AI recognition of differences among book-length texts

Abstract

Can an Artificial Intelligence make distinctions among major works of politics, philosophy, and fiction without human assistance? In this paper, latent semantic analysis (LSA) is used to find patterns in a relatively small sample of notable works archived by Project Gutenberg. It is shown that an LSA-equipped AI can distinguish quite sharply between fiction and non-fiction works, and can detect some differences between political philosophy and history, and between conventional fiction and fantasy/science fiction. It is conjectured that this capability is a step in the direction of “M-comprehension” (or “machine comprehension”) by AIs.