An Interview with Luc Lussier
The Artificial Intelligence Designer/Programmer of Defcon

Victor Feltes:     Of all the people at Anonymes Informatique [now GolemLabs], Defcon fans know the least about you.  Tell us about yourself.

Luc Lussier:    Well, I guess it's because I'm the least present of the Anonymes Informatique family members that less is said about me.  But for those interested, here are some details.  I'm still an undergraduate student of computer engineering at the University of Sherbrooke having one last semester to go before moving to something else.  My current areas of interests are artificial intelligence, fuzzy logic and dynamic FPGA configuration.  Because the University of Sherbrooke has an internship program, I already have some working experience, the latest being for CAE, a Montreal based company building flight simulator for airline companies all around the world.
 

I understand that you are currently in Japan.  What has brought you there?

       As far as I can remember, I always had some fascination for Japan and it's culture.  I really wanted to work and live in Japan because just visiting for a few days is not good enough when it gets to know a new culture.  So when I found out about this internship program called "Co-op Japan" in which I could participate, I just had to give it a good try.  That's when I started learning Japanese!  Not the easiest task to undertake.  After studying for more than a year, I still can't talk much...  And here I am, working at ATR, an international research center located in the Kansai area, between Kyoto, Osaka and Nara.  Here, I'm working in a speech recognition group, an area that is quite new for me.  In my department are also some big names in artificial life and intelligence like Thomas Ray and Hugo de Garis.  It's a really stimulating environment, I feel like waking up in the morning just to learn more each day (weekdays that is, don't expect too much)!
 

How did you come to work at Anonymes Informatique?

    Coincidence?  Good fortune?  Actually, Jean-René is a good friend of one of my cousin.  For some times I had heard about this game called Defcon and how good and great it was and all and everything, you know...  And one day, my cousin told me Jean-René was looking for someone to help him with the AI of Defcon.  Since I'm quite interested in AI, he suggested I get in touch with the Anonymes Informatique team and get introduced to this game to be.  Wow, now that was a quite impressive!  Such a huge amount of possible actions, a great interface and some real facts under the cover.  What a game!  That's when I decided to join.
 

Tell us about Defcon's Artificial Intelligence.  How is it better than others we have seen?

     When I first started talking with Jean-René about the AI in order to understand what kind of behavior he wanted the computer to generate, we often came across the expression "more humane".  So the Defcon's AI as for main goal to make the computer controlled countries in the game act in a human like fashion as much as possible.

       But just what is being more humane for an AI?  To define such a complex behavior, we have set some basic guidelines.  First of all, humans make mistakes.  Now this does not mean that once in a while we are just going to make the AI do a stupid random mistake!  No, most humans (or so I believe) do make mistakes for various reasons even if they think they are actually doing the right thing. Secondly, humans usually posses only a limited knowledge.  But some peoples do know more then others! This means that the AI does not know about how its actions really affect the game.  It can only rely, like the player, on knowing that performing this action bring those kinds of results.  Finally, humans can adapt, that's how we learn.  If you can't solve a problem one way, you'll eventually try to solve it differently and so will Defcon's AI.

       This way of doing things is part of one school of thinking related to the AI.  Some peoples think the computer should know every possible solution to every possible problem the AI will be faced with.  Unfortunately, this leads to endless if A then B reasoning and this is but the tip of the problem!  What if those who made the program forgot to include a specific A or implemented B as a poor solution to the A problem?  You end up with an AI which has some flaws that will sooner of later be found and exploited!  Now this is not the way I believe things should be done.  Sure, I gave a reason, which should be good enough already.  But more then that, this kind of implementation is deeply rooted to a specific problem.

       What I prefer is a general approach.  Defcon's AI uses a problem solving approach just like us.  If you are hungry for example, you will find a way to get something to eat.  How are you going to do it?  This is very individual dependent, you might go to buy some food at a grocery store or go to a restaurant, but what kind of restaurant, cheap or expensive?  If you have no money, you might have to beg for some food.  Again, this is very individual specific and so are most problems encountered in Defcon.  Different countries have to solve the same problem in a different way!  Do I know how?  Not at all!  Does Jean-René know it?  Maybe, maybe not?  Who knows?

       So because I just can't go to work on Monday morning and say, "Ok, let's program each countries specific reaction to a lack of energy resource", a different approach has to be taken.  That's why each country is controlled by a unique AI entity that will learn, evolve by acquiring experience and learn from it's mistakes.

       Now as to whether or not and how it's better than others we have seen?  This is a good question!  I don't believe a neither black nor white answer can be given to this question.  As I almost consider Defcon's AI as a scientific AI prototype experiment, I would say that I lack the context to really compare it to others.  The reason is that you must either apply the Evolutive Human Emulator (EHE), that's how we call Defcon's AI, to a similar task that a previous AI was used for or try using other AI to pull the strings inside Defcon to really make any comparison.  Two different experiments we have not really performed.

       At first glance, I would say that the EHE could take on the challenge of previous AI tasks and that most other AI implementation would have a hard time playing Defcon.  But just who am I to say such thing?!?  I do believe that the EHE's three basic guidelines set it apart from the others.  As for any bad sides, there's always a bad side, Defcon might not be a graphic intensive game, but it sure is a computation intensive one!  Remember that allusion to a scientific experiment?  The EHE does process a huge amount of information.  Future versions of the AI could most probably make a good use of multi-processor computer architecture to speed up things.  But this is an other story!
 

Are all the computer's decisions made through the EHE?  For instance, what about at the WarArt aspect of the game?  Will the computer test and apply unit tactics, or is that hardwired into the program?

       This is actually a very good question.  Unfortunately, the only answer I can give you is that I don't know!  Since I left Canada for my internship, a lot of things probably changed in Defcon planning and architecture.  When I left, the EHE was not used yet to control the units movement or any other aspect of the battle simulation.  Anyway, the EHE could certainly not be used alone to control the units.  Some hybrid approach would probably be better.  But then again, I didn't really look at the question.  It could help to chose which strategy to use but some path finding algorithm would still control the unit's movement.

(Post Script:  The Warart A.I. is currently under development, but it will probabily use a simplified EHE.)

Is there anything else you would like to add about the creation and future of Defcon?

       What I like the most about Defcon is that the game is some kind of platform that will eventually let players customize the game to their taste.  The wide range of actions and the amount of details also impress me a lot.  On the humane aspect, I really think that Defcon is some kind of adventure for it's creators and the peoples around them.  Before I joined the Anonymes Informatique team, a lot had already been made on Defcon but the completion of the game was only a far away goal for us.  As I write this, Defcon is still a work in progress but if I judge by what I have seen of it, I'm quite sure that simulation game fans will certainly like it!

       Thank you very much.

      Thank you Luc.  I know I'm not alone when I say I'm eager to put your brainchild to the test.