Recently there has been a lot of progress in brain reading; for instance Here is a nice piece done by CNN, here is a nice article on brain reading video games, and here is a link to Frank Tong’s lab, who may be familiar to those who regularly attend the ASSCor the Tuscon conferences. This stuff is important to me because it will ultimately help to solve the empirical question of whether or not animals, or for that matter whether we, have the higher-order states necessary to implement the higher-order strategy for Explaining What It’s Like so I am very encouraged by this kind of progress. The technology involved is mostly fMRI, though in the video game case it is scalp EEG. But though this stuff is encouraging fMRI and scalp EEG are the wrong tools for decoding neural representation, or so I argued in my paper “What is a Brain State?” (2006) Philosophical Psychology 19(6) (which I introduced over at Brain a while ago in my post Brain Statves Vs. States of the Brain). Below is an excerpt from that paper where I introduce an argument from Tom Polger’s (2004) book Natural Minds and elaborate on it a bit.
Polger argues that thinking
that an fMRI shows how to individuate brain states would be like thinking that the identity conditions for cricket matches are to pick out only those features that, statistically, differentially occur during all the cricket games of the past year. (p 56)
The obvious difficulty with this is that it leaves out things that may be important for cricket matches but unique (injuries, unusual plays (p 57)) as well as includes things that are irrelevant to them (number of fans, snack purchasing behavior (ibid)). The same problems hold for fMRI’s: they may include information that is irrelevant and exclude information that is important but unusual. Irrelevant information may be included because fMRI’s show brain areas that are statistically active during a task, while they may exclude relevant information because researchers subtract out patterns of activation observed in control images.
I would add that at mostwhat we should expect from fMRI images are picture of where the brain states we are interested in can be found not pictures of the brain states themselves. They tell us that there is something in THAT area of the brain that would figure in an explanation of the task but they don’t offer us any insight into what that mechanism might be. Knowing that a particular area of the brain is (differentially) active does not allows us to explain how the brain performs the function we associate with that brain area. We need to know more about the activity. Consider an analogy: we have a simple water pump and want to know how it works. We know that pumping the handle up and down gets the water flowing but ‘activity in the handle area’ does not explain how the pump works. Finding out that the handle is active every time water flows out of the pump would lead us to examining the handle with an eye towards trying to see how and why moving it pumps the water.
And, as I go on to argue, after examining those areas to find what the actual mechanisms are neuroscience suggests that it is synchronized neural activity in a specific frequency that codes for the content, both perceptual and intentional, of brain states. So, multi-unit recording technology (recording from several different nuerons in the brain at the same time) is the right kind of technology for looking at brain states. This is not to say, of course, that the fMRI and EEG technology is not valuable and useful. It is, and we can learn a lot about the brain from studying it, but it must be acknowledged that it is ultimatly, explanatorily, useless. To find higher-order thoughts or perceptions we will need to use advanced multi-unit recordings.