authentication, reliability, findablitilty
Ignorance is very much a legitimate approach to ai - not that you are going to get past a Ph.D. thesis dissertation with the matter but I hear a lot ringing in you proposal.
OCR is a really hot topic in several code camps, it's relationship to ai is well proven and in no undergraduate course I would sit still for would there be any less than ten percent of the effort definitely scheduled for ai or related topics. There is significant risk, as in burned sheepskin, to set a definite and wire bound course on solely this goal. Use the 80/20 rule. ( not hard to find ) temper with common sense and invited rebuttals. Telltale for me is that there is no such thing as a class in the backplane of a 7700 - it is all wires. Class and many programming practices we see sold wholesale derive or descend from some good problem solution ideas but any code camp that cannot go with an Owl swooping on it's furry little hide is sure to die in an ide.
On the short, it will be ineffective to ignore having not been in the field and as such having something to ignore. Ignoring non significant information in a scan done in an Office would be forefront work. Technology represents the human internalization of an ideal world, and is getting too far adrift in a world model of it's own fantasy. This is well known in contemporary forefront work at the Doctorate level in funded research, real sincerity and moderated thought would be given non-trivial examination by a wide array of workers in the arena. In paticular: Intentional near systematic acceptance of incorrigible sub-prime seekers as quotidian may be correlated with a nearly failed world banking system.
I seriously doubt the near failure could have propogated with clear & concise examination of all results, there are thesis subjects not as worthy, the challenge would be to not become one of the ignorant. Easier said than done if people can see right through you.