Strategy Games: The Next Move

This being the title of yesterday’s GDC panel featuring Tom Chick, Soren Johnson, Dustin Browder, Ian Fischer, and Jon Shafer, of which Gamasutra just posted a summary. Unfortunately it’s very brief and I can’t really discern much here, other than everyone making different points and disagreeing with everyone else. Does someone have a complete transcript or recording of the panel?

"I don’t think it makes financial sense to make great AI…


I suspect he’s right. I mean, if it was a real selling point then there’d only be one real success title every 2-3 years - if that. Even great strategy game AI isn’t all that impressive, or different or better-feeling than the norm.

Chris, I know that GDC posts videos of the presentations, but I think you have to log in as an attendee. I’m not sure how open they are to the public. As for a transcript, I’m not sure how that works.

I feel bad because I didn’t mean to ambush Jon and I think it came across that way. But he made a point that needed to be made. I’ve complained about the AI in many strategy games, most recently Civ 5 and DOW2: Retribution, but the fact that it doesn’t bother many players – and that many reviewers apparently don’t even notice! – makes me wonder if developers need to bother. It seems they don’t. From a business perspective, if people are going to play strategy games as puzzle/RPG/city builders (Civ5), or action-RPG campaigns/multiplayer games (DOW2: Retribution), then it makes sense not to fuss with something that only a few of us are going to care about. It sucks, and I’ll continue to bitch about it, but I understand.

And I was glad that Dustin Browder jumped in at that point, because I think Starcraft II has a very good AI, as have Blizzard’s RTSs all along. However, it requires a very different kind of work to make a good AI for an RTS and it’s easy for someone at Blizzard to say “nah, we can afford the resources!”.

Of course, it’s also a matter of design. Understanding the capabilities and limitations of your AI are an important part of the design process, and I suspect that’s part of why Dawn of War II uses that silly action-RPG corridor crawl where a traditional skirmish-based campaign usually goes. You don’t need as much AI if you’re just activating little clots of enemies at discrete script triggers.

Anyway, I hope Jon doesn’t get dinged for making an honest and understandable observation about how games are made. He’s right. And I’m glad he’s at Stardock, where Brad Wardell seems to care about guys like me enough that he’ll give Jon the resources he needs. :)


And Shafer now works at Stardock. They might want to look into laying off whoever codes the AI since that job isn’t very important.

Okay, I found the GDC Vault site that hosts all the conference recordings, but none yet from GDC 2011. The list of free videos from GDC 2010 is very small, and the full set is indeed limited to members only. That’s too bad, I wish they would allow outsiders to pay a few bucks to watch one video.

This. In some strategy games as you have said, the AI is not as important. They have a campaign/story/immersion, or they are about asymetric confontation, or they are strategy/puzzle games. In others, where the quid of the game is direct confrontation against a similar oponent, it’s much more important. In the latter type, i agree with you, without good ai there is no game. But the important thing here is that using generalizations is not a good idea. The strategy genre is very broad, very heterogeneous, and in some games the ai will be essential, in other there won’t be any ai by the nature of the game, and in between there will be every type of stance you can imagine about AI between these two extremes.

Civilization is a hybrid case. In theory, it’s about confrontation against oponnents who have the same start in resources to see who wins. In practice, lots of people play Civ a a big tycoon game, or to direct their own civilization in a more “immersive” just for the thrill of the alternative history of mankind.

Speaking of Starcraft 2, has anyone heard anything about the release schedule for the other two campaigns? Anything? Are we talking about 3-4 years between each release of SC2?

Regarding strategy AI, I have to agree with Tom. Only the hardest of hardcore players care about it. Everyone else plays strategy games like big puzzle games.

With all due respect to Mr. Tom Curtis, I don’t think he characterized my position very well. Doesn’t help that my comment on the article doesn’t seem to have gone through…

I don’t think good AI is evil or a bad thing. Simply that, in most cases, making a GREAT AI instead of just a good AI is not something which will help you sell more copies. It’s much, much easier for publishers to point to cut scenes or great graphics or celebrity voices or whatever else as having a measurable impact on sales. And as long as people AREN’T judging game AI harshly, that won’t change. So I guess what I’m saying is that people SHOULD make a bigger deal out of bad AI, so that this problem goes away.

One of the reasons I’ve joined Stardock is because Brad makes games first because he enjoys it, and second because he can make money doing it. He enjoys making writing AI, so that’s what he does. I’m also in this business to enjoy it and make games people enjoy playing, so coming to Stardock is definitely exciting for me. Does the company’s games philosophy make good business sense? Nah, not really, but making games for the middle market allows us to do that and still be profitable (which was really what I was hoping the takeaway from my rants would be!).


The question isn’t so much whether gamers ding you for bad AI, the question is whether your gamers ding you for bad AI. I’d say Stardock gamers tend to expect a great AI (and don’t care about multiplayer at all if the AI is great).

So it may not be good business sense in general, but it is good business sense for you guys. (and maybe very few others)

I know that I didn’t buy Star Ruler when it was on sale last week on Impulse, because of hearing about questionable AI. Civ V I didn’t buy initially due to Steamworks, but ended up not buying it period due to the AI.

I suspect you’re wrong about that. My impression is that what most strategy gamers want is AI that can challenge them in any way. Where hardcore RTS gamers wants to be surprised & occasionally humiliated by the AI like we are by strangers in public… I mean, like we are online by fellow humans, the majority of RTS gamers are perfectly happy to lose because their APM can’t keep up with a team of AIs trying to drown them in a sea of bad decisions, deadly only because of it’s enormous volume.

Of course, games like Civ are big puzzle games. On harder difficulties you win by manually optimising every single thing you control, every single turn. Beyond having figured out a build order, strategy doesn’t really enter into it. And in terms of strategy, that’s really no different that formulating a strategy for beating a TD map in the TD game of your choice - which again is more puzzle than strategy.

Was hoping for a bit more time to respond to this during the panel, but it’s cool since we had a lot to go over.

Writing AI for an RTS is definitely different from doing so for a TBS game. In an RTS, time is on the side of the AI programmer (since the computer can consider a lot of precise possibilities simultaneously, which is really hard for a human), whereas time works AGAINST a developer in a TBS game (“why do these turns take so long? Geeze…”). The AI simply can’t use as much time to think as a human does. So you’re already forced to take shortcuts.

As you note, design definitely plays a huge role as well. In an RTS like Starcraft, if you build a big enough force and send it at the enemy when he’s not looking, you can wipe him out without a lot of trouble. You might be able to write the logic for something like this in, I dunno, a few hundred lines. It may not be great, but it’s at least a viable threat. The closer you get to games like chess or go the tougher it gets. A few hundred lines of logic in chess is going to give you an AI that gets beat by a human the very first time they’ve played the game.


Yes, it definitely varies from game to game and company to company. Stardock has built its brand on good AI, so it’s expected we’ll deliver on that. There aren’t really any other companies I can think of for which that’s the case, aside from maybe a couple of really small indies. I’m just speaking in broad generalizations, because it’s more fun that way.


One thing that GC2 did which I liked in that regard was the “allow max AI” option- so that the AI can take the time it needs , if it needs it, but it’s optional.

I do think strategy games should be designed to be AI-friendly whenever practical. There are things in Elemental that seem to be very, very AI-hostile, such as the sovereign death rule.

At the very least, they need to make sure that the person who writes the AI isn’t involved in game development anymore.

Jon, I appreciate and understand the issue that most people don’t really care that much about whether a game has a great AI that provide an almost human-like challenge. A lot of the people who buy and enjoy a game like Civ 5 struggle to win even if the computer isn’t blatantly cheating. To be fair though, at times it honestly does feel like the AI in that game will simply build up a big force and send it at you when you’re not looking. I think that’s still a credible threat even in TBS.

I guess my concern revolves around what you consider to be a good AI. In Civ 5 there were significant game features that the AI practically didn’t participate in or did so in such a poor fashion that it ought not have bothered - notably the use of naval and air power. The AI could still be a challenge for a lot of players, but is it still “good” merely if it provides a challenge in general or should that moniker require that the AI provide a challenge across all aspects of the game design?

To me it’s like if you had a high-level chess AI, except that the only thing it ever did with rooks was to castle. I wouldn’t consider that to be a good AI even though I’m sure it could beat a lot of people.

Not only does it not make sense financially, great AI has never really been done before – no Civ-like strategy game ever made has had AI that can come close to challenging even a merely competent player on anything close to an even footing.

Game companies really can’t afford to invest overly into such bleeding edge research, as nobody really knows just how feasible it is, and the return on investment is speculative at best. The extent to which they do (and they do!) is out of a sense of craftsmanship, love for what they do, and aspiration for something better – not for any financial gain.

I’m doing this right now and it’s a huge fucking pain in the ass – non-developers frequently have no grasp of just how brain-busting hard it is. I’m spending a bunch of time on my AI because I want it to be “good”, yet I will be forced to stop far before I can hack out “greatness”, and am still stuck with a gussied up heuristic hodge-podge that I know will inevitably have game-able flaws. From a purely financial perspective this is a mistake, a money pit.

The best way to “improve your AI” isn’t to make it play better anyway, but to address the psychology behind people’s assessment of AI. Make it more personable, etc. Civ 5’s problem isn’t so much that it plays badly, but that it makes a few glaring mistakes that even beginner’s don’t make – e.g. with naval transports.

I tend to agree with Jon’s position here, and having been running a team with a tad bit smaller resources that either Stardock or Firaxis its hard to cost justify development of AI past the “somewhat challenging” level. Games want to win in their games, not get defeated.
We still put 25-30% of coding time on AI, and I doubt we’ll ever go lower…

This is probably the most important lesson I’ve learned in the last decade. It doesnt matter how cool a feature sounds, and how fun it is, if the AI can not handle it.

Beware that Chess analogy, because Chess is easy to hack out!

I would say instead that the further you get from a simple game like Chess, the more exponentially difficult writing a good AI gets. Chess has the luxuries of perfect information of the present, completely non-random outcomes, only 1 move at a time, and relatively few possible moves. A game like civilization is vastly more complex.

Yeah, Chess is not a great example of hard AI. I bet you could beat 97% of the world at Chess with a hundred lines of code, given computer speeds now. Even Go is reasonably tractable – it’s ‘hard’ compared to other board games, but it’s still easy to represent internally and using probabilistic evaluation functions you can at least play decently. Tougher games are ones that you aren’t even sure how to represent it with a data structure.