GDC 2011: Strategy Games - the next move

Animal crossing was actually a pretty fun and highly polished game, anyone remember Mr Resetti? :D
 
Toyota can atest to what happens when a company switches its focus from making the best product possible to making the most money possible.

I don't know about that. Toyota is still the best car company on the planet. And, to the extent they developed some problems, they were mostly in the area of complacency.

Dell would be a better example of a company that has suffered (to some extent) from a focus on controlling costs rather than making great products.
 
Animal crossing was actually a pretty fun and highly polished game, anyone remember Mr Resetti? :D

Yes, that's my point. So is Farmville. But they aren't strategy games. If people aren't interested in making strategy games, and all of the things that come with that (which means either a decent AI, or a game design that makes it interesting to play strategically without a great AI), they should just go make games in one of those other genres that can be very successful and don't require that.
 
Whoever allocated that extra money into the intro video that we all skip was really a forward thinker. In the meantime I play an AI that leads with archers and drops knights in a lake.
 
Yes, that's my point. So is Farmville. But they aren't strategy games. If people aren't interested in making strategy games, and all of the things that come with that (which means either a decent AI, or a game design that makes it interesting to play strategically without a great AI), they should just go make games in one of those other genres that can be very successful and don't require that.
still the challenge must come from somewhere(something).

in general there are 3 types of opponents(challenge-providers):
1) AI controlled(aka civ. other nations are AI-controlled)
2) other humans(aka MP civ)
3) yourself (tetris, minesweeper, etc.)

relating to civ 3) is not possible. therefore the devs must either provide a decent AI or a decent MP.
 
Yes, that's my point. So is Farmville. But they aren't strategy games. If people aren't interested in making strategy games, and all of the things that come with that (which means either a decent AI, or a game design that makes it interesting to play strategically without a great AI), they should just go make games in one of those other genres that can be very successful and don't require that.

Oh i agree, it may not be a strategy title but despite appearances it is a surprisingly deep, long lasting game with huge polish and attention to detail, i'm just thinking that to make a similar game in that genre that could even approach animal crossing's quality may not be all that easy either, it's a bit of a gem.
 
Before anyone blames Jon or Firaxis for that attitude, they need to examine their own habits. Games are the feature bloated messes they are because that is what gamers ask for.

OH OH OH! I know the perfect reply to this!

Ben "Yathzee" Croshaw said:
Fans are clingy complaining diphorsehockys who will never ever be grateful for any concession you make. The moment you shut out their shrill, tremulous voices the happier you will be for it. Incidentally, why not buy a Zero Punctuation t-shirt?
 
I find it wonderful that people with no AI experience say these things are trivial or just the matter of hiring the right people. They aren't. If they were, we'd have nothing to research.

I love it when someone cites, say... XCOM as an example of wonderful AI. It just proves my point. XCOM aliens aren't a threat because their AI is great, it's because the aliens get such an absurd number of resources and advantages on their side it's amazing they ever lose. If a human had what the aliens had on a terror mission, they wouldn't just win, they'd make XCOM look like the AI on settler.

Yes, yes, we've got algorithms to solve, say... TSP. Okay, great, I can completely solve a 200-300 node instance of TSP in a reasonable amount of time (few seconds, because that's all most players will tolerate because you're going to be doing a lot of these searches). But what's that ACTUALLY get me?

Comparison time, folks. Consider SOP. That's just TSP, but you put constraints on the order you visit cities in (basically, you must visit city X before Y). State of the art in SOP? Over 30, you're starting to look at instances where you're taking too long. We don't even know if we've got the best solution on some 40-instances (though we get close). All I've done is add prerequisites and I've slashed the size of the viable problem to 1/10th. Take an actual, complex game problem? gg, gentlemen.

Also were mentioned were minimax trees. Minimax trees are great at chess. But they utterly fail at, say, Go. And Go is a very simple game. Why? (hint: it isn't the branching factor, you can cheat a lot with what node you expand next to get around that) Minimax sucks at understanding formations. It also fails horribly in chess working around fortresses. Attempting to apply minimax to some formation sensitive like 1UPT is just ludicrous. Also, the state of the art in a formation-dependent game like Go essentially boiling down to simulating a lot of random moves. But it still requires wonderful things like well-tuned evaluation functions.

I mean, yeah, the game industry is miles behind in AI. I mean, a few years ago they touted A* as the holy grail for pathfinding and we've known about A* since around the time the first things you could consider video games came out. But that doesn't mean the problems they're trying to solve are trivial or anything.

Consider the state of the art. The best AIs for say, Brood War? (that's the new thing nowadays since someone cracked the API) Scripts. Very, very, very good scripts, but scripts nonetheless. No algorithmic fanciness or anything (okay, the good ones have some small amount of algorithmic fanciness, but all the legwork is in the script). Now, do you hire someone to spend hundreds of manhours developing a finely tuned script they have to toss out every time you make a balance change or do you crank up the little variable that says "let the AI cheat more"? Guess which is easier.
 
I find it wonderful that people with no AI experience say these things are trivial or just the matter of hiring the right people. They aren't. If they were, we'd have nothing to research.

I didn't say it was trivial, I said the programmers could have done a good job if allowed. Current AI is much stupidier than it should have been.

I love it when someone cites, say... XCOM as an example of wonderful AI. It just proves my point. XCOM aliens aren't a threat because their AI is great, it's because the aliens get such an absurd number of resources and advantages on their side it's amazing they ever lose. If a human had what the aliens had on a terror mission, they wouldn't just win, they'd make XCOM look like the AI on settler.

It is interesting that from a list of games you picked XCOM. However, I did not say it has a "wonderful" AI I said that compared to what Civ 5 has even that game had a vastly better AI, especially if you take in account hardware and budget resources. Plus, the AI at these games at normal level do not possess huge advantages.

[snip]

Also, the state of the art in a formation-dependent game like Go essentially boiling down to simulating a lot of random moves. But it still requires wonderful things like well-tuned evaluation functions.

You must be joking! Well-tuned evaluation functions is the number one thing to do in TBS. Better to say perfectly tuned evaluation functions. If you say this is something special or extra difficult, I am starting to doubt about what you think you know about AI. It looks like you have some AI-related "math" dictionary, but you didn't think very much before writting the above sentence.
If you cannot write good eveluation functions, you cannot write a strategy game. As simple as that. It is not an excuse, it is a disgrace!

I mean, yeah, the game industry is miles behind in AI. I mean, a few years ago they touted A* as the holy grail for pathfinding and we've known about A* since around the time the first things you could consider video games came out. But that doesn't mean the problems they're trying to solve are trivial or anything.

If you consider the depth of time passed and the amount of progress done, the world have progressed from i386 to Hyperthreaded Quad-core 3Ghz+ CPUs, manufacturers produced self-moving Robots to do housekeeping and other tasks(a very much NON TRIVIAL AI, here. I hope you agree, since human lives could be at stake!), and something as simple(yes, although you seem to want to make it look more difficult than it actually is) as programming the AI for a strategy game has remained, relatively, in the ancient era!!

Consider the state of the art. The best AIs for say, Brood War? (that's the new thing nowadays since someone cracked the API) Scripts. Very, very, very good scripts, but scripts nonetheless. No algorithmic fanciness or anything (okay, the good ones have some small amount of algorithmic fanciness, but all the legwork is in the script). Now, do you hire someone to spend hundreds of manhours developing a finely tuned script they have to toss out every time you make a balance change or do you crank up the little variable that says "let the AI cheat more"? Guess which is easier.

I do not think so. I say that what you have to do is program those evaluation functions properly. As simple as that. Understand your game, and teach the CPU play well. Decission making is the core of modern programming. They use it from embeded microprocessor programming to Robot programming. It is not as difficult as you claim, sorry! ;)
 
in general there are 3 types of opponents(challenge-providers):
1) AI controlled(aka civ. other nations are AI-controlled)
2) other humans(aka MP civ)
3) yourself (tetris, minesweeper, etc.)

relating to civ 3) is not possible. therefore the devs must either provide a decent AI or a decent MP.

I guess I fundamentally disagree that #3 is not possible. In fact, #3 is where most of the challenge in most successful strategy games comes from. In Civ4, the challenge isn't primarily generated by the AI (even a novice can play better than the AI with no handicaps); the challenge is generated by the asymmetry built into the game (i.e., the AI gets advantages over the human player). Certainly, in a strategy game like Panzer General, almost none of the challenge comes from the AI, which basically just sits there and waits for you to try to kill its units.

A starting point for design of any strategy game has to be that the human is going to be better at strategy than the computer. That means the computer is going to have to have other advantages over the human, to make the game interesting. This is easier to do in games like Panzer General, that use preconstructed scenarios. It's harder to do in games like Civilization, which are supposed to provide the illusion of competing against a roughly equal opponent. The fundamental problem with ideas like 1upt is that humans are much better at dealing with them than computer AIs, so they increase the magnitude of the advantage that you have to give to the AIs, which was already a problem.
 
It is interesting that from a list of games you picked XCOM. However, I did not say it has a "wonderful" AI I said that compared to what Civ 5 has even that game had a vastly better AI, especially if you take in account hardware and budget resources. Plus, the AI at these games at normal level do not possess huge advantages.

You're just wrong. XCOM has hardly any AI, and the scenarios are designed with truly gigantic advantages for the aliens. The only thing that makes it remotely playable despite the total lack of AI competence is that huge asymmetry. Civ5 has way, way, way more invested in its AI than XCOM ever did.
 
Oh i agree, it may not be a strategy title but despite appearances it is a surprisingly deep, long lasting game with huge polish and attention to detail, i'm just thinking that to make a similar game in that genre that could even approach animal crossing's quality may not be all that easy either, it's a bit of a gem.

I don't understand your point. I never said anything negative about the quality of Farmville or Animal Crossing. I just used them as examples of successful non-strategy games that can appeal widely to players regardless of how good or bad they are at playing them. Of course they are well designed and executed, that is exactly my point.
 
You're just wrong. XCOM has hardly any AI, and the scenarios are designed with truly gigantic advantages for the aliens. The only thing that makes it remotely playable despite the total lack of AI competence is that huge asymmetry. Civ5 has way, way, way more invested in its AI than XCOM ever did.

Exactly, my friend! XCOM has hardly any AI! But what it matters is that it looks like the Aliens are smart when tracking down targets and they had included other aspects that made the game competent.
The point is not how much you invest, but how smart you do the programming!
However, I would say it would be better if we do not focus the discussion on an XCOM vs Civ5 AI comparison. They are different games and I include it in a list with other games just to illustrate a point. Let's, better, focus on the point.
 
Exactly, my friend! XCOM has hardly any AI! But what it matters is that it looks like the Aliens are smart when tracking down targets and they had included other aspects that made the game competent.
The point is not how much you invest, but how smart you do the programming!

No, the point is that XCOM is an asymmetrical game with asymmetrical forces, and so it's easy to make the scenarios interesting and challenging even with no AI to speak of. You can't do that in a symmetrical game like Civilization, because it means giving the enemy civilizations huge handicaps, which players don't like, plus when you require huge handicaps it becomes very, very difficult to make those handicaps present the right level of difficulty all the way from the early game to the late game. XCOM doesn't have that problem, either, because the early-game scenarios and the late-game scenarios can be independently designed and balanced.

I personally would enjoy playing an asymmetrical version of Civilization, which simply gives up on the idea that you are contending against other civilizations that start and develop in the same way that yours does. You could build a scenario-based version of Civilization, that would be asymmetrical like XCOM, in that as the game goes on you would come up against tougher and tougher foes, who are more and more advanced, but not because they built up as you did but just because the designer preconstructed them to create the correct level of difficulty. But I think this game would never have the broad appeal of Civilization, because it doesn't have the immersive element that Civ does.
 
[snip]
I personally would enjoy playing an asymmetrical version of Civilization, which simply gives up on the idea that you are contending against other civilizations that start and develop in the same way that yours does. You could build a scenario-based version of Civilization, that would be asymmetrical like XCOM, in that as the game goes on you would come up against tougher and tougher foes, who are more and more advanced, but not because they built up as you did but just because the designer preconstructed them to create the correct level of difficulty. But I think this game would never have the broad appeal of Civilization, because it doesn't have the immersive element that Civ does.

I would enjoy playing a version of Civilization that has the AI properly evaluate combat and production challenges. As simple as that. I hope they find someone that can do just that, it is not so hard, I know a few modders in these forums that could do it in a, relatively short amount of time(I said relatively, do not take on that and start claiming that I am irrational, again!:cool:).
 
I still think 1UPT is an exponentially hopeless problem. Panzer general covered up with specific scenarios and massive bonuses.

I want to give another example: the Total War Series, where friendly units passing each other do not take up space, resulting in ridiculous tactics like squishing 10 knights into the space of 1 pikemen and then breaking the enemy lines. Why? Because AI's could not handle 1UPT. If this rule did not exist, the AI will literally be walking units in circles forever if they ever need to move one unit back and another in front.
 
I see that there is a lot of confusion about how much money should someone "pour" into programming a "great" AI...
I have been playing strategy games for years and I can say that the first ones I played (Sid Meier's Colonization, Empire II, XCOM-TFTD, XCOM-UFO Defence, Fantasy General, Panzer General, Steel Panther etc) had a vastly better AI, while they had to cope with much lower budgets and much lower computer resources.
THIS IS SIMPLY NOT TRUE. I played all those games extensively, and still play a few of them, and in all of them the AI is benefiting from massive statistical bonuses, setup bonuses of much more material, and setup bonuses with positioning on the set maps. It is completely not comparable to a game like CiV. Sure I could set up a hard scenario in CiV, that doesn't mean its a strong AI. If you played a random map on fantasy general with EVEN settings, it is pretty much impossible to lose.


I agree with most of the rest of what you wrote though.
 
A quote from the designer of Civ5:

"I don't think it makes financial sense to make great AI - budgets can be more efficiently allocated to other areas of development"


He is correct. What good is a great AI, if the game is not great?
 
But the problem with not focusing on the AI from the get-go vxma (or at least hiding it's deficiencies behind clever handicapping), is that you are automatically (in a game like Civ) saying "we don't want to make a great game". So you're handicapping yourself from the start.

From what I can see of the code of Civ 5 (I'm no coder, I just read a lot of what people who know what they're doing are saying), is that the base AI code in war is the same as the one used for Civ 4. Now while Civ 4 eventually got the basics right, it was for a war system which was radically different than for Civ 5, which means that the Civ 4 code is starting off inadequate for the job, leaving the task of adequately coding AI for 1UPT (IMO) a Sisyphean task, which I don't think the designers even attempted.

Shafer's admission reads to me more like: "Yeah I wanted 1UPT, and I knew I'd need new coding for it. But do you know what, I couldn't be bothered putting the time, effort or money in to get it. So I went with the Civ 4 code, that does fine doesn't it?"
 
Thanks for posting this.

...

The biggest concern I have with Shafer's quote is not the idea that you can make good games in other ways than investing huge budgets in AI. The biggest concern I have is that he doesn't seem to have any understanding of those other ways, either. He just seems interested in making games that don't go beyond the 10 year old level. The other quotes in the article tend to suggest that also. That may well be where the biggest business opportunity is, but, if that's your attitude, why put any strategy in your games at all? Why not just make Farmville or Animal Crossing?

Yay someone who gets it! I have been begging game developers to cut back on management tools for years because the more management tools you give the human, the more ways it has to bludgeon the AI. Planet focus in Galciv II, Instantly re-arrangeable Sliders in the paradox games, all the ways to convert science/gold/production into each other in civ games.

You can have a complicated game, but you need to keep the players inputs to manageable things you can reasonably replicate with an AI. If you are coding in tools for the player to customize that the AI cannot use, well you really don't understand what the genre should/could be.

Toyota can atest to what happens when a company switches its focus from making the best product possible to making the most money possible. Further proof that the MBA syllabus does not challenge the brain.

Actually that whole Toyota issue was completely overblown. Way to identify yourself as one of the sheeple.

I find it wonderful that people with no AI experience say these things are trivial or just the matter of hiring the right people.

..

Now, do you hire someone to spend hundreds of manhours developing a finely tuned script they have to toss out every time you make a balance change or do you crank up the little variable that says "let the AI cheat more"? Guess which is easier.

Yay!, someone else who has half an idea what they are talking about. If making a good AI was easy all games would have good AI. AI is so poor you really need to design the game around the AI, or have pre-set scenarios/massive handicaps. Those are pretty much your options.
 
Top Bottom