In Your Brain and You, I have a brief section on the need for caution in interpreting the “lighted up” regions commonly featured in reports of brain scan studies. I’ve been reading an article that’s very informative on this matter and since it is somewhat densely written and is available only in a subscription journal it seems worthwhile to summarize one or two points that it makes here.
The article is by Carrie Figdor at the University of Iowa. The title is “Neuroscience and Multiple Realization of Cognitive Functions” and it appears in Philosophy of Science, vol. 77, July 2010, pp. 419-456. The points I’m going to draw from this article are not its central thesis; they come up in the course of defending the main thesis against an objection.
An important piece of background is that many studies use “subtractive logic”. Briefly, scans are taken during the performance of several tasks that are highly similar, except that one requires a cognitive operation (the focus of interest) that is not required by the others. Activation in the other tasks is then subtracted from the task that requires the operation of focal interest. Regions whose activation levels remain significantly different after the subtraction are taken to be specially associated with the focal cognitive task.
Figdor calls attention to the fact that, since the differences of activation are small, results from several participants need to be collected, and then these results are averaged. This method has the consequence that the activation that shows up after averaging may represent regions that are smaller than those that are required to perform the focal task. That is because performance of the focal task may rely on many regions, not all of which are the same in different participants. In that case, subtractive logic plus averaging would show only regions that are common to several participants, and these may be fewer than are required for the task in any single participant.
An analogy may help here. I’ll venture one – it’s my own, not Figdor’s, so if it has some unintended misleading feature, that’s my fault, not hers. Consider a record of office workers assigned the same task (call it T1), which they all successfully complete. But they have different methods. One is more inclined to talk with colleagues, another does more internet searching, one does a lot of calculation, another uses heuristics to estimate values. These activities may wash out in averaging, even though each participant could not have completed T1 without doing his or her particular subset of them.
Now, perhaps they all send a request to archives for a file that contains a key fact, and they can do many other tasks that are similar in many ways to T1 without consulting with the archive department. Then the request to archives will show up distinctively in an averaged record of work on T1. But that request would not be “how the task is done’ – it would be only a common part of several ways of doing it.
Discovering the location of an operation that is required for a task would still be interesting, even if it were not sufficient for the task of focal interest. But subtractive logic plus averaging may not even reveal a region whose activation is required. In terms of the analogy, perhaps one worker is able to infer what’s in the archive file from other documents, and never puts in a request for it. This would lower the average value for the request somewhat, but if it were only one worker who did this, the archive request would still “light up” as distinctive of the task, even though it is not required to perform it. Figdor’s article shows an instructive pair of images that were generated from an fMRI study. One is the averaged result, for a certain layer, of the 12 participants, and the other is the image from just one participant who contributed to this average. The difference between these images is dramatic.
Figdor describes research that uses more sophisticated variants of the subtractive logic paradigm, and that holds promise of avoiding the difficulties to which I am calling attention. The lesson to be drawn is emphatically not that results from brain scanning have nothing to teach us. It is, instead, that more caution is needed in interpreting these results than is often found in journalistic reports of this work.