Over the last few years, I have been dabbling in both game studies and practical game design, but, being a M.Sc. in artificial cognitive systems, my favorite part of development has always been the artificial intelligence – specifically, the rather unique ways it is used in video games. As I have read about the subject, I have compiled a list of illuminating resources that give a good overview of the topic that I wish I had when I myself started out.
- Artificial Intelligence: A Modern Approach [Book] by Russel and Norvig is the standard textbook for university-level cognitive system studies, so it is a general recommendation for everyone getting into artificial intelligence. For the purposes of this list, however, only Part I (chapters "Introduction" and "Intelligent Agents") is of particular importance, as it will dispel many myths that non-scientists have regarding AI (e.g. that it is synonymous with machine learning). The book can be found in any university library, but you can also find it online if you just google.
- "A Panorama of Artificial and Computational Intelligence in Games" [PDF] by Yannakakis and Togelius is an awesome and exhaustive overview of all the ways artificial intelligence is currently used in games, beyond the most common understanding of making NPCs move about in a somewhat sensible pattern. The article is unfortunately no available publicly, so you'll have to either sign up with IEEE or access it from a university network (which is how I got it) to read it.
- "What Is Game AI?" [PDF] by Kevin Dill gives an entry-level introduction to the specific meaning of "video game AI" that game players and developers put into the term, including its purpose and typical challenges.
- "Behavior Selection Algorithms: An Overview" [PDF] is a brilliant introduction to all of the key AI architectures used by video game developers: from (hierarchical) finite state machines, through behavior trees and (dual) utility systems, to goal-oriented planners and hierarchical task networks.
- "Deciding on an AI Architecture: Which Tool for the Job?" [GDC Vault] is a discussion panel by some of the leading game AI designers, exploring which architecture is best suited for which game design ideas.
What makes a good game AI?
- "What Makes Good AI?" [YouTube] by Mark Brown presents the argument that the purpose of good video game AI is to support the game designer's intended game play experience and lists a number of best practices and guidelines for it.
- "Game AI - Funtelligence" [YouTube] by Extra Credits argues that game AI's goal is to make the game more mechanically engaging and lists three main concerns that guide its design: discernability of behavior, potential for interesting player decisions, and economic viability.
- "Less is More: Designing Awesome AI" [GDC Vault] by Kimberly Voll is a short but illuminating talk where she explains the general purpose of game AI (echoing Kevin Dill) and illustrates it with a rather mind-blowing example from her game RocketsRocketsRockets.
- "Playing To Lose: AI and Civilization" [YouTube] by Soren Johnson elucidates the quintessential difference between the academic "good" AI (like AlphaGo and its ilk), which "plays to win", and the traditional "fun" video game AI, which "plays to lose", and how it influenced the AI he wrote for Civilization III.
Nuts and Bolts
- "The Structure of a Game AI System" [Web] by Daniel Sanchez-Crespo Dalmau is a really low-level list of basic components of a game AI architecture that are so trivial, most other sources don't even bother talking about them.
- "Scopes and Frames of Video Game AI" [Web] is my own tentative take on AI agents' positioning within the context of a particular video game – something I have not seen presented as a coherent systematic model before.
- "Three States and a Plan: The A.I. of F.E.A.R." [PDF] by Jeff Orkin is a classic text explaining how he had adapted the (hitherto academic) goal-oriented planning algorithm STRIPS (read more on it in Chapter 10 of Russel-Norvig) to the ground-breaking AI of F.E.A.R. (2005), as well as how it actually works. For some context on why GOAP is not used much in games anymore, see also "This Year in Game AI: Analysis, Trends from 2010 and Predictions for 2011" [Web] by Alex Champandard (the planner AI advocate from the "Deciding on an AI Architecture" panel linked above).
- "Behavior trees for AI: How they work" [Gamasutra] by Chris Simpson is a great introduction to game development's most popular (and arguably the most flexible) AI architecture. For some perspective on how behavior trees came to dominate game AI, see also "Are Behavior Trees a Thing of the Past?" [Gamasutra] by Jakob Rasmussen (the CEO of a company specializing in utility AI, so take his opinions with a grain of salt).
- "Improving AI Decision Modeling Through Utility Theory" and "Embracing the Dark Art of Mathematical Modeling in AI" [GDC Vault] by Kevin Dill and Dave Mark are two hour-long introductions to dual utility-based AI architecture. You can also read up on it in Russel-Norvig (Chapter 27), or in Dave Mark's Behavioral Mathematics for Game AI [Book] (which I actually bought and read while working on the AI for my Winter Palace).
- "The Simplest AI Trick in the Book" [GDC Vault] is actually just one episode of a series of regular GDC AI Summit panels (search for others on the Vault), but I find this particular installment especially interesting thanks to the segment about realistic human-like agent reaction times.
- "Remember to Relax! Realizing Relaxed Behaviors in AAA Games" [GDC Vault] by Anguelov and Shroff gives an in-depth look into the so-called "external actions", i.e. outsourcing the logic for non-core peripheral behaviors into the game world objects.
I also strongly recommend the Game AI Pro website as a resource for all things game AI, and freely admit to have totally stolen this reading list format from Mark Brown's awesome Game Maker's Toolkit. Lastly, I suggest Dr. Tommy Thompson's AI and Games YouTube channel for a more visual kind of AI content.