Summary

AI development is fundamentally an engineering task:

• The requirements and design goals must be quantified.

• The traditional approach to game AI is highly controllable, closely integrated with the game engine.

Recent developments in modern games have improved upon this, notably by borrowing modern ideas from AI. This trend progresses toward embodied animats with learning capabilities:

• Genuine embodiment can improve behaviors and reduce the need to "fake" behaviors individually.

• Embodiment is good practice because it formalizes the interfaces between body and brain, providing guidelines to separate the AI from the logic and simulation.

• Learning comes in a variety of different flavors, appropriate in many situations.

• Learning can be done online or offline, in supervised or unsupervised forms.

Achieving this can be surprisingly difficult:

• Much background knowledge is required; programming skills are assumed in this book.

• There are guidelines for the AI development process, but this is a rough sketch that needs customizing.

• Iteration and testing is the backbone of success in AI, just as in game development.

The next chapter covers reactive techniques and AI architectures, showing how they can help improve game development.