The ‘AI Conference Round-Up’ is dedicated to breaking down the most interesting talking points of recent events in AI circles. In this first entry, I talk about my recent attendance to the Game/AI conference in July of 2014.
This year was my first time attending the Game/AI conference in Vienna. Having travelled to academic game AI conferences numerous times over the years, I felt it important to take the plunge and visit the other side. ‘Game/AI’ is ran by Alex Champandard and his team at AIGameDev.com, which is the largest site dedicated to AI practices in the video games industry. As such, Game/AI is a conference where like-minded professionals come together to discuss their design philosophies and practices.
Overall it was an interesting and engaging experience, with some strong presentations and plenty of fun discussion throughout. What follows is some of my observations and thoughts having returned from the event.
The Need to Avoid Complacency
One issue that Champandard acknowledged in his opening speech and was re-iterated by many throughout the conference was the need to avoid complacency. In fact in some circles it was argued that developers had already been complacent and instead they should be focussed on how to reinvigorate their ideas and practices.
It’s perhaps not a surprise that these comments come at the beginning of a new console generation. Given that consoles often carry the largest market share for genres, they will help to define the size and scope of future projects. The Playstation 4 and Xbox One have raised the minimum bar for performance for games on future systems (the WiiU of course being an exception). This means that worlds are becoming bigger, richer and more complex. As a result, there is a need for more intelligent, reactive and engaging AI within games and the industry needs to keep ahead of players expectations.
What followed throughout the conference was some interesting talks about how developers from across the world are addressing this issue.
Multiple Systems & Emergent Play
If I were to summarise the core philosophies discussed in many of these talks, it would be as shown above. Many developers, ranging from Cryteks’s Crysis 3 team, to Digital Extreme’s work on Warframe and Ubisoft’s team working towards the imminent release of FarCry 4 iterated this point: a need for multiple systems that are – in principle – simple, yet work together to create scalable, interesting gameplay.
This is often the reality of industry practice versus academic endeavour, given that there is a need for game AI to be fast yet as flexible as possible. Often this results in what may be considered ‘basic’ methods pervading within the industry. Of course in recent years this has been discredited, what with the work on introducing planning technologies as discussed previously on this site. What is exciting to see from these talks is that old ideas have evolved into words of wisdom and a enthusiasm for new ideas is emerging, while acknowledging that sometimes reinventing the wheel is not necessary.
Two talks that caught my interest during GameAI were focussed on adopting AI in areas that we typically do not consider. Often the emphasis is on non-player characters (NPCs) within the domain of bombastic shooting or action games. Hence I was intrigued by two works: one that changed the genre, the other changing the platform.
Nicholas Bonardi delivered a fantastic keynote on the AI behind Ubisoft’s Rocksmith 2014, a game designed to aid novice and advanced guitar players to continue to hone their skills. The focus of Nicholas’ talk was on the ‘Session Mode’, where players practice alongside a virtual band. This band is ultimately an AI system that listens to the player and tries to accommodate for what they hope to achieve. We will be coming back to talk about this game in more detail in the ‘AI & Games’ series.
The second talk was focussed on mobile title Republique by Camouflaj, a stealth game that makes players consider how to move a vulnerable protagonist in a hostile environment. Given that the adoption of AI in mobile titles has weighed heavily on my mind in the last couple of years (to the point I have previous and current work in the area) so I am pleased to see some movement on that front.
MCTS In The Wild
A big surprise of the conference came in the very last talk by Tim Gosling on the work by Creative Assembly on Total War: Rome II. The focus of the talk was on the application of an algorithm known as Monte Carlo Tree Search (MCTS).
MCTS is one of the most interesting developments that have emerged in academic circles in the past 10 years. The algorithm adopts principles from tree search, while using Monte Carlo inspired random simulations of the world to predict future outcomes. In short, MCTS is powerful and smart, but often comes with the cost of CPU power, given it needs to conduct a large number of simulations of future outcomes.
Total War is to my knowledge the only game that has in fact adopted MCTS as part of its AI systems and potentially shows a future for further adoption within the industry. We will be looking at MCTS (and no doubt Total War: Rome II) in a future post on the site.
I must say I thoroughly enjoyed my time at the conference, given that working more on the academic side (while still dabbling in my game development) there can be frustrations when seeing the (increasing) gaps between research concepts and industry practices: an issue that I don’t feel either party is more to blame for than the other.
It is interesting to see not so much to see how things work on the ‘other side’, but more to sit and talk with developers and listen to the burning issues: what is it that they are dealing with on a daily basis that keeps them tweaking algorithms, tools and systems? I look forward to returning in 2015 and what the conference will bring.
Don’t forget to check out the content that is hosted up at AIGameDev.com, given that they are responsible for hosting and managing the GameAI conference.