This is certainly an interesting approach, but does anybody have any firsthand experience with learning a technical discipline from primary historical sources? While I do enjoy reading the original papers for many mathematical ideas, it seems like if it were done in a course, it would be very slow-going and leave much of the material—especially in a huge subject like Discrete Math—uncovered.
I have. All of my knowledge of theoretic CS comes from research papers.
There's pretty much no such thing as computer science education in my region (Ukraine) -- no theory is covered whatsoever. Long story short, I'm studying physics instead.
Anyway, my interest in programming eventually lead me to interest in Scheme and specifically, compiler construction. And the only source of information on this subject are research papers published online.
So for the last five years I've read half of readscheme.org, and some papers from the home pages of various researchers. Topics include algorithms and data structures, various programming techniques, garbage collection, control flow analysis, and so on.
I can attest that productivity of this method is low. It's hard to read about relations between CPS and Monadic style, while you don't know the definition of either. But eventually it comes to you; after a number of papers with lambda-calculus inside you can pretty much infer the notation.
Another issue is that some large chunks of knowledge are missing; some topics are out of scope of these papers, some are considered to be common knowledge. And you often just don't know what have you missed; needless to say such education is fragmentary.
On the other hand, over the years I've mostly lost interest in physics, but interest for CS is higher then ever; and I believe it's no coincidence.
So there. While this is not the fastest way of learning, it certainly has advantages.
It seems like a really bad idea to me. Learning from primary sources more-or-less forces you to learn things in the order they were discovered, which in turn requires you to repeat and then correct every mistake that was made along the way. This is still largely how physics is taught, and Eliezer Yudkowsky correctly points out how unwise this is: http://www.overcomingbias.com/2008/04/quantum-explana.html (look for the paragraph that starts with "The order in which humanity discovered things"). In the case of discrete math, this approach would leave you with a lot of silly misconceptions like that the axioms of Boolean logic are "laws of thought".
