Storytelling as Inverse Inverse Planning
Kartik Chandra,Tzu-Mao Li,Joshua B Tenenbaum,Jonathan Ragan-Kelley
DOI: https://doi.org/10.1111/tops.12710
Abstract:Great storytelling takes us on a journey the way ordinary reality rarely does. But what exactly do we mean by this "journey?" Recently, literary theorist Karin Kukkonen proposed that storytelling is "probability design:" the art of giving an audience pieces of information bit by bit, to craft the journey of their changing beliefs about the fictional world. A good "probability design" choreographs a delicate dance of certainty and surprise in the reader's mind as the story unfolds from beginning to end. In this paper, we computationally model this conception of storytelling. Building on the classic Bayesian inverse planning model of human social cognition, we treat storytelling as inverse inverse planning: the task of choosing actions to manipulate an inverse planner's inferences, and therefore a human audience's beliefs. First, we use an inverse inverse planner to depict social and physical situations, and present behavioral studies indicating that inverse inverse planning produces more expressive behavior than ordinary "naïve planning." Then, through a series of examples, we demonstrate how inverse inverse planning captures many storytelling elements from first principles: character, narrative arcs, plot twists, irony, flashbacks, and deus ex machina are all naturally encoded in the flexible language of probability design.