I think I wasn't fair in this comment, because it suggests that it is easy to do, when it actually isn't. For example, a cogntive control task might have an easy and hard condition. You'll end up with significant activation all over the brain for either of them, and their contrast will still have significant differences between the conditions all over the place too (e.g. SMA, pSMA, Anterior Insula, Anterior Cingulate, Posterior Cingulate Cortex, visual and motor cortex, cerebellum, etc. etc. etc.). Because the brain is doing all kinds of things during the task, and any "subroutine" is still going to involve communication between lot of places. But if the idea that function isn't localized, i.e. going back to a naive connectionist paradigm, then, sorry, no. The distribution of activations we get are often quite stereotyped, and if we are more careful and specific with our models we can localize function often even better.