AI Conference Round Up: AAAI 2014

From the 3rd to 7th of October 2014, North Carolina State University played host to the annual Artificial Intelligence for Interactive Digital Entertainment (AIIDE) conference.  This is a big year for the conference as it celebrates 10 years of research focussed on digital entertainment.  Speaking as someone who is attending the conference for the first time, it is interesting to note the differences in culture when compared to the likes of the Computational Intelligence and Games (CIG) conference.  While I would argue that CIG is one of the few that relates heavily to AIIDE, it’s clear when reading their mission statement this is not exactly the case.

Sponsored by the Association for the Advancement of Artificial Intelligence (AAAI), the conference is targeted at both the research and commercial communities, promoting AI research and practice in the context of interactive digital entertainment systems with an emphasis on commercial computer and video games.

The most obvious difference is that there is an explicit emphasis on commercial video game applications and research.  I would argue that CIG (and other conferences) do not share this sentiment – while they may be interested in commercial games, that is not their primary focus.  As such, there was a lot to process as I walked away from the event.  This is due not only to the range of interesting work being explored at AIIDE, but also that previously described mission statement: one that does not represent the community that attend the conference.  I try to summarise my time in a nice paragraph-driven format.

Games Are A Platform

One need only look at the range of research being published to note there is a greater emphasis on using games as a platform for research rather than a benchmark.  At CIG there is often a significant body of work dedicated solely to the application of algorithms in specific games to test their effectiveness.  Certainly the large body of research in Monte Carlo Tree Search presented at CIG supports this.  Even this year we had a session dedicated solely to MCTS applied to game-inspired benchmarks.  However, this is merely a reflection of the current state of computational intelligence research.  One need only peruse the conference proceedings of earlier years to see a distinct focus on neuro-evolution, temporal difference learning and Monte Carlo methods. This is largely the ethos of the CIG conference, to apply computational intelligence algorithms in a variety of game (or game-like) domains.  This is in many respects what CIG is interested in and the appropriate location to publish such research.

AIIDE feels like it has a different philosophy: where researchers are engaging in research that uses games to achieve something.  In some senses this can be to do something interesting that, as a system, games can support – though is not a reflection of the conferences mission statement.  This is indicative by the number of researchers interested in computational creativity that present at this event and the body of work at the EXAG workshop.

The EXAG Workshop

The first couple of days of the conference are focussed on workshops. one of which was the Experimental AI in Games (EXAG) workshop hosted by Alex Zook of Georgia Tech (pictured above) and Mike Cook of Goldsmiths in London.  The emphasis is expressed in the title and I don’t think it need be stated any clearer.  I attended the whole day of talks (and if you look towards the end, you may even see me presenting a short paper) by researchers, Ph.D. candidates and others discussing their work.  What was equally fun and exciting about this workshop was that not only was the work displayed rather varied, but the mood was very relaxed and encouraged discussion of work that either is still in its infancy or outright at the idea stage.  For that I am grateful given that the paper I published with Michele Vinciguerra had only started two or three months from the time of writing.  What made EXAG so engaging was the creation of a platform that allowed for healthy discussion without the fear that our work is not legitimate or lacks merit: a rather prevalent issue in the academic world, given the need for impact factor of published work at conference or journal level.  This is something that will prove immensely useful for doctoral candidates whose work may not conform to ‘norms’ established by the community: despite the fact that AIIDE Doctoral Consortium took place later in the week.  I look forward to seeing EXAG return in the future and I hope to return with more work to discuss.

Industry Applications

During the fourth keynote of the conference by Squirrel Eiserloh, a significant debate opened on the validity and applications of games-based research for industry.  As mentioned earlier, this conference has now been running for 10 years and the arguments over whether games-based research should be focussed on ‘solving’ industry problems have been raised before.

It caused a bit of a stir and did lead to lengthy discussion once more. Personally, I feel that games-based research does not necessarily need to lead towards industry application.  Naturally, there are a lot of great examples where research has been applied in commercial titles both AAA and Indie.  In addition, there are researchers who want to have impact in the commercial sector through their research.  Regular visitors will know that both parties are largely is one of the main reasons why this blog was started.  However, despite this, there is no problem with research that simply embraces games as a means to an end and gains something from that experience.   Will that new knowledge ultimately inform the games industry?  Perhaps in time it will.  You never know what research ideas will perpetuate into commercial applications (and not just in games).

My impression was that the discussion raised frustrations that had been dormant (perhaps since a keynote in a previous year). While research in games has only really flourished in the last decade, an image has been established in some circles as the community being a gathering of naive and enthusiastic researchers, complete with a list of publications aimed at ‘solving’ the games industry’s problems. These ‘solutions’ often being unfit for purpose given the requirements of the AAA sector.     This does not reflect how this community works, given work often can be expressed in a number of ways:

  1. Many researchers are more interested in the ‘games as a benchmark’ philosophy, as mentioned previously.  Games provide interesting problems that are interesting, challenging and scalable.  Often providing a nice location to conduct research that – in real world domains – would prove too expensive.
  2. A lot of research is still in its infancy or lacks the scalability and robustness to become a staple of commercial products.   This is not to say that work is poor quality or insignificant: this is how research works!  It takes time!  For people working in these areas, games are an interesting application to experiment with. It allows researchers to assess how well their work operates in this context as well as learn of new routes to take their work.
  3. Research that achieves breakthroughs in the AAA industry is often carefully crafted through strong relationships.   Consultation is made between academia and industry and this needs to respect each others unique environments, markets and considerations.
  4. Conversely, this can often happen without any real involvement from academia, as established research is considered as a viable alternative that has not yet been considered in industry (e.g. HTN planning).
  5. It’s not all about AAA: work can be achieved in the indie sector: either by making connections within the community or making games yourself that work as part of research.

Is it perhaps time that the mission statement be a reflection not of the original intent, but the community’s interests?

A compounding effect is that the community itself has changed and grown tremendously in 10 years.  To the point that the AIIDE mission statement, which implies research published is aimed at industry application, seems out of date and inappropriate. AIIDE’s successes are driven by her community of talented and passionate researchers.  Is it perhaps time that the mission statement be a reflection not of the original intent, but the community’s interests? Sure, there are risks associated with such action, given that search interests and trends fluctuate over time. However, when the attendees of a conference no longer associate themselves with its mission statement, yet are publishing work that is deemed relevant by its own community (i.e., peer reviewed), something is clearly amiss.

For more on this discussion, I would point you to fellow researchers Mike Cook and Julian Togelius who have published their own feelings on this topic. Both of these are well worth a read.

Oh and Let’s Not Forget!

This post could easily become very lengthy: there is a lot of really exciting work happening all over the world that was on display.  I figured I would write a couple of lines to summarise some of the exciting and interesting work I saw.

Phil Lopes (Malta) and Chong-U Lim (MIT).
Phil Lopes (Malta) and Chong-U Lim (MIT).
  • The AIIDE 14 Doctoral Symposium had some really interesting work from PhD candidates.  Notably Phil Lopes and Chong-U Lim shown above.  Phil is discussing community-powered creativity and how to harness that type of information.  Meanwhile Chong-U is looking at how to harness player-data to see if correlations can be made in social media behaviour or player aesthetic customisation.  I look forward to what they bring in the future.
  • Doctoral Consortium organiser Noor Shaker presented work on authoring tools for platforming levels that looks like merely the beginning of another branch of engaging work.  Noor is one of the best and most active researchers in PCG for platforming games.  If you have never heard of Noor and her work, I encourage you not just to read this paper, but pour through her entire publication history.
  • Eric Jacopin presented a summary of his research on plan quality when systems such as GOAP and HTN planning were adopted in AAA games.  This is a topic that I have discussed previously on the site.  It was a pleasure to meet Eric in person and discuss his work at length.
  • Peter Ingebretson from Maxis delivered an interesting keynote on the improvements made to the AI of The Sims between the third and fourth iteration of the series.   Notably discussing how previously conducting multiple actions required linear execution.  This talk went into detail about how the constraint-guided search and improvements in animation and agent control allowed for multiple actions to be executed simulatenously (e.g., drinking milk, eating cake and watching TV at the same time).  In addition, there are a number of spatial considerations that have to be factored when executing these behaviours.

Like I say, there is a lot of really exciting research presented that you need to be reading.  The full proceedings of the conference are freely available online and I encourage you to see what catches your imagination.  I am grateful for my time at the conference and to meet friends old and new.  I look forward to attending next year (better start lining up my next EXAG paper).

Enjoying AI and Games? Please support me on Patreon!
Tommy Thompson Written by:

Tommy is the writer and producer of AI and Games. He's a senior lecturer in computer science and researcher in artificial intelligence with applications in video games. He's also an indie video game developer with Table Flip Games. Because y'know... fella's gotta keep himself busy.