Programming a Computer for Playing Chess (1949)

  The chief weakness is that the machine will not learn by mistakes. The
  only way to improve its play is by improving the program. Some thought has
  been given to designing a program which is self-improving but, although it
  appears to be possible, the methods thought of so far do not seem to be
  very practical.

I went looking for papers doing any evolutionary programming wrt avoiding mistakes. TD Gammon [1] seems pretty close. It's not a terribly long paper:

  The only thing which prevents TD-Gammon from genuinely equaling world-class 
  humanplay is that it still makes minor, practically inconsequential technical 
  errors in its endgame play

[1] http://www.bkgm.com/articles/tesauro/TDGammonAchievesMasterL...