«

[velik_m]

Originally posted by Mecron@May 8 2006, 04:46 AM
certainly its not secret that the more choices you have tactically and strategically the harder it is for an AI to keep up. SotS has a great, dynamic AI but I will be the first to say that there is nothing like another human to give you an ugly surprise or two. Luckily SotS gives you full multiplayer and makes it easy to join and leave games on the fly as well as just resuming an MP match later.

that's crap, if anything - the more options there are the more (properly developed) AI will trive. The problem with AI is usually, that it's static (or almost static) - which means sooner or later you will figure him out. Also too much focus is going towards (easier) tactical part and too little towards strategic - which is where most humans get advantage over CPU.

[Mecron]

crap? really? Where are you getting this AI knowledge from? Cause if you are correct you must have a helluva GO AI programmed somewhere. Are you aware at all the ONLY "actually better than human" AI is for Chess. A VERY constrained game as far as move and ability choice is concerned. For any game involving concepts more difficult than projecting constrained move choices into the future, game AI's simply become a finite option expert system that one tries to make as robust as possible.

[velik_m]

First, i would like to apologize, crap might be to strong of a word. My intention was not to insult anyone.

i would like to apologize for taking this thread off topic, so this will be my final response on this subject.

Originally posted by Mecron@May 15 2006, 07:25 PM
crap?* really?* Where are you getting this AI knowledge from?* Cause if you are correct you must have a helluva GO AI programmed somewhere.* Are you aware at all the ONLY "actually better than human" AI is for Chess.* A VERY constrained game as far as move and ability choice is concerned.* For any game involving concepts more difficult than projecting constrained move choices into the future, game AI's simply become a finite option expert system that one tries to make as robust as possible.

i don't even know what "actually better than human" means? Does that mean AI is better than human on average? or always? better than all humans or better than most? Using such a fuzzy definition makes it very hard to argue. Chess was for many years a proof of human superior intelligence over computers, that is why AI research pushed in development of AI chess player. However chess is far from ONLY field where AI is succesfull. AI has proven itself on many fields as superior than human and not just in finitive space.

On the July 4 weekend of 1981, while many Americans were preoccupied with barbecues or fireworks displays, players of an immensely complex, futuristic war game called Traveller gathered in San Mateo, California, to pick a national champion. Guided by hundreds of pages of design rules and equipment specifications, players calculate how to build a fleet of ships that will defeat all enemies without exceeding an imaginary defense budget of one trillion credits.

To design just one vessel, some fifty factors must be taken into account: how thick to make the armor, how much fuel to carry, what type of weapons, engines, and computer guidance system to use. Each decision is a tradeoff: a powerful engine will make a ship faster, but it might require carrying more fuel; increased armor provides protection but adds weight and reduces maneuverability.

Since a fleet may have as many as 100 ships?exactly how many is one more question to decide?the number of ways that variables can be juxtaposed is overwhelming, even for a digital computer. Mechanically generating and testing every possible fleet configuration might, of course, eventually produce a winner, but most of the computer?s time would be spent blindly considering designs that are nonsense. Exploring Traveller?s vast "search space," as mathematicians call it, require the ability to learn from experience, developing heuristics?rules of thumb?about which paths are most likely to yield reasonable solutions.

In 1981, Eurisko, a computer program that arguably displays the rudiments of such skills, easily won the Traveller tournament, becoming the top-ranked player in the United States and an honorary Admiral in the Traveller navy. Eurisko had designed its fleet according to principles it discovered itself?with some help from its inventor, Douglas B. Lenat, an assistant professor in Stanford University?s artificial-intelligence program.

"I never did actually play Traveller by hand," Lenat said, three years later. "I don?t think I even watched anybody play it. I simply talked to people about it and then had the program go off and design a fleet?When I went into the tournament that was the first time that I had ever played the game."

Eurisko's fleet was so obviously superior to those of its human opponents that most of them surrendered after the first few minutes of battle; one resigned without firing a shot.

Afterall i can design an AI for tic-tac-toe that will never lose or make a mistake (which humans do) and so can claim it to be superior to any human.

But all that has very little to do GAME AI. I am full aware of problems of game AI.

What are the problems of game AI?

PROBLEM NO. 1: No matter what game players say they want, what they really want is to WIN. Off course they don't want the opponent to just roll over - no they want a glorious battle in which they have to use all their ("superior") brains, knowledge and experience. And they want to win... There is no fun in losing, especially against a computer. This demands put AI on the losing side from a get-go. It also creates bunch of subproblems:

• SCALABILITY - everybody wants to win from newbs to pros, to make a system scalable, it's better if it's simple, using things like state-finitive machines and weighted decision rules.
  
  
• NO RISK - game AI has to put good fight ALWAYS, therefore he can't engage in high risk tactics (tactics where they either win or lose badly - both unpopular results from players point of view).
  
  
• IMPROVEMENT - a player wants a sense of improvement, the longer he plays a game the more superior he expects to be. He doesn't want AI that will develope and improve at the same rate as he is. He doesn't want an ever challanging game - he expects that at some point he will "master" the game.

OTHER PROBLEMS:

• playing field is not even from the start (eg. in RTS AI can order 100 troops in a fraction of a second, while human has to deal with user interface)
  
  
• time - not just CPU time (which AI has to share with graphics, UI...) but thinking time: computer player is expected to think fast - in RT game that is a given but also in turn based games: AI taking 1/10th of time human takes for himself will annoy players.

In general players don't want intelligent opponent - they want an illusion of compotent enemy. Afterall CS bots could headshot you everytime they see you but who would to play against them?
In many games AI has to be dumbed down to make the game even playable.

This all leads to simple systems that are based on untested heuristics, cheats and stacking odds against players, which are easier to develope and better suit the purpose.

IN CONCLUSION:
To address your original question: NO, i don't have "helluva GO AI programmed somewhere", but that has nothing to do AI "having hard time keeping up", it has everything to do with it being a GAME AI and it's natural limitations as well as it's goals.

Knowing all this, i rarely knock on game AI (which is getting better all the time), i just wish game developers would make more effort on strategic part of AI (not just tactical and pathfinding), instead of saying it's next to imposible and just do some simplistic model that can be figured out and countered in about 15min of gameplay.

p.s. i apologize for any spelling errors.

[efthimios]

As a gamer, I do not want only to win. This is not why I play, online or single player.
I do want an intelligent AI/player. Intelligent doesn't mean headshot every signle time, that is accurate and knowing where you are, that is not a matter of intelligence but of design/cheating of the AI since not playing with the same rules.

The joy is not in winning but in playing.

I accept that some gamers see it as you do, but, not all do so, so please don't say that every gamer thinks the same thing/wants the same thing.


I want better AI, not cheating AI. BIG difference.

[Mecron]

LOL I PLAYED Traveller in real life. A LOT. I even entered the Trillion Credit Squadron contest. Sniff! Me am so old.


It is VERY funny though that as proof of your strategic AI argument you chose and ENTIRELY tactical example. TRAVELLER High Guard rules did NOT contain much about logistics or empire management. The contest you refer to is about ship design for tactical combat. While there were rules about assault ships and marine boarding parties I dont see anything in that article about using them. Now IF it had said the AI then went on to play FIFTH FRONTIER WAR and kick behind at it...THEN your point would have been made. k:


Your points about what players want and need from AI's is very good. Your point asking that "more time" be spent on Strat AI's is a gross oversimplification. I agree that often you find inferior strat AI's but that is not always due to lack of effort. I was trained to be an AI programmer. I know the difference between a moment of true AI inspiration and just good guys trying their hardest. There is NO cut and dried path to strat AI programming. It is a wildly unconstrained problem. Sometimes in games like GC2, you can constrain portions of the game itself so the AI can deal with it easier, but that is not always an option. Choosing examples based on MIT research teams is NOT your average game dev situation either. And besides...My point was about INNOVATION. Until you get a self aware lil program running. Innovation will always be the human high ground.

But whatever...its not like people haven't made money convincing folks their AI was the be all and end all of things when its was just cleverly disguised average. I could follow the same path and babble on and on about how our AI will steal your wallet and rent porn while you play. Instead I choose to treat gamers like grownups and say the SotS AI is damn good, but the game runs so deep that a human opponent will always be more surprising a year later. (Hence my deep seated belief in multiplayer) Though I know marketing guys wince when I say stuff like that. :blink:



(For the record, while SotS ship construction is not quite so open ended as High Gaurd (which was based on tonnage and percentages, while Sots is based on a large a variety of section choices with relatively fixed attributes) the actual tactical options when it comes to weapon selections and ship performance is at least an order of magnitude more complex. And of course there is the whole actual strategic portion. :whistle: SotS also requires an AI that can handle 4 different drive system rule sets and of course racial tendencies and diplomacy)

[gregor]

But he was correct on many things. let's say for example C&C:
you start off with a few footsoldiers and MCV. you start building your base. but hey the computer already has the base and all upgrades, yet he decides to send only a fraction of an army to attack you possibly even only his footsoldiers instead of tanks and all. is that intelligence? no. i am sure that is just made to make the game more interesting. because otherwise you would loose immediatelly.


As for more strategic, turnbased games - i haven't seen a good one yet. but then again i haven't played much of the new startegy games. yet i think the built in idiocy is still present - i.e. no attacking human untill human is strong enough to defend and even then - no full scale, risky, suicidal attack.

as for bots in HL mods and others - i don't know much about programming but i was a tester. and no they don't cheat when they give you a headshot, they just calculate it more acuratelly. they calculate better how to counter the recoil and also their reaction time is better. they dont' need a wall hack to do it.
how do i know this? because we gave them same parameters as a human would have (to their weapons, recoil...) but the result was that they were godlike when you faced them and the only way to kill them was to shoot them in the back. they managed to calculate the parameters much faster then you and to counter the recoil of rifles and pistols. so the only way to make the game interesting was to make them more dumb - slower reaction rate, high rate of fire and artificially made recoil and misses.

now that's not cheating. it's a fact that computer calculate faster then humans. otherwise we wouldn't be using them at all.


why would marketing guys be "against" multipalyer? personally, in this time, i believe that single player should be made good, so you can train there while multiplayer is to play with some intelligent opponet that is actually a person. not to mention that multiplayer can bring in quite a lot of money, especially if subscription in involved (e.g WoW).

[Eagle of Fire]

But he was correct on many things. let's say for example C&C:
you start off with a few footsoldiers and MCV. you start building your base. but hey the computer already has the base and all upgrades, yet he decides to send only a fraction of an army to attack you possibly even only his footsoldiers instead of tanks and all. is that intelligence? no. i am sure that is just made to make the game more interesting. because otherwise you would loose immediatelly.

The whole point of not being attacked immediatly when you got there is that it's story related. Would it make sense that the computer start to attack you with everything it got as soon as your start screen apear while he should not even know you are there yet? Absolutly not. Most of the C&C missions involved that you got there unseen and those missions were coded that way. That's why you always had a sporting chance to create your base in the first few minutes of each mission unless you got reckless and gave your position to the ennemy.

If you want a very good strategy game, I recommend Original War.

[gregor]

actually, even when he knows you are there he never attacks with full force. in many missions he attacks you MCV as soon as it get's on the map. but they only seems to be the guards (they are on those possitions on the start). so when they already know you came they still don't do nothing with their heavy force. how do the know where are your buildings if they never saw them? how can them make airstrikes on them if they don't know where they are? cheating?
nah they are made a bit dumb intentionally.

but i did notice in scirmisch game in red alert that he builds new turrets and may attack even with full force. but the problem is that you start building up at the same time. so his full force in that case is no match for yours. unless you double the number of opponents.

problem is they don't know when to pull out, do strategic simultaneous attacks - frontal attack and a few second later smaller attack on some othe risde...

[efthimios]

Are you basing your argunment about AI on Command and Conquer? Hardly the best AI , even for RTS.

I am very surprised to read that you have never played a good turn based game. Is this solely on AI performance? GalCiv II can be quite good, Conflict Europe is good, and there are so many hundrends TBS games that you never liked any one? Wow.

Perhaps, if AI is the problem, you should devote the rest of your life to develop a great AI for games. :blink:

[Mecron]

I say the same thing everytime I see on of those commercials trying to suck money out of young folks to train to be game designers. If they REALLY wanted to help and to get people good jobs in the industry the first thing on the list would be a 4 year course on AI and advanced game theory.