The issue with this circumvention is that it assumes wave function collapse, which is equally mysterious. If you do away with collapse, you are left with a many-worlds type interpretation, in which the wave function simply evolves unitarily forever. It is in this sense that the current state "dictates" the previous states---the histories consistent with one measurement or another separated based on the kind of measurement made. If you do the interference experiment, the histories consistent with outcomes of that experiment are ones in which the photon takes no definite path. If you do the path-determining experiment, you no longer get interference patterns but also the consistent histories are ones in which the photon takes a definite path---either one slit or the other.

Since you get to decide which kind of experiment to do after the photon passes the slitted screen, your decision seems to force the histories to become either the interference kind or the definite-path kind after you might expect that to have already been determined. That's the sense in which the future seems to control the past.

I agree that this was not explained clearly in the article. It's a very slippery subject.