Looking Inside the 2010 GDC AI Summit

In mid-2008, the AI Game Programmers Guild (AIGPG) was born. While there were plenty of other noble goals on the agenda, our first major public undertaking was setting up the AI Summit at the 2009 GDC. It was our opinion that, with the improvement of AI becoming a focus of the game industry, there needed to be a better, more concentrated representation of all things related to artificial intelligence for games.

With 14 sessions staffed by over two-dozen AI professionals from the US, Canada, and Europe, the inaugural AI Summit was an incredible success. (At one point, the Fire Marshal actually had to turn people away at the doors due to overcrowding! Yes, we’re getting a bigger room this year). A year later, members of the Guild and the industry public alike still refer back to last year’s Summit sessions as they discuss game AI.

Even before GDC 2009 was over, the Summit participants and the Guild as a whole were already looking forward to this year. We were analyzing how the various sessions were received by the attendees in order to make sure that our second time through was even better than the first. As the contents for this year’s event have crystallized over the past two months, we believe that we have achieved that goal.

This year’s AI Summit again brings a host of the top AI professionals together for 12 sessions. While the Tutorials and Summits days of GDC were shortened this year, the AI Game Programmers Guild and Think Services agreed that there was too much valuable content to cut it down so significantly. As a result, while the other tracks will end at 5:15 PM, the AI Summit will continue until 6:00 on both days. This allows us to bring two more quality sessions that would have had to be removed. We appreciate this accommodation by the Powers That Be—and are sure our attendees will as well.

One of the changes we made this year is increasing the number of short talks. Rather than one person taking up an entire hour on one subject, seven of the eight lecture sessions are comprised of three people giving 15-20 minute presentations. (The eighth lecture involves two people.) Three panels and one “rant” session round out the schedule.

As with last year, the Guild wanted to present a higher-end level of content. There are no “beginners” sessions in the Summit. While some of the topics may present new and educational information to attendees, it is not a basic AI tutorial. We do believe, however, that there is a place for attendees who are new to AI programming. In fact, we feel that non-technical people will be able to benefit from the treatments of many of the presented subjects. Many of the positive comments we received last year were from designers and producers! Regardless ofthe skill level that an attendee comes in with, they will most certainly leave with a significant amount of knowledge and understanding of the latest and greatest in game AI.

This year’s sessions cover a wide range of topics. Once again, we tried to cover not only what has been done in the past and what is being done at present, but where the future of the industry is going.

As with last year, we look at a few of the high-profile titles that were released in the last year—Killzone 2 (Alex Champandard –Guerilla), Dawn of War 2 (Chris Jurney – recently of Relic), and Brütal Legend (Tara Teich – Double Fine). Rather than take the typical “post-mortem” approach, these short presentations specifically address some challenges thatwere faced in the design and programming of the AI and how they approached them.

Some of the sessions are specific to particular technologies. For example, Alex Champandard (AIGameDev), David Hernandez Cerpa (LucasArts),and Michael Dawe (Big Huge Games) give a group lecture on three different approaches to the upcoming industry standard architecture of behavior trees. Kevin Dill (recently of Rockstar New England) and I (Dave Mark – Intrinsic Algorithm) present a session on how utility modeling techniques can be used to not only benefit existing architectures but act independently as well.

Other sessions seek to answer the common question that AI developers are often presented with. In one session, we tackle “which is the right tool for the job?” We have collected a panel of “technology zealots” –Steve Rabin (Nintendo of America), Michael Dawe (Big Huge Games), Alex Champandard (AIGameDev.com), and me, Dave Mark (Intrinsic Algorithm). The moderator, Charles Rich (Worchester Institute of Tech.), will present typical game situations and let the panelists argue their respective cases for their own technology and against the others. The result should not only delve into the pros and cons of each approach, but also provide insight into the give-and-take process of deciding which one to use.

Another panel, moderated by AiLive co-founder, John Funge, seeks to explore a question that many companies ponder at some time oranother—“why so wary of middleware?” Steve Gargolinski (Blue Fang), Borut Pfeifer (recently of EA-LA), Brett Laming (Rockstar Leeds), and Chris Jurney(Double Fine) express their various positions covering the spectrum from “middleware saved us!” to “middleware is evil.” They explain their respective histories with products that formed their opinions and what (if anything) would change their mind.

Two more sessions tackle the decision of architecture from different angles. One session deals with “architecture mashups”. Three different people—Steve Rabin (Nintendo of America), Kevin Dill (recently ofRockstar New England), and Brian Schwab (Blizzard)—give examples of how they combined different types of architectures into one hybrid system in order to solve their particular problems. This promises to show that sometimes the best tool for the job is a Swiss Army knife. The second session discusses getting“the most AI bang for your buck.” Rob Zubek (Zynga), John Walker (recently of High Voltage), and Phil Carlisle (University of Bolton, Mindflock) give working examples of how to develop robust AI even when faced with limited resources of hardware, time, or manpower.

As games move away from linearity, developers have beenfaced with many challenges in managing interactive storytelling, drama, andfiction. Three short lectures by Dan Kline (Crystal Dynamics), Façade co-creator Michael Mateas (UC-SantaCruz), and interactive fiction writer Emily Short, look at various ways the AI world can assist designers and writers in dealing with these challenges.

Similarly, as games move toward photo-realism, the challenge has been to create behavioral believability that keeps pace with the visual aspect. Steve Gargolinski (Blue Fang), Phil Carlisle (University of Bolton, Mindflock), and Michael Mateas (UC-SC) present short lectures on how characters can be made to seem more “alive.”

As with last year, some of our sessions are more forward-looking. In a rapid-fire eye-opening session, Ian Holmes (UC Berkeley), Richard Evans (Maxis), Steve Rabin (Nintendo, Digipen), MichaelMateas (UC-SC), and Adam Russell (University of Derby, B-Block Studios), give 10-minute presentations of working demos designed by themselves or their students on cutting-edge experiments—any one of which could become the “next big thing” in game AI.

Additionally, in what promises to be the most “blue sky” session of the week, we “answer the designers’ wish list.” We asked big name, marquee designers (to be announced later) questions about what they want and need from their game characters and environments. We will present their answers to a select panel of AI programmers (Soren Johnson – Maxis, Richard Evans –Maxis, Chris Jurney – Double Fine, Brett Laming – Rockstar Leeds) and give them the opportunity to discuss how they would approach solving the challenges. In some cases, they may have an answer. In others, the programmers may have to present a wish list of their own – such as what they would need from technology before a solution to the problem is feasible. Either way, the session should expose both what designers would like to have and how AI programmers could approach complex problems that may, at present, seem to be intractable.

The final session brings eight of the Summit presenters to the podium for 5-minute topical “rants”. Rather than simple griping about the industry, these expressions are about issues that are very important to AI developers and theindustry as a whole. Some will make you laugh, some may require a fire extinguisher… but all of them are sure to make you think.

In all, the AI Summit at GDC 2010 promises to be an interesting, educational, and fun event. To put things into perspective, it wasn’t only the attendees that found last year’s event valuable. We presenters came away saying that we learned a lot. For many people, it was the best AI-related event of the year. This year’s Summit looks to be even more engaging and informative than the last. We hope that you will join us!