Early in the history of the field of artificial intelligence (AI), a paradigm known as microworlds emerged in which researchers constructed computer simulations of aspects of the real world from which their nascent AI systems could learn. Although microworlds were ultimately abandoned, AI researchers have recently called for their return, this time borrowing from the literary genre of interactive fiction, whose forms and conventions they might use to represent the world in text for the purpose of teaching machines to speak. This confluence of literary form and scientific method invites a closer examination of the relationship between word and world in AI research. The author argues for a reading of microworlds research and of AI more broadly through the lens of literary realism and through the literary texts that comprise its data sets and from which researchers expect artificially intelligent machines to learn about the world. The question of what kind of knowledge literature represents lies at the heart of AI research and thus presents an opportunity for a deeper engagement between AI research and literary, game, and media studies.