I look at this article as a good example of thinking about the incompleteness of most of our probability modeling.
Clearly, just like we hope to be able to defend our planet by striking an asteroid with a spacecraft impactor to deflect it enough, early enough, for it to be made to miss Earth, there is an opposite side to the same physical concept - a natural object collision with an asteroid could make a miss turn into a hit on Earth.
Looking at it as carefully as we know how, including what sized objects could move Apophis' path 19,000 miles towards Earth by 2029, how many such objects are in the area of its orbit, and what the probability is for such an impact to move the orbit to the "wrong" path instead of a "better" path, seems like good scientific thinking to me. It helps us gauge the reliability of the projected path and prediction that we are "safe" (from this one) for 100 years or more.