Extrapolating from the article and associated research, what you're describing as a mental "flex" may be an ability to consciously weight higher-level models over lower-level ones, and vice-versa. I'd be interested to see the fMRI results of a bunch of people who can do that "flex."