The optical character recognition is not itself to blame. I used it to digitise the early, handwritten, chapters of my mother's book which I was preparing for distribution amongst family and friends. However, like the human eye it can misread characters in poor quality documents (typically, an o which has lost part of its right hand side is taken for a c). This means that every thing produced needs to be vetted for errors which appears not to have been done by the contracors employed by Ancestry.
Also the software needs to be told where to look for the information to be extracted. It doesn't think for itself. If it looks in the wrong place you get the wrong answers. My parents and I appear to have lived in four or five different streets between 1945 and the early sixties despite the registers for that period be being of identical format.
Don't shoot the inventor, sack the contractors.