% of what’s going on in their analysis (I see the phrase “Gabor wavelet pyramid,” and my eyes glaze over), and apparently I’m not alone, because despite the fact that the ability to guess what someone’s seeing from their brain activity is really, really cool, the paper got no love from the blogosphere. And I was really hoping someone would post about it, so they could explain it to me.
However, a paper by Soon et al. published just this week in Nature Neuroscience(2, which makes predictions about people’s behavior from fMRI data is getting all sorts of attention, with an article in Wired and the Boston Globe, as well as in diverse areas of the blogosphere (e.g., here, here, here, and here). Now, I find the other paper more interesting, because it uses underlying theory about how the visual system works in order to predict patterns of activation from patterns of input, whereas this week’s paper just attempts to correlate activity with behavior, but apparently that’s just me. To see why people find this new paper so interesting, though, we should take a step back, all the way to February 2007.
Back then, when time’s were simpler and the dollar was worth slightly more, Haynes et al. (Haynes is one of the authors of the Soon et al. paper) published a paper in Current Biology(3) that’s basically the same as this week’s Nature Neuroscience paper. They asked participants to choose between two tasks (either addition or subtraction), then to “covertly maintain their intention” (I guess that just means keep it in mind, in neuroscientist-speak) during a delay, after which they would be presented with two numbers and the two possible answers (one for addition and one for subtraction), from which the participants were to select the correct one given the decision they’d made.
Since this is an imaging study, participants were having pictures of their brains taken all along, using fMRI. After everything was done, Haynes et al. used pattern recognition techniques to look for, well, patterns in the fMRI data taken during the delay (while participants were covertly maintaining their choice), and then correlate those patterns with each of the decisions. They then used the results of the pattern recognition analysis to predict which of the two decisions a participant had made. Activity in several brain regions were able to predict participants’ choices, including the anterior medial prefrontal cortex, the posterior medial prefrontal cortex, the left lateral frontopolar cortex, the left frontal sulcus, the right middle frontal gyrus, the left frontal operculum, all regions associated with memory, executive functioning (e.g., planning and resource allocation), and motor control. The medial prefrontal cortex had the highest prediction accuracy, correctly predicting participants’ choices 71% of the time. When thinking about this, keep in mind that this activity occurred before people actually carried out their chosen activity, so this is all activation associated with the choice itself. Or so the logic goes.
The Soon et al. paper isn’t much different from the Haynes et al. paper. Once again, they gave participants two options from which they could freely choose, though in this case the choice was which of two buttons to press. While participants were making their decision, letters were scrolling across the screen, and after making the choice, participants were told remember the letter that was present when they made the decision (i.e., when they first became consciously aware of having made a choice). Once again, fMRI pictures were being taken of their brain.
As in the previous study, Soon et al. used pattern recognition models to correlate brain activity with the two choices, but this time, there was a twist. Using the letters that participants indicated were present when they made their choice, Soon et al. could estimate (within a couple hundred milliseconds) when they became consciously aware of t heir choice, and could then look at brain activity in various regions at various time intervals prior to conscious awareness. They found that activity in the frontopolar cortex (executive functioning), signals occurring up to ten second before participants became consciously aware of their decision could predict which choice they’d made with accuracy greater than we’d expect by chance. Several areas within the parietal cortex were also accurate at predicting choice at above chance levels.
What does this mean? It means that prior to our becoming consciously aware of choosing something, at least in these simple tasks, our brains have worked out what we’re going to choose below the level of awareness. In other words, the unconscious mind is doing all the work. Duh. This is not all that surprising, and the only truly impressive thing about it is that the choice seems to be set so long before we’re aware of making it, even when the choice is so simple (it’s not like there’s anything riding on it). I’d have thought things worked a bit faster than that, though I suspect part of the reason they work so slow in this case is because there are no real time pressures.
From my perspective, there’s not really anything to write home about here. Unlike the Kay et al. paper, Haynes et al. and Soon et al. aren’t using any new techniques or models. They’re just using correlations, and my son could do that, if he had the statistical software to analyze fMRI data that is. But from the very first sentence of the Soon et al. paper, it’s clear why this study has drawn people’s attention. They begin their paper with:
The impression that we are able to freely choose between different possible courses of action is fundamental to our mental life. However, it has been suggested that this subjective experience of freedom is no more than an illusion and that our actions are initiated by unconscious mental processes long before we become aware of our intention to act.
Aha! Free will! People love that shit, and that’s how they’re thinking about the paper, as evidenced by the headlines and post titles for articles about the study, such as “Free will? Not as much as you think,” “Free will as illusion,” “Letting go of free will,” and just, “Free will?” It’s clear that people, encouraged by the paper’s authors, think this data calls into doubt the existence of free will, if it doesn’t debunk it altogether. But does it?
The answer, for me at least, is no, it has absolutely nothing to do with free will. Putting aside for a moment the fact that I’m not really sure what free will looks like in a non-dualist metaphysics, it’s pretty clear that Soon et al.’s data really only speaks to the existence of free will in a dualist metaphysical system in which consciousness is totally separate from the physical (and unconscious) mind. Or at least, you need to posit some sort of homuncular consciousness in order for it to say anything about free will. In other words, for this study to have any relevance to free will, there would have to be this conscious system (physical or not, it doesn’t matter), separate from the unconscious one, that is sitting around getting input from the unconscious system and making its own decisions freely (whatever freely means in this context). You’d have to have a conscious mind that’s watching the unconscious mind, and acting separately. If this is your model of how things work in the head, then you’ve got more problems than this data — you’ve got a hundred years worth of data to contend with, along with some difficult logical and engineering problems. If, however, you treat the conscious and unconscious minds as part of the same system, then any decision made by the latter are as free as decisions made by the former. That is, there’s no reason to treat decisions made unconsciously as less free than decisions made consciously. Unless there’s some property of conscious awareness that gives freedom to choices, but I have no idea what that property would be, and I don’t think anyone else does either.
In sum, then, the Kay et al. study is really cool, but I don’t understand it, while the Haynes et al. and Soon et al. studies are mildly interesting, but have absolutely nothing to do with free will. I blame this confusion on our continued love-affair with consciousness, which leaves us blind to the fact that consciousness is doing very little of the work in our minds, while the under-appreciated unconscious mind is doing everything and getting none of the glory. I blame Homer.
1Kay, K.N., Naselaris, T., Prenger, R.J., & Gallant, J.L. (2008). Identifying natural images from human brain activity. Nature, 452, 352-356.
2Soon, C.S., Brass, M., Heinze, H.J., & Haynes, J.D. (2008). Unconscious determinants of free decisions in the human brain. Nature Neuroscience.
3Haynes, J.D., Sakai, K., Rees, G., Gilbert, S., Frith, C., & Passingham, R. E. (2007). Reading hidden intentions in the human brain. Current Biology, 17, 323-328.