The Allure of Games. The conference draws these crowds each year because the allure of games runs deep. Before computer games became commercial, they were a secret vice: programmers wrote games like Adventure and Space War for each other in their spare time (and had to steal cycles on machines dedicated for other purposes to do it). But then Syzygy's Computer Space hit the market in 1972, later spawning Pong and Atari; and since then, games have matured into a full-fledged industry, with tens of thousands of dedicated game developers producing games for millions of gamers (most of whom own hardware dedicated for nothing but games).

Games are a multibillion dollar industry - over 6 billion in 2001 and still growing. Ignore for a moment the legions of distributors unloading battalions of CDs and cartridges on the shores of CompUSA and WalMart. Behind the front lines, a huge infrastructure has sprung up to support the builders of games. Industry giants like Intel and Microsoft, desperate to ensure that the next generation of games are destined for their chips and operating systems, courted game developers vigorously with show booths the size of houses filled to the brim with tools and incentives. Surrounding these landmarks were a sea of smaller companies, trying to make their mark with the latest solutions for building virtual worlds (Lithtech's Jupiter game engine) or populating virtual worlds (Biovirtual's 3DMeNow character animation) or making the virtual populace dance (Vicon's motion capture solutions) or painting virtual tatoos on the virtual populace's faces (Right Hemisphere's Deep Paint) or, toughest of all, keeping track of the tens of thousands of pieces of data that define your tatooed populace dancing over your virtual landscape (NxN's AlienBrain content management system).

(Note that being simply being mentioned above does not constitute an endorsement; a statement like "Fractal Design Painter 3.1 Rules!" would constitute an endorsement. On that note, procreate Painter 7 Rules!)

But it is more than just big business: games are a fundamentally creative industry. While movies are still a bigger industry, worldwide game sales have surpassed American box office receipts - a large enough chunk to notice, and still growing. Media giants like Sony get it: one of the conference's chief sponsors, it inundated attendees with t-shirts and talks and beer mugs promoting its plans for its game consoles, present and future. But the game phenomenon has strong grass roots: like Hollywood, where you can't catch a cab without meeting a driver working on his script, the GDC was filled with young aspiring developers either hatching their own game idea or working with a team out of a friend's basement, scouring the floors for inexpensive development tools and new ways to deliver their games to a wide audience.

The game industry courts these new teams and new developers vigorously. Premiere visual effects company Alias|Wavefront released a free personal version of its flagship Maya animation package, desperate to crack the market held largely by Discreet's 3D Studio Max; game engine creator Lithtech is willing to discuss bargain pricing for new studios using its game engine, whereas Epic offers a free version of its Unreal engine and level editor with every copy of the game. The conference in many ways was geared to young professionals trying to break into the field: several talks and roundtables focused on becoming a game professional, and numerous booths and private rooms were set up to help prospective employees and employers mix and mingle.

Creativity, AI, and Software Development. The conference also drew a crowd that you might not expect - authors, artists and academics of all descriptions. Cyberpunk author Bruce Sterling of Mirrorshades fame canceled his talk on the undeground of games, but comic artist Scott McCloud, engaged "The Sims" designer Will Wright in a battle of dueling laptops. Recreational mathematician John Conway, creator of the "game" called mathematical game called "Life" and a longtime favorite of Martin Gardner's mathematics column in Scientific American, gave a keynote talk on mathematical approaches to games. In that vein, the Independent Game Developer's Association sponsored a summit designed to improve the relationship of academia and games, and one of the conference's chief sponsors is the newly formed, entertainment-focused Digipen Institute of Technology.

It was the conference's catholic appeal to all creative disciplines that was of greatest interest to me. Readers of the Library probably know that I like to tell stories, like to build software, and like to design minds. Computer games combine all three: a happy nexus of creativity, software engineering and artificial intelligence - and I found the Game Developer's Conference an ideal place to expand my mind on all of these topics.

The conference hosted a dozen sessions on games and artificial intelligence in several different "tracks" - AI roundtables, AI in strategy games, tutorials on AI techniques. Most of the sessions were repeated, but unlike similar sessions at the conference, which were held more than once simply for scheduling convenience, the AI sessions seemed to burst their bounds, with topics spilling over from one session to the next and finally culminating in an AI Programmer's dinner Saturday evening.

During this debate, in which designers of games like Starcraft bantered with Ph.D's in artificial intelligence, a few key concepts emerged. The most important idea was that the best solution is the most effective one. While that sounds obvious, the artificial intelligence community is engaged in a very long range project, and researchers naturally organize into schools of thought each pursuing their own path to the prize. This competition leads to vigorous arguments between schools, usually between champions of "good old fashioned AI" (such as yours truly), usually seen as symbolic, serial, and programmatic, and champions of competing schools, who distinguish themselves by being neural, parallel, dynamic - pick your false dichotomy of choice.

These kinds of debates were refreshingly absent; the debate focused on what techniques work and how to effectively use them. Finite state automata and fuzzy state machines ruled the day, though decision trees and other more complex control structures were beginning to be discussed. Technology for pathfinding based on the classic A* algorithm has become so refined many attendees did not wish to discuss it; one key advance which made this technology even more powerful was terrain analysis to break a large and complex map into a simpler network of locations.

Game AI programmers made these basic control techniques powerful using several simple methods. Hierarchical controllers enable simple agents to organize into squads and groups. "Idle" animation sequences create the appearance that agents have an inner mental life; combining the idle animation with simulated perception can create a powerful illusion of intelligence if, for example, a guard doesn't see you until he idly glances in your direction.

The concept which struck me the most was the power of simulated perception working hand in hand with both human and computer terrain analysis. Level designers are in a unique position to label significant features like traversable terrain and resupply points as a level is built. The game engine can then preprocess these levels, extracting connectivity, visibility and choke points directly from the map structure. With this rich set of features at hand, it is relatively easy to build a game AI capable of very complex behaviors simply by triggering off the intelligence embedded in the map.

For example, with visibility, connectivity and resupply information, an AI engine can pick a good sniping spot, or, to take a more famous example, with enough smarts mapped into the environment a Sim "knows" to hit the fridge when hungry and go to bed when sleepy, even though the Sims basic programming knows nothing about fridges and beds - only about what the Sims basic environment affords to them.

The creative sessions were similarly rewarding. After a few lively roundtables in the Hilton and a few circuits of the glittering Expo floor, I left the white walls of the Convention Center and skipped over the streetcar line to the Spanish Mission styled Civic Auditorium, where Will Wright, designer of the Sims, and Scott McCloud, author of Understanding Comics, whipped out their laptops and had a friendly duel over the nature of time in movies, games and comics.

Movies are dynamic, existing only in time and requiring viewers to devote a block of time to appreciate them; books and comics are "frozen" in time, granting readers far more freedom to appreciate them over any number of sittings or in any order. Games, in contrast, are in the middle: like movies they are dynamic and must be experienced in time, but like books they can be paused and resumed and replayed in many different ways. Unlike both, games are fundamentally nonlinear and interactive: a game's stories and visuals are created collaboratively between the designer and the gamer through the game engine. Even in a game with a linear plot like Half-Life, a gamer must be an active participant, making thousands of minute choices which ultimately make no two experiences in a game being exactly the same.

Unfortunately, the unique nature of each gaming experience makes testing games a bear. When a problem occurs, replicating the error can be difficult - just one more problem which makes game development a particularly challenging software development problem. This is particulary rough for complex systems like game AI - a recent book on AI Game Programming Wisdom devotes an entire chapter to tools for debugging game AI. As another example, an ideological debate erupted in the AI Programming roundtables when a visiting professor questioned the table about the role of machine learning; one seasoned developer (himself a veteran of a commercial AI company before his life in games) rejected the use of machine learning on the grounds that if you make your game engine learns from each session, it has become a fundamentally nondeterministic software artifacts and cannot be reliably tested (or, more colorfully, "your QA department will kill you"). After some debate the room came to a consensus that while it is best to agree to disagree, no-one wanted unpredictable game engines but everyone wanted game characters that appeared to learn, regardless of whether the characters used machine learning techniques or not.

The Point. These kinds of discussions and debates illustrate to me why games are important for computer science. Clearly games are important for their economic impact as a sizeable industry and for their social impact as a new literary medium. But for computer scientists, games are a challenging testbed for computing advances. The rigors of game development schedules and the requirements for a polished gameplay experience puts the cherished technologies and methodologies of computer graphics, artificial intelligence and software engineering to the test, and researchers should pay close attention to how effective these technologies are in practice.

Computer graphics researchers already know this well, but it is my impression that software engineers and artificial intelligence researchers have a lot to learn from the game industry. AI researchers need to hear that for the game environment complex control structures and deep machine learning algorithms are not as effective as simple state machines operating over rich perception and environment maps. Software engineers need to hear the lively debates over team size, software methodologies, and coding standards, hearing the experiences of coders on both sides of each debate - and measuring which groups tend to come in on time and under budget. (It amused me to no end to sit down in a random chair at a software engineering session - only to find myself in the middle of the only other extreme programming practicioners in the room.

Games are a tremendously rewarding field for both creative artists and computer scientists. Games are a new medium: my experiences playing Myst, Half-Life or Elite Force stick with me as vividly as any in a book or movie. Games are challenge for computer science: the techniques used to port Zork to microcomputers, or to run Quake in real time, or to have your foes sneak up behind you in Unreal, are as technically brilliant and as worthy of study as anything done in any other area of computer science.

For those involved in it, game development can be frightfully challenging, but immensely rewarding. Each year I have gone to the Game Developer's Conference I have learned a great deal, and not just about how to make games - about how to build AI systems, about how to engineer software, about how to be a creator.

I'll be back next year. I encourage you all to do the same.

See you in 15.
- The Centaur
Renaissance Engineer