Even beyond our current limitations in brain imaging technology, it's not at all clear that slapping some sort of apparatus on someone's head and then setting them loose in the world could be very informative even in principle. The brain imaging technology itself is only half the story in brain imaging studies; the other half is very strict experimental control over what the subject does and doesn't do. Letting people do whatever they want while we scan them is basically the definition of a poorly controlled study.

In order to identify the neural correlates of some cognitive function, we have to compare the imaging data that we get from a subject when they are doing that task to the data that we get when they are doing a separate task which is exactly like the first task in every single way, except for the exact thing being studied. Further, we have to observe the subjects doing each of these tasks very many times consecutively in order to wash out incidental noise in the data. Then we use a sort of "subtraction" method, whereby we take the aggregation of all the experimental trials (trials where the subject is doing the thing of interest), and we subtract from that the patterns we see in the aggregate of all the control trials (trials where the subject is doing the matched task). What we are left with is a fairly good idea of which brain areas are active when the subject does just that one thing. (How useful that information actually is to us is another story entirely.)

A major problem with this naturalistic imaging data is that it would be extremely hard to identify "control tasks" that we could use to make sense of the data that we observed when the subject was doing the thing we are actually interested in. Imaging data is only informative relative to other imaging data; it's meaningless on its own. And when/if we could identify (purely through luck) a reasonable control task to use in examining a particular cognitive function, we are virtually guaranteed within this paradigm to get very noisy and low-quality data. Moral of the story: proper experiments are crucial.