Why Learning and Reactive Behaviors?
The animat approach emphasizes the creation of intelligent behaviors in genuine creatures. This underlines the essence of AI development, instead of just programming and optimizing academic algorithms. A balance of responsibilities is necessary in professional game AI.
This approach—which we dub nouvelle game AI—shows how to deal with modern academic AI in practice by blending it with popular technology. As game AI development faces new challenges (such as increasing intelligence, efficiency, and realism), wisdom from other well-established disciplines becomes indispensable. The approach to game AI in this book can be seen as a combination of three fields of research:
This book sheds light on nouvelle game AI as the application of these contemporary ideas to computer games. We believe that they are applicable to games in four ways (as discussed in the following subsections), improving believability and simplifying the development.
Naturally, there are some drawbacks to this approach compared with "standard" game AI. We will present both approaches when possible, deciding when nouvelle game AI is applicable, and when it's not as desirable. Notably, we will find ways to combine learning with fixed approaches and make embodiment efficient. This will enable us to use an ideal blend of technology to create the most suitable AI for NPCs, combining the advantages of both approaches.
Embodiment Makes Synthetic Creatures More Genuine, Which Improves Believability
Behavioral research shows that believability flows from accurately simulating the body of creatures, notably their interaction with the environment. When imposing such biologically plausible constraints on the perceptions and actions, realism is a byproduct of applying the AI as a brain. Each detail of the design no longer needs to be faked, and animats remain believable outside of their intended domain.
Situatedness Facilitates NPC Development and Improves the Benefits of Learning AI
Embodied systems are mostly affected by their immediate surroundings. Using senses, only local information is gathered, similar to the way humans or animals interact with their environment. This automatically filters out less-relevant information, and tremendously simplifies the problem of developing intelligent behaviors—both in theory and practice. The process of computing an action based on the perceived situation is almost straightforward. Learning techniques perform better for this same reason, reducing potential problems and increasing efficiency.
Reactive AI Architectures Are Ideally Suited to Controlling Game Characters
Like living creatures, animats must react to stimuli from the environment. Because there are various types of situations, it is often appropriate for different AI components to provide reactions. Together, these independent components can be organized into reactive architectures to increase their capabilities. There are many architecture types, but reactive ones are suited to intelligent control because of their reliability and simplicity, often providing a foundation for more elaborate techniques.
The Game AI Development Pipelines Can Benefit from Methodologies Used to Create Embodied Creatures
Embodiment separates the brain and the body. In the game engine, this explicitly distinguishes the AI from the game logic and even the world simulation itself—greatly simplifying design decisions. As for development, the animats are usually built incrementally, validating the system by experimentation. This is a great methodology to ensure realism and robustness. Finally, reactive architectures are intrinsically modular, allowing the divide-and-conquer paradigm during both implementation and testing.