key word..."occasionally". :)

Actually, a lot depends on the kind of programming involved. I could imagine an AI historian that was exceptionally good at noting anomalous and exceptional data, which is often what triggers our interesting questions (Isaac Asimov said that most discoveries arise not out of 'Eureka' but out of 'That's odd....'), but to be a good historian it would then have to construct a new pattern/paradigm/narrative that included the new datum.

One problem "knowledge management" (yes, they already exist; I've read descriptions of intra-corporate systems which attempted to encapsulate all the procedural knowledge of their employees [making the employees enter data on any new procedure or decision], so that losing an employee did not mean losing any institutional knowledge) systems will have if they try to address the messes of human history is going to be the evaluation of data: contradictory and biased sources. No, they can't just be coded "Republican" or "probably wrong": we all use sources that are inconsistent and biased and sometimes you have to use them creatively and counterintuitively.

I could see a database/encyclopedia from which one could construct a "history to order" based on keywords and boundary conditions. But, as Ralph points out, it will not be original (I don't think we can program something to not plagiarize yet). We still haven't actually cracked the language barrier yet with AI. We can construct limited algorithms, but nobody has successfully linked them to knowledge bases, nor are they truly creative.

Maybe it'll come; around the same time that the professorial role becomes limited to monitoring on-line individual learning modules and supervising the construction and installation of new ones based on previous models.