Abstract

How should we incorporate algorithms into humanistic scholarship? The typical approach is to clone what humans have done but faster, extrapolating expert insights to landfills of source material. But creative scholars do not clone tradition; instead, they produce readings that challenge closely held understandings. We theorize and then illustrate how to construct bad robots trained to surprise and provoke. These robots aren’t the most human but rather the most alien—not tame but dangerous. We explore the relationship between the reproduction of tradition and the generation of surprise and then show how formalizing a particular humanistic theory as a Bayesian prior allows us to identify readings that disrupt it through a process of algorithmic abduction. Among the exploding universe of surprising (and mostly ridiculous) possible readings, algorithmic efficiency allows us to select readings that nonetheless garner substantial support from the total archive and so merit interpretation and engagement. It is in such interaction between alien and human readers that meaning is made and understanding disturbed; it is by extending alien readings to novel texts that the scholarly community tests the value of those readings.