Design Pattern Analysis of Autonomous NPC Data Visualization in Games

Alexander King

Extended Abstract

Non-Player Characters (NPCs) controlled by artificial intelligence (AI) are widely featured in digital games, providing game worlds with adversaries, allies and general inhabitants. However, a small subset of games foreground the use of these autonomous AI agents such that their interactions are the primary focus of gameplay itself. Although the design affordances of NPCs which are able to interact with their world and each other are rich and surprising, there are unique challenges in communicating the complexity of these systems to the player. Though there is a growing body of research into methods of architecting systems of emergent AI actors (Park, 2009), this research has focused on creating interesting interactions for an audience with a high degree of systems literacy. Similarly, much work has been done into methods for creating believable and relatable NPCs (Warpefelt, 2013), but such work is focused on ways NPCs can be made to mimic human interactions, not in how to better express an underlying NPC behavior system.

In Expressive AI : Games and Artificial Intelligence (2003), Mateas explains an interdisciplinary approach for describing the use of AI in ways that are not “purely technical.” (pg. 1). The author brings together terms from “game studies, design practice and technical research” to establish a new vocabulary for analyzing what he calls “Expressive AI”. He describes how AI can be the focus of a game experience, and uses Pac-Man as an extended example of a game where AI agents are the primary driver of gameplay; foregrounding AI agents, in particular their behavioral interactions, can open up novel design spaces. A design pattern analysis is proposed to better understand best practices for games which foreground NPCs in this way. The purpose here is to promote better player understanding of these underlying models of behavioral interactions. The challenges of making believable NPCs in a social simulation is not “strictly an AI challenge” (Khandaker-Kokoris, 2015), but also a challenge of design. Because one of the primary challenges facing complex social simulations is providing feedback to the player, which is difficult “when what you’re communicating are thoughts and feelings” (2015).

From the existing body of research, we can see established traditions for examining NPC-Player dynamics, emergent and expressive AI, and gameplay data visualizations, all of which are separately exploring many of the same underlying questions. Thus, the question to be investigated is: Games which foreground AI are designed to provide opportunities for spontaneous & lifelike interactions with and between autonomous NPCs. How can these games communicate the inner workings of these NPCs and relationships to the player without overwhelming them in details, or providing so little detail that they seem random? Any single approach is insufficient to answer these questions, an interdisciplinary and empirically based approach is proposed to answer this question. This will help advance understanding of one of the intractable problems of AI design in games.

A content analysis based approach is proposed that draws simultaneously from artificial intelligence research, character believability analyses, and data visualization, in order to identify common problems and best practices amongst extant digital games which foreground NPCs.