“Why most games are dumb”

Dave O’Connor, the boffin behind the smartest, most plausible strategy game AI I’ve ever had the pleasure to pit wits against, has been sharing some of his secrets with students in Canberra. His hour-long lecture sheds light on many of the features that make the Airborne Assault wargames so singular: the micro-management eliminating delegation system (31:10), the ingenious route-finding routines (36:50), the inertia modelling (18:45), the representation of ‘soft’ factors like leader temperaments (25:30)… If mainstream RTS developers adopted just a fraction of these ideas the world of strategy gaming would be a far more interesting place.


  1. Homunculus says:

    I would baselessly posit that AI coding is often the most neglected aspect of development because it’s pretty difficult, time consuming and (most importantly) isn’t perceived as being terribly sexy by marketing types to leverage as Unique Selling Points. It certainly doesn’t lend itself to being as succinctly demonstrated as the latest phong shading or whatever that can be packaged in a simple screenshot or promo video.

    Which is a shame, as when done well it can really elevate the experience a title offers, as can be seen with Dawn of War’s dastardly skirmish AI mod, or Dale’s work on Civ 4, which Firaxis eventually embraced into the game proper with Beyond the Sword.

  2. Nick says:

    I think game AI has suffered also because the current zeitgeist is for multiplayer.

    I surely can’t be the only one who buys a lot of RTS games yet rarely if ever plays them against other people? Maybe I’m weird.

  3. Morte says:

    I was kind of tempted by the Airborne Assault games after reading this, and I went looking for a demo, but I couldn’t find one. Anybody know of one?

  4. Tim says:

    AI is very very unsolved, it requires deep thinking and research. Where as other areas of development are quite well defined (eg: make good art assets). Games seem to rely more on reliable and robust (if still flawed) psuedo AI.

    Most aren’t very adaptable and seem to be centred on balance and scripting rather then any knowledge representation and artificial reasoning.

    I get the impression there’s a certain amount of fear associated with it. What if they invest all that effort learning and developing an AI that is just a bit crap and scripting turns out to work better?

    I like the sound of what this guys been doing. Making AI tracktable. It makes much more sense to seperate AI into specific tasks that need reasoning, rather then tack an “AI opponent” onto a finished game, not that any company would do that right? I wish my AI classes had been better, but I think that about a lot of comp sci classes.

    Comp Sci!! You call yourself a science?! Go back and do more maths, it’s good for you.

  5. Pishtaco says:

    There’s a demo from Red Devils over Arnhem, the first game on this engine, from a while back: link to gamershell.com

  6. Chaim Krause says:

    Unfortunately there is no demo. (The RDoA is very outdated and doesn’t do the current game engine justice. Although it will give you some idea of what it is like to play.) I am a Panther Games fanboi and beta tester, so you can’t trust anything I say, but HTTR and COTA rock. :)

    The above is true, but I have also been involved in wargame development as both a hobby and as a profession for many years, and the AI in HTTR and COTA is some of the best I have seen from any military simulation. That includes the multi-million dollar sims built with DoD dollars.

    You also don’t have to just take my word for it. In 2006 The Wargamer awarded COTA The Readers’ Choice Awards – Game of the Year and The Readers’ Choice Awards – Wargame.

  7. James Sterrett says:

    I’m another Panther fanboy. We may come out of the woodwork on you here. ;)

    In lieu of a demo, you might try the following discussion threads:

    MarkShot’s Mini-guide:

    link to matrixgames.com

    Eddy Sterxe’s 5-Minute Guide:

    link to matrixgames.com

    Discussion on why they do or do not appeal:

    link to matrixgames.com

    MarkShot’s Quick Overview:

    link to wargamer.com

    And, tooting my own horn, my take on them:

    link to matrixgames.com

  8. Adel says:

    I should’ve picked up Highway to the Reich when I saw it for 20 bucks at my local EB. I know Panther Games sells it direct for a higher price, but I can’t convince myself to pay that much for a game that I’ll most likely be overwhelmed by. Panzer General II was nowhere near this detailed and in-depth.

  9. Draco says:

    The main problem behind AI would be solved by developing some special hardware, some kind of AI acceleration card.
    With the hardware companies making profit of it we should be seeing AI bell & whistles and special features en the reviews pretty soon.

  10. Nick says:

    I despise playing RTS games against other people.

  11. cliffski says:

    join the club, unless its someone you know well.

  12. Yhancik says:

    @ Draco :

    As proven by the success of the Physics Special Hardware ;p

    No really, we have multi-core processors, we don’t need special hardware. It’s only a question of priority ; they just need to spend less time/money on on realistic sweat shaders and volumetric blood, and more on AI.
    (If only “most” gamers would stop to ignore games that looks “so 2 years ago”.. *sigh*)

  13. brainwashed says:

    @ Tim – well said.

    @ Draco –

    afraid I have to disagree with you there. An AI acceleration card would only be useful if there were common AI routines applicable across a variety of game types – unfortunately (with the possible exception of pathfinding) this is not the case.

    Afaik, state-of-the-art AI research is nowhere near being able to construct some kind of algorithm with enough adaptability to perform reasonably at a variety of different tasks – and in any case, the vast majority of commercial games do not utilise any of the developments from recent AI research (with some exceptions, such as FEAR’s use of a planner) and instead tend to employ very domain specific bags of tricks, scripting, and plain cheating.

  14. Andrew Armstrong says:

    Heh, definately agree with Yhancik and brainwashed, AI accelleration is not really possible at all…general CPU’s with more cores/faster processing, and better memory access times, will do better wonders for AI. More console memory too would help – a lot of console games don’t do much complex AI since it would require a whole host of the memory to do things like leader temperance or other soft things (although these can be easily faked).

    Definitely getting better, although the fact that most strategy games barely have any available “strategy” options, means the AI is limited by the games own gameplay in many cases – if it ever has “personality” it loses, outright, because all but one strategy wins, even against a notice player.

    Strategy games could be improved a lot, shame a lot needs to be done in real time, and feedback is tough – Sins of a Solar Empire simply does “chat” (which is an improvement over nothing!) but even that is limited, a shame, since games like that do have a few manoeuvres and other things the AI needs to think about.

    Also does “interesting” (ie; more complex) sell anymore to publishers? They’ve always wanted dumbed down RTS games, since consoles here they come! (sigh, just look at what the Civilization series is doing)

  15. Turin Turambar says:

    Ahh… the problem i have with these wargames is not their complexity or their underdeveloped graphics. It’s the controls and/or interface and/or feedback.

    Some of these developers should try to expand their scope a bit and play some modern games, perhaps they could learne then how to integrate a easy and clear interface, fast and easy controls and how to show information and proper feedback to the players without need of artificial methods.

  16. Draco says:

    I just spend 200+ € in a graphics acceleration card and feeling a little cynical, but my English isn’t enough to show that note in my comments.

    None the less, good appreciations from the three of you.

  17. Cigol says:

    Tim Stone should write more often on the site.

  18. Homunculus says:

    Agreed. The secret bonus RPS Stone / Smith third axis receiving play is always a joy.

  19. Dave "Arjuna" O'Connor says:

    Tim “the boffin” Stone,

    Thanks for the publicity mate, but I can do without the boffin monicker. ;)


    It always raises a chuckle in me when I hear people talk about “recent advances in AI”. When I was doing a contract for Defence in 1992 I was introduced to a whole group of scientists ( real boffins replete with PhDs ) who were all working on the latest AI theory. They asked me what my approach was going to be. I told them I am going to build an expert system ( ie rule based ) based around a system that models real world decision making in a hierarchical structure – ie I was going to model reality and eschew abstraction. They all laughed. That group was abandoned long ago as nothing meaningful was produced from it. Defence are now looking at our engine as a concrete way forward.

    I agree that scenario specific scripting is never ever going to provide a challenging AI opponent beyond the first few plays. Once you know the approach/options the AI will take it becomes predictable and easy to beat. Even if a number of conditional options are provided, a scenario specific scripting engine cannot cater for all the possible permeatations, and these are numerous on the battlefield.

    That’s why our approach is to design an AI that is “generic” – ie one that has situational awareness and can assess, plan and react to the situation it confronts regardless of the scenario, terrain, weather, opposing and friendly forces. By and large we have succeeded with this approach. There will always be plenty of room for further improvement, but we have at least built a foundation on which we can blaze the trail.

    I would also like to add that not all wargames can be criticised for their poor user interface ( UI ). We have received high praise for ours. Sure we would like to spruce things up with a nice 3D map etc, but in terms of user functionality I think some of those so called “modern” games could learn a thing or too.

    It is true that we do not have a current demo available for Conquest of the Aegean ( COTA ) – our latest title on sale. However, we will be putting out a demo for our new upcoming title, Battles from the Bulge ( BFTB ). Check out these links:

    COTA website: link to cota.matrixgames.com

    COTA Forum: link to matrixgames.com

    BFTB Forum: link to matrixgames.com

    In particular, the “New Stuff” thread on the BFTB forum is worth a look.

  20. brainwashed says:

    hi Dave,

    cheers for the response – your ideas sound pretty interesting – I’ve been looking for a decent strategy fix lately – pity there isn’t a demo out for COTA, the five minute video was pretty interesting in itself.

  21. Dave "Arjuna" O'Connor says:


    Go ahead. Take the plunge. You won’t be dissappointed. But don’t trust me alone. Check out the many awards won by COTA and the generally outstanding reviews it received.

  22. JeF says:

    Hi Tim,

    Thanks for sharing this with your readers.
    I’m another fan boy and beta tester of the AA engine. And occasionally the creator and maintainer of the website where the presentation is hosted.

    I’m not posting any more links to game stuff : my website tells it all. :-)

    Take care,


  23. Jason says:


    “I should’ve picked up Highway to the Reich when I saw it for 20 bucks at my local EB.”

    I hope Dave doesn’t mind me pointing this out, but you can get Highway to the Reich for $15 at NWS Online. This is the boxed version with the printed manual. Of course, it’ll be close to $20 with the shipping. If you’re worried about being overwhelmed, just set aside some time for the tutorials and you’ll be fine.

  24. Tim Stone says:

    Highway to the Reich is a good place to start if you’re new to the series. It’s more sophisticated than Red Devils Over Arnhem, and more manoeuvre-friendly than Conquest of the Aegean (flatish Dutch terrain vs. craggy Greek mountainsides).

    All the games are surprisingly approachable. Learning the subtleties takes a few months/years, but the delegation system means you can mount competent attacks almost immediately. Just find an HQ, select an order type (probe, attack, move, etc.) and destination, and you’re away. Child’s play.

    @JeF. Thanks for sharing the lecture. Your wargamer.com forum post was what piqued my interest.

    @Dave. It’s great to hear 3D maps are on the agenda! With those and a few polygonal panzer-shaped counters I reckon AA could conquer the world.

  25. Dave "Arjuna" O'Connor says:


    I was in a meeting today with some Defence officials and when I mentioned about an option to do 3D maps they said “Oh No, we don’t want them.” You just can’t please everyone. ;)

  26. Novack says:

    Ping Back from AIGameDev.com

  27. Vercinger says:

    I’m surprised I’ve never heard of these games before. Delegating duties, competent ‘AI’ and smart pathfinding (why the bloody panzer is it so hard for so many studios to get that last one right?) are things I’m very much interested in seeing combined. Definitely will check out the games.

  28. NSGrendel says:

    An astute and insightful look into how the loading of expectation onto players affects the long term commercial success of a game.

    Oh, no, wait. I’m thinking of journalism. This is a fl/puff piece. And lo and behold, it did fill column inches.

    • Tagiri says:

      . . . Welcome to 2008? Why did you even feel a need to comment on this article?