If you start thinking along the lines of first principles you could just as well start with a NAND gate.

..because some people were even smart enough to pull this off: http://www.nand2tetris.org/ :)

McCarthy had something to say about this... http://www-formal.stanford.edu/jmc/recursive.pdf

Which paper is the one with the symbolic differentiator ?

Recursive first principles. I don't think n2t will put the emphasis on that (no offense, I love their work and results, something I wished for years).