Nelson Goodman’s Fact, Fiction, and Forecast42—written in 1954 and perhaps one of the most important and underappreciated books on machine learning—suggests a kind of knowledge called an overhypothesis, because it helps to define what the space of reasonable hypotheses might be.
