AI coding...oh the pain

I hear plenty of complains from friends about games where the AI "cheats". Both by giving the AI knowledge of user strategy/location and supplying the AI with extra resources...

For example, in some gametypes of Unreal Tournament where it's "all vs all", the AIs team up on the user.

I'd love it if you could show the gamedevelopers how a AI should be And it makes the game more enjoyable if one knows it's fair play... "I can be as good as that...if I just train a little bit harder an play more"

NONONONONONONONONONONONONONONO...that's what makes the AI so good is that they don't cheat!!!! I havn't posted in a long time becuase there wasn't anything I could post on. Anyways, f**k the reviews!!!! besides, like you said, stardock isn't screwd totally if it makes galciv into an exact nibbles clone

Okay my $1/50. First make the AI system open so that people can create their own modifications to the AI -- and the best AIs can be put on the downloads page. Second, you will need an AI that will work on low end machines. Third you should have an AI that will beat the average player without cheating 5/6 times.

Now I think that having the open system is the best route -- Suppose that you let your fans program the actual AIs. Then you let people upload their AIs. When people play "online" they download these AIs and the outcome of the game gets posted to the metaverse. Quite soon you'll be able to see which AIs (and I'm thinking that a really good AI will have different strategies for the different players) will win most often against the non-programmers. You could also have tournaments where each person submits a single AI (rather than a full set of five AIs) and they are played against each other. Stardock might even want to award prizes to those programmers that provide the material for version 1.5 AIs.

Okay -- now about the 1.0 AIs. I'd say you might want to do two different sets -- one that will require high end machinery (or a reduced graphics option) and not cheat to become difficult -- and another set which does cheat at the higher levels but is responsive on lower end machines with a full graphics engine.

I think that this would be an optimum strategy for Stardock to persue.

Thanks!

Frogboy wrote:
"The discouraging thing is - nobody cares about good AI anymore."

You can't be serious! A good AI is the MOST important factor in a good strategy game. I am one of the deepest fans of Sid Meiers "Civilization" and the AI is the HOTTEST topic discussed among fans in the newsgroups.

With Civ they took following approach:
Level 1 and 2: the AI has disadvantages (slower growth)
Level 3: the AI has the same resources as the player (fair game)
Level 4-6: the AI has advantages, but not in the form of cheating, but in the form of faster growth

Most people I know play with Level 3 - that is a fair game. Level 1 and 2 is for beginners. Very experienced players use Level 4 and above, but the game only is enjoyable here if you
- have splendid start conditions (otherwise your are not able to catch up)
- play very aggressive (a war game, that I personally do not like)

So I would suggest to develop a fair and square AI that plays human at best. If you really succeed in this, then you might slow it down for beginners or accelerate it for advanced players.

Of course you are right: many games these days use a cheap AI that cheats - and many reviewers only having some hours to test a game and are not noticing it. But in the Internet age the game - and especially GalCiv - will (hopefully) spread via word of mouth, so you really should make a fair a square AI your number one priority IMHO, as only this will guarantee a long lifetime.

Of course I second the idea to have a user customizable AI. But for this you would have to be one of the best programmers of the world, as no other topic is more demanding and no other strategy title succeeded in this AFAIK. You probably only have a chance if you try it the "Unix way", that is to develop very small and basic scripts that may become mighty once you concatenate them. But this is extremely hard stuff and you easily can get lost on this vast terrain, so beware. My deepest respect goes to everybody who develops a good AI as this for sure is the most difficult and time consuming aspect of a game.

Definitely put the effort into the AI. There is a very big difference between a challenging game and a challenging AI. A challenging game tends to happen once or twice before you find a strategy which always wins. A challenging AI can adapt to your changing strategies and provide challenging games more often.

I agree with the make the AI open as it would hopefully add lots of extra AI programmers to the team. The suggestion of having a low end and high end AI is probably a good one. The average player will never need a very diffucult AI player to make a fun and challenging game. Aim for these players first and allow the advanced players to modify or train the AI themselves and share their results with everyone via the website. These mnodified AI's will require more resources but will hopefully provide the advanced difficulty for those players who need it.

Paul.

Yea I know, some people care about AI. I care about AI. I can't imagine releasing a game where the AI just sits back and cheats and doesn't play the game.

But the amount of work to do it right becomes discouraging when it apparently makes no difference in the bigger picture. I've been reading glowing reviews of the 4th generation of a particularly well known strategy game where the AI basically doesn't do anything. It's little more than a "Get X amount of free stuff per turn with it increasing per turn based on difficulty level". And yet the reviews have been amazingly high.

With 10 minutes, I could just increase the AI's production/money by 10X the difficulty level and you'd have a challenging game.

The coding I have to do to make the AI truly intelligent (like the OS/2 version) revolves around having the AI put together fleets that are balanced to take out particular enemy groups, forming alliances, looking at the polticial and economic conditions of opponents and their allies, and the worst part of all - calculating the prediction of what it thinks the game will look like X turns in the future (which requires a lot of multithreading). All that work and what will the reviews say about it? probably nothing, they'll probably just rip on the game not having a 3D rotatable map ala Homeworld or simulated battles ala MOO or something completely irrelevant to strategy games.

I'm going to put in the work because ultimately game development is also about making a game one wants to play themselves. But it sure is discouraging to see how little computer players apparently mean to how a STRATEGY GAME is reviewed or viewed by the general public.

BTW, as cool as making the AI "open" sounds, it's not really practical. Ultimately in a strategy game, the people who make the game need to make the AI deeply integrated.

I have never played a game as good as the OS/2 version of GalCiv. I have played CIV, all of them (1-3), MOO2, AlphaCentauri, Age of Empire,
Ceasar 2 & 3, WarCraft 2, SpaceCraft, and a few real time games. CIV and A/C come the closest to GalCiv, then MOO2 IMO. The others are so
simple they are not worth playing more than a couple of times. And I will bet they don't have an AI that deserves the name.

Keep playing up the re-playability aspect of GalCiv in the interviews and you will build an expectation in the reviewers. Keep talking about
how you get a different game every time you play GalCiv because of how smart the AI is and you will build another expectation in the
reviewers. This all has to help when it comes time for the reviewers to review GalCiv (specifically, does it live up to expectations - the
expectations YOU convince them to have). Don't give up on the AI just because you think the reviewers will pan it. Convince them follow your
game plan, don't let them force you to play theirs.

Cheers,
Jack

Also, may I add that if you paint a picture of what to expect from GalCiv for the readers of the interviews that if the
reviewers fail to review GalCiv with an eye to what you say in the interviews that the reviewers will loose credibility?

Cheers,
Jack

Maybe you can get some interviewer to ask about what you expect from the reviewers, perhaps something about
how they seem to ignore the strength of the AIs in games. Then you can respond about how games whithout a
good AI is so bland that noone wants to play it more than a couple of times ( before tossing it into the
circular file).

Cheers,
Jack

Hello,

just my voice for AI. AI might be not so important for casual Age of Empires player, but real strategy player tends to get disgusted by cheating AI. IMHO, game with cheating AI has no much of replayability value - this is why I like Imperium from EA (extremely old game) from 4X games. Terrible controls, poor access to information, but computer does not cheat (just has more resources on start). Please, dont give up AI - you are my only hope for really good game in these bad times for strategist

I am even considering purchasing Drengin.net access now, as soon as I find the way how to make payment from Czech republic

I thought the whole point of you folks being independant was to produce the games you want to play, not being content with the junk out there now.
I've played several games with cheating AI's and they are the ones collecting dust in the corner. Guess that's why most of my board wargames have been gotten rid of- no one wants to use their noodle enough to give you a good run of competition, or spend the time required, maybe?
I think an interesting-and underated- AI is that used in Age Of Wonders. The AI is somewhat user configurable in the Scenario Editor. I'm not saying its perfect or couldn't be improved on, but its one of the best I've encountered yet.
Don't start feeling rushed and throw something together just to get the Beta out for those of us excluded from the Alpha. I'd rather wait a while longer and have something with value than slop. Just my opinion.

For me, a good AI and UI are the most important for a strategy game. The ulimate goal is an AI that doesn't cheat and has no advances compared to the player, but still beats you. Like losing from a chess computer.

To me personally AI and Gamplay are most important things to a game period, unless of course if it's a online game. It doesn't matter if it's a FPS or a Sports game, AI should be important in today's games but there may be some hope. Look at a game like Jedi knight 2 one of the main reasons I like it so much is the AI is pretty damn good! I mean it's not like playing a human player or anything but it's pretty challenging and the bots seem generally smart. Another FPS is Halo believe it or not the AI is really good. As for strategy games, I would think they would be way more complicated to do well.

Even a game like Civ3 had problems with the AI but it was still very good compared to what else is out there. I think we should raise the bar, Brad you are know for AI, I say run with it and don't look back.

Brad, if you decide to have a dumb brain cheating, let me know. I'll keep my P100 with OS/2 and GalCiv Gold. It's still enyoyable.
I have to admit also the version I liked more was 1.07...
You know the code, any reason for that?

My vote is for the best AI you can get. Forget low end and high end computers. If you have multithread then you can make it.

If your only point is to have great reviews then forget coding and start writing checks :VOMIT:

I personally think a good AI is a good thing. I don't play games online for the most part, so AI is very important to me. I absolutely hate games that are intended to be played online and the single-player mode is an after thought, and thus usually cheats and takes the fun out of the game.

As for reviews, I hardly ever read them. I look at what people (ie the players) say about the game. This is where a good AI will help you. People will talk about the challenge of a game, and not the cheating nature of it.

I say keep at it, it'll be worth it in the end. The reviewers may not notice as much, but those who matter most, us, will.

Don't get me wrong, the AI in Galciv is going to be good.

The whole reason I write games is to write AI. I'm just dispapointed that STRATEGY games can get good reviews that have no AI in them.

Well I know of a gamesite / tv show that loves good AI maybe you should ask them the tv show is called electric playground and they loved the AI in HALO. The guys there were freaking out on how good it was and it was nice to see a good AI for a change. I know personally that AI is what makes the game fun I hate when you go up against the AI all he has is about 2 or 3 planets and his F**KING fleat is just as big as yours or has 3 times as many small ships that retreat all the time you encounter them. And If I were to try just having 3 planets in my control the AI would destroy me in 2 seconds. Also you said it yourself and I quote.

The coding I have to do to make the AI truly intelligent (like the OS/2 version) revolves around having the AI put together fleets that are balanced to take out particular enemy groups, forming alliances, looking at the polticial and economic conditions of opponents and their allies, and the worst part of all - calculating the prediction of what it thinks the game will look like X turns in the future (which requires a lot of multithreading). All that work and what will the reviews say about it? probably nothing, they'll probably just rip on the game not having a 3D rotatable map ala Homeworld or simulated battles ala MOO or something completely irrelevant to strategy games.

If you don't do it then save yourself the time and just make this into a meele type of game where you give everyone ships and let them go at it. Because with out alliances fleet groupings ect.ect. thats what you have.

I hope to hear more on this

Jason

As I said above, the AI is a major focus of GalCiv. In some ways, the game is AI with a game around it for the person to interact with. But it is disappointing that fewer and fewer strategy gamers seem to worry about it. It's not just reviewers who rave about strategy games with no AI, you can read reader reviews and they'll rave about the features of some strategy game where the AI clearly isn't completed or even fully functional.

You did a great job with the OS/2 Galciv AI, and you will do a great job with this one. Don't worry about the reveiwers.

Cheers,
Jack

Brad,
unfortunately most people nowadays buy a game, play it for maybe 40 hours and then move on to the next game. It's probably not that hard to make a game challenging for 40 hours even with a crap AI. The question is whether you want a game which will keep people happy for 40 hours (which is a fair return on your money) or a game which people will still be playing a few years later.
I vote for the few years later, but let's be honest, the amount of effort in programming to achieve this will never make the same return (effort versus sales)for a company. Especially as people won't buy the next game if they're still enjoying the last one!
Both reviewers and players are right to praise game features and enjoyment even if the AI is lacking. They should however state the lifetime of that enjoyment (10 hours, 40 hours 100+ hours). Then we can choose what we want to buy more carefully.
The biggest bonus in development a strategy game with good AI is the fame. It will become a classic. Go for it!

Paul.

Paul

I have to agree with you but i believe the problem is that people have been made to believe that if a game keeps you occupied and you enjoy it for 40 or 60 hrs that it's a great game. That's why i believe that people no longer look into how complex the AI is because in the end they expect the game to only last 40 or 60 hrs of game play. I on the other hand i'm very disaponited and always try to make the ai harder by changing its stats well I always came to the same conclusion if I wanted a game to be as complex as I am I would have to play against someone else. Aren't people tired of the game AI being super strong then when it hits you it takes out half ur fleet and territory then you attack and in one sweep you kill the AI? I don't know but personally I've yet to see a good AI.

I agree that AI is both very important and very difficult to get right. Oh yes, I have just signed up for GalCiv as it appears to come close to my ideal galactic game I heve looked for forever. One inportant thing about AI, it should not cheat. It should struggle under the same handicap as players do. It must be able to learn however from play and thus grow with the game itself.
About me: I am a long time game player and have been a beta tester for a number of online real time games going back to the early Kesmai games and Sierra network. I did the beta for Ultime and Everquest, Shattered Galaxy and one called C.O.R.P. that dissappeared quickly. Well enough for a start. I look forward to GalCiv, GreyWolf

Cyberfluf,
I agree with your comments, but must argue that 40 to 60 hours of entertainment for your money is a very good investment. Compare it with a night out or a cinema trip and you start to realise that computer games really are good value. Even a game with poor AI which cheats but is carefully scripted for once off play can give this value.
Our problem is that we're all greedy! We want to be able to play strategy games (and usually FPS as well) again and again. But for the same price!
I love playing against good AI. Not AI with a huge advantages or better starting locations or cheats, but AI which can adapt play style to what you throw at it. It doesn't have to beat me (I like winning) but it does have to make me unsure of victory and keep me challenged.
Paul.

Paul,
You have stated the two most important things to me in computer games.
1). Let me win.
2). Keep me in doubt about my winning until near the last (let me see my success coming, don't
suddenly suprise me with sudden victories).

Cheers,
Jack

ps., I have decided I don't like CIV3.