I completely agree here. Good information here, and lots of potential when you think about it could be applied with other machine learning algorithms that are currently being used with "real-time fMRI". Indeed, I don't think we're far off from reducing the gap between neuronal activity and behavioral patterns, but we need to keep in mind that the brain is incredibly complex, highly variable between individuals, and most of the time, presents quite a poor signal-to-noise ratio with the technology currently available.
Perhaps it's more practical and better with temporal resolution, but it depends on how concerned you are with spatial resolution too (which EEG is quite poor with).
That's the point of this article. The experiment routes around the lack of good spatial resolution in the data. It's like a much more sophisticated, much less easy to game polygraph.
