No one would write a program with the intent of doing this either. Machine-driven processes are inherently imperfect. If a program is attempting to differentiate text from non-text, for example, there may be stray pieces of text that the program is unable to recognize as text.
Only if you start out with that premise. Any business that writes document software of the sort Adobe does with the intent of as perfect a reproduction as possible given the limitations of digital reproduction isn't going to settle for a program that mish-mashes the original just because the first person they gave the job to didn't know what he was doing.
Try playing chess against a $60 software package sometime. We humans are much more imperfect in many ways than our machines.
ML/NJ