Non-Player Characters and Artificial Intelligence

Introduction

Non-player characters (NPCs) are entities that embody a role in a game context that is usually performed by a person. For example, in a bullying scenario in which a player plays the role of a bullied child, diverse NPCs are needed to play the roles of the bully, the bully’s friends and the teacher (Vannini et al., 2011). These characters are controlled by artificial intelligence (AI) algorithms, which generate behaviours based on different human characteristics according to the design goals. The modelled behaviours can range from simple wandering entities that populate a scenario (Thalmann et al., 2004; Ulicny & Thalmann, 2001), e.g., members of a crowd, to more complex and human-like entities that intelligently interact with a player (Gratch et al., 2002; Magnenat-Thalmann & Kasap, 2009).

However, at its core, what is a Non-Player Character? The answer is an autonomous agent (for games) that senses its environment and acts intelligently and independently from the player. In fact, different types of characters¹ represent attempts to mirror human behaviours either at a specific or a holistic level depending on the game’s requirements.

Common approaches to modelling of NPCs result in the creation of different types of entities: scripted (Spronck, Ponsen, Sprinkhuizen-Kuyper, & Postma, 2006), reactive, deliberative and hybrid (Wooldridge, 2002). Each of these approaches contains different design purposes and applications. From a designer's point of view, to establish a parallel with real people, we can conceptualise others in different ways according to the context of the situation at hand. For example, if we consider the management of exits in a stadium, we focus on people's steering and the resulting flow behaviours through bottleneck points in the environment. In this situation, people are usually more adequately modelled in NPCs by reactive characters that can exhibit emergent crowd behaviours based on simple rules (Thalmann et al., 2004; Ulicny & Thalmann, 2001). In another example, if we consider a situation of emotional support for a learning situation, we are interested in the advisor's (character) knowledge, skill and ability to help the user (João Dias & Paiva, 2005). In this case, modelling of the advisor person in an NPC role is better achieved by means of a deliberative or hybrid (combining different types of behaviours) entity that contains the required mental models necessary to respond to the situation. Beyond these examples, several other focused behaviours can be applied, e.g., genetic algorithms (Krovi, Graesser, & Pracht, 1999), to evolve behaviours or machine learning used to exhibit human-like learning capabilities (Spronck et al., 2006).

Another important aspect of NPCs is their form. Given the game-like types of scenarios, the NPCs we discuss in this chapter are usually virtual characters. The visual representation of the characters aims to approximate the experience of player-character interaction with that of human-human interaction. If we combine the virtual characters with the previously described creation of intelligent behaviours, we create intelligent virtual characters. These virtual characters are intended to give the illusion of human behaviour by simulating many human aspects (physical and cognitive) and modalities (perceptions and action) for dynamic social environments. To many, the ultimate goal is the creation of synthetic characters that “are not only believable but also as remarkable and unforgettable, as humans are”².