Throwing Bottles at God: Predictive Text as a Game Mechanic in an AI-Based Narrative Game

Abstract

We present Throwing Bottles at God, an experimental interactive narrative game that makes use of a predictive text writing interface as both a game mechanic and a means by which to deliver narrative content. The player steps into the role of @dril, a well-known pseudonymous social media personality with a distinctive writing style, and authors short snippets of text while receiving suggestions from the game as to which word @dril might use next – suggestions supported by word pair frequency data extracted from the corpus of all existing tweets by the actual @dril. The game represents a first attempt to use AI-based game design to heighten the player’s awareness of AI algorithms, specifically predictive text algorithms, as they play a role in the player’s day-to-day life. It also blurs the line between player-authored and developer-authored narrative content by inviting players to freely mix snippets of developer-authored text into their own in-game social media posts as they compose them, resulting in player-assembled messages that embed sequences of words drawn both from an external corpus (the @dril corpus) and from developer-authored narrative content.