The problem of AIs in strategy games (Full Version)

All Forums >> [New Releases from Matrix Games] >> Distant Worlds 1 Series



Message

Wicky -> The problem of AIs in strategy games (10/27/2011 2:16:14 PM)

When an AI battles against humans in a strategy game, it always ends that certain technologies need to be balanced, also certain prices for technologies, up to the trading of entire colonies etc...
Can't that be done easier???

I ask myself, there's always something that a human will exploit!
In conclusion: No matter how much precision is put into balancing and development of an AI, they will always "be stupid" for one reason:

The lack of imitation !

AIs cannot imitate wtf lol.. They compute resources demands ahead of time, can react in milliseconds to human threats, never loose the view over the battlefield, and yet the most important ability is missing. Now, an example:

What if the AIs in distant worlds would try to trade techs with each other everytime the human trades techs? If the AI would copy the ship design of the winner everytime it looses a ship battle?

Such is the human intelligence: in their left brain they have a certain tactic to win, and with the right brain they copy and imitate everything they do not know. If the AI could only do 1/10 of this imitating, we'd all be doomed!!



Brainsucker -> RE: The problem of AIs in strategy games (10/27/2011 4:57:13 PM)

Well, AI in a game is created not to beat human players. They are created to give us the feel of riding a roller coaster. That's mean, they give us thrill, but also fun. They are not created to kill us. The problem is, our fun rating is different one to another, and sometimes, our skill to play demands more challenge. And that is what the AI lack. They are not adapted to the challenge.

Human has the capability that AI not. Adaptation. They can remember, then analyst and find a way to beat the challenge that given to them. Well, not always successful, but if you play and re-play the same challenge for many times, you will become adapted and understand how to beat the threat. The problem is, AI hasn't the same capability. They will always have the same value from the first time they fight you, to the time when you have already adapted and know how to beat it. AI can't face new challenge that they are not prepared to handle. That's why, AI manual update from the developer is always needed. But... it also not always give you the best solution for the game.

Because, sometime, your skill, as a hardcore player has surpass the DEV. So even that the DEV themselves come to fight you, they won't be able to win. If that condition becomes true, you will consider that the AI of the game is stupid. Well, because no one know how to handle your skill to play anymore.



LoBaron -> RE: The problem of AIs in strategy games (10/27/2011 6:48:14 PM)

AIs lose when they have to work with non-values. At the moment there are no input variables to calculate the human brain is
better, because it creates and categorizes its own variables.

My grandparents lived in Linz. In Voltastrasse, near the Kepler University. [:)]
Loved the place when I was a kid.




feelotraveller -> RE: The problem of AIs in strategy games (10/28/2011 6:37:42 AM)

The AI isn't an AI. That is, it does not think. It computes. Two very different things.

Unfortunately computing (crunching sums or following algorithms) has come to be understood as thinking.

I blame the stockmarket. [:D]

p.s. Actually a programmed game is the "AI" home world. Imagine if it tries to debate us!



adecoy95 -> RE: The problem of AIs in strategy games (10/28/2011 7:03:34 AM)

you can make decent ai in strategy games, but it requires a few things:

they might have to cheat, yea it sucks, but especially in real time strategy games, some things just need to be cheezed.

for example: in my personal opinion, company of heroes has a great AI, even better than relics later DOW2 ai. however, it cheated.

for one, it got a little extra fuel (rare resource needed for vehicles), and it had no fog of war. this allowed it to use long range weapons such as anti tank guns without needing to be programmed to effectively use scouts. while i wont pretend to understand ai coding, i would imagine that trying to teach a computer to use scouts without them getting killed in real time while still being useful would be an impossible task.

in galactic civilizations (or was it civ 4? i cant remember lol, playing those two games were so identical i forget) the computer would trade techs like crazy. this is a good suggestion by you because it helps keep the ai competitive. they would make sure that any techs you obtained that you traded were spread around like wildfire. to the point were you could not even exploit the tech for trade because everyone had it in a turn or two.

it requires commitment:
a lot of developers that make strategy games these days assume the player will do multiplayer or campaign. and statistically this is probably true. however, this mindset gets slippery when it comes to games with only one game type and no multiplayer.

the best and (maby) the easiest way to add a difficulty curve is most defiantly a difficulty slider. return of the shakturi had it half right; however, (spoilers) capturing their homeworld gave you their "cheat" bonuses, and that was bad for keeping the game challenging end game.

i am hoping that the combination of difficulty sliders and spheres of influence will be able to jostle me right the heck out of my usual playstile with a swift kick in the face. something i haven't gotten yet without the most extreme use of sliders for empire development levels.

fortunately i think it has been confirmed a slider is in for this expansion



feelotraveller -> RE: The problem of AIs in strategy games (10/28/2011 8:26:47 AM)

Brainsucker hits it on the head. The computer players are there to make the game fun. The difficulty is providing variable levels of challenge.

Giving the computer bonuses is a shortcut which programmers use to avoid building decent AI simulations. Worse is giving them bonuses of kind rather than degree (e.g. access to a special resource rather than quicker mining).

So is failing to code them with a FOW. It takes the fun out the game by making stealth, ambushes and deceptive tactical manuevers impossible/useless.

Some of the (potential) fun of computer players comes from having them do different things at different times rather than race X always doing Y. Their gameplay should vary from game to game only partly based on the situation they find themselves in.

Generally a lot more could be done to make the computer players responsive. For example coding for an investment in PD after having a fleet walloped by fighters (but only after a significant battle and not when the player first starts building fighters). The data is there in game to allow this responsiveness but it should be under FOW conditions - no omniscience for the computer player.

Tech trading was (IMO) stupid in the games you mention. It just forced you to trade it to every computer player at the same time. It must always be considered what a player will do in response to the 'behaviour' of the computer players. Another reason that their play should not be strictly scripted.

I suspect I won't be disappointed by Legends but generally sliders mean sloppy programming. A sheer quantitative change which does not result in a qualitative shift in game play just forces players to exploit stupidities and shortcomings of the programming. The computer players end up not playing any better but only being bigger.



Darkspire -> RE: The problem of AIs in strategy games (10/28/2011 8:34:52 AM)

The only game ive ever played with AI that really gave the impression i was playing 'hotseat' would be the original GalCiv, on the higher levels it really did seem to react correctly in a given situation and give the feeling you were not just playing a computer. Ive been into AI since i saw Wargames and have been hoping with all the increases in processing power that someone would come up with a decent AI engine, like the unreal 3d engine is licensed out to make games. 99% of games, DW included, just seem to work on the cause and effect principle which isnt really AI IMHO. DW is a really great game but with a proper AI it really could be something special, at the moment it plays more like an arcade game, as an example look at the freighter problem with fuel, with an AI subroutine that wouldnt be a problem, AI, as i was taught many moons ago, can be viewed like a flowchart, yes and no creating new paths to choose,
as an example with our freighter;
check the players space stations for fuel levels, are they full? no, has the nearest gasmine got enough fuel to warrant a trip to fill up? yes, go to the gasmine, fill up and return to the nearest low fuel level space port. no, check next nearest mining station
rinse and repeat ...
Dual core processors have been around a while now and still no one seems to be able use all that extra power to make a decent game AI, maybe they are all worried after watching Terminator [8|]

Darkspire



feelotraveller -> RE: The problem of AIs in strategy games (10/28/2011 8:53:56 AM)

Well, yes and no. [:D] Not to start a debate or anything... [;)]

Have we all been waiting for better 'AI', absolutely. What have we got - lovely arcade 3D graphix.

The flow chart stuff, um, I don't think that's intelligence. It's certainly not how I think strategically. Okay, we all do the coffee or tea stuff too, but thinking seems a lot more complicated than a yes or no and/or logic allows.

I really hope that Elliot, somewhere down the track, takes a shot at doing something real with 'AI'. He's got the rest of the package as far as I can see, except the fancy 3d images that I (and my poor old processor) can easily do without.



Wicky -> RE: The problem of AIs in strategy games (10/28/2011 10:45:17 AM)

quote:

ORIGINAL: Brainsucker

Well, AI in a game is created not to beat human players. They are created to give us the feel of riding a roller coaster. That's mean, they give us thrill, but also fun. They are not created to kill us. The problem is, our fun rating is different one to another, and sometimes, our skill to play demands more challenge. And that is what the AI lack. They are not adapted to the challenge.

Human has the capability that AI not. Adaptation. They can remember, then analyst and find a way to beat the challenge that given to them. Well, not always successful, but if you play and re-play the same challenge for many times, you will become adapted and understand how to beat the threat. The problem is, AI hasn't the same capability. They will always have the same value from the first time they fight you, to the time when you have already adapted and know how to beat it. AI can't face new challenge that they are not prepared to handle. That's why, AI manual update from the developer is always needed. But... it also not always give you the best solution for the game.

Because, sometime, your skill, as a hardcore player has surpass the DEV. So even that the DEV themselves come to fight you, they won't be able to win. If that condition becomes true, you will consider that the AI of the game is stupid. Well, because no one know how to handle your skill to play anymore.



Makes me sad, that the only option after beating the AI in every way possible is to stop playing. Considering it took them so many updates to create our current DW AI, and the first chess computer to beat a human champion took 20 years, there's probably no hope for me anymore. :(



MartialDoctor -> RE: The problem of AIs in strategy games (10/28/2011 11:10:55 AM)


quote:

ORIGINAL: Wicky

Makes me sad, that the only option after beating the AI in every way possible is to stop playing. Considering it took them so many updates to create our current DW AI, and the first chess computer to beat a human champion took 20 years, there's probably no hope for me anymore. :(



How are you trying to make a challenging game in DW here? What settings are you using for yourself and for other AIs?

Although I will agree it's difficult to make a challenging game(although with Legends, it should be much better as there is a difficulty setting and improved AI), it's still possible...



Data -> RE: The problem of AIs in strategy games (10/28/2011 11:11:05 AM)

There's always hope....but a lot of work also. Good things come to those who wait [;)]



Bingeling -> RE: The problem of AIs in strategy games (10/28/2011 12:56:10 PM)


quote:

ORIGINAL: Wicky

Makes me sad, that the only option after beating the AI in every way possible is to stop playing. Considering it took them so many updates to create our current DW AI, and the first chess computer to beat a human champion took 20 years, there's probably no hope for me anymore. :(


Get even sadder. Making a very good chess AI is trivial to making a very good Distant World AI. Chess is a game of simple rules and a limited number of positions, after all. And even with that much limitation, the game has a massive (and then some) number of possible plays.



Abraxis -> RE: The problem of AIs in strategy games (10/30/2011 4:52:10 AM)

As previously mentioned, AI will only fail when it's meant to imitate other players.

When viewed as a simple tool to antagonize the player with varying levels of difficulty, "cheating" loses meaning and suddenly you can do some pretty interesting things.

AI War for example. Never was really my style, but it embraced the idea. You literally were fighting a crazy AI of some (any number of) sorts. They didn't try and dress it up like a human and hide the circuitry, they just ran with it and it worked really well.

That's not to say I don't think they should stop trying to make convincing an human like AI opponents, but maybe sometimes it's over-valued and emphasized.



J HG T -> RE: The problem of AIs in strategy games (10/30/2011 7:30:16 AM)

^ Agreed with that. I've played AI War a little (should probably play more of it at some point) and the "war against an AI" setting works and lets the devs to create truly challenging enemy to fight. "AI" is indeed AI wars main selling point.  



inf666 -> RE: The problem of AIs in strategy games (10/30/2011 12:13:05 PM)

I have to disagree. I think it IS possible to program an adaptable AI. The hard part is to make them scalable so that the game does not become too hard or easy for players of varying skill. In essence the AI has to observe the player and calculate his "skill" and react accordingly. Player is very inefficient = be inefficient as well. Player has no sense for strategical movement = make lots of fleet movement mistakes as well.

If a talented developer understands the game mechanics well, he could implement an AI that would crush everyone every time. Most developers rather choose to make "bad" AIs because the player has to win no matter what. After all where is the fun in losing every time to a fricking machine? After three games most players will stop playing and make bad publicity. Thus the game AIs are deliberately being kept dumb to ensure that players do not lose. *sigh*




Erik Rutins -> RE: The problem of AIs in strategy games (10/30/2011 4:53:18 PM)

I think we've provided enough options that if you want the AI to crush you, it will oblige, but the default game mode is a bit easier. The ROTS AI was tougher than the original DW AI and the Legends AI is IMHO much tougher than the ROTS AI, plus we have added the difficulty slider where you can give the AI a boost if you desire.

Regards,

- Erik



MartialDoctor -> RE: The problem of AIs in strategy games (10/31/2011 11:11:20 AM)

quote:

ORIGINAL: inf666
If a talented developer understands the game mechanics well, he could implement an AI that would crush everyone every time. Most developers rather choose to make "bad" AIs because the player has to win no matter what. After all where is the fun in losing every time to a fricking machine? After three games most players will stop playing and make bad publicity. Thus the game AIs are deliberately being kept dumb to ensure that players do not lose. *sigh*


I'm going to take a guess and say you are not an AI programmer.

Until neural networks can be developed where AIs can adapt and think, there is no way that an AI can beat a good human opponent. AIs can't think or develop new strategies; they only follow pre-set, programmed routes that follow set paths in certain situations. The most that can be asked for is a good programmer can develop strong AI routines that will have the AI make reasonable decisions. An AI can not adapt and learn at this point in time.

The only AI that has ever beaten a good human opponent was in chess. And the only reason that was possible was because they had a freaking supercomputer that was so powerful, it could calculate every single possible move and calculate the best move based on all possible moves from that move. It had no way of adapting or truly thinking...

In the game Go, they still haven't invented an AI opponent good enough to beat a good, pro player. The best one invented yet lost to a pro player even when given a 9 handicap (the highest available).

I won't beat this bush any longer. But if you think an AI opponent can be designed to beat a good human opponent, in a game like Distant Worlds (or just about any computer game out there), you are very mistaken.



martok -> RE: The problem of AIs in strategy games (10/31/2011 11:30:06 AM)


quote:

ORIGINAL: Erik Rutins

I think we've provided enough options that if you want the AI to crush you, it will oblige, but the default game mode is a bit easier. The ROTS AI was tougher than the original DW AI and the Legends AI is IMHO much tougher than the ROTS AI, plus we have added the difficulty slider where you can give the AI a boost if you desire.

Regards,

- Erik

A question, Erik:

Assuming we have the original DW plus both expansions, at what difficulty level is the AI going to play its best game, minus any major cheats/bonuses?






ASHBERY76 -> RE: The problem of AIs in strategy games (10/31/2011 12:57:10 PM)


quote:

ORIGINAL: MartialDoctor
I won't beat this bush any longer. But if you think an AI opponent can be designed to beat a good human opponent, in a game like Distant Worlds (or just about any computer game out there), you are very mistaken.


Really? I consider myself a good human player who has been beaten many times by the A.I.SOTS1 can be pretty dam tough if you fall behind early.

Human players tend to restart failed early games a lot if early colonization is poor,ect,same with ingame events like a bad battle defeat,reload,reload.The A.I cannot do that.I think we would lose a lot more without this option.



Nedrear -> RE: The problem of AIs in strategy games (10/31/2011 1:10:42 PM)

There are several ways to seemingly imrpove an AI over humans:

- Counter strategy by speed
- Counter strategy by bonuses
- Counter strategy by omniscence

What does that mean?

A PC can not calculate quicker than a human. It is only capable of focusing all his capabilities on a single task. Humans can not! That gives the enemy the edge if combined with a slow interface. Starcraft II uses that on brutal by forcing you to play quicker against an enemy who is a little bit more refined than hard... but only a little. Sometimes he gets bonuses too.

Bonuses are most common in games of the Total War series. The enemy always wastes money on large troops and moving them about. This does not bode well with economy. To let them advance further on they cant get a negative income or get an increased income of provinces. That is a definite bonus to counter lacking strategy as the opposed solution, splitting empire troops and moving them in small packs, is too much a hassle to programm in a tactical manner.

Omniscence is a way some people create a mighty AI. It is not that good though. Imagine you can see everything your enemy does. Expecting attacks and creating countermeassures is not that much of a deal than. In Starcraft that means sending out AE hitter when a group of bursters are rolling to your base etc. You can still wear down the enemy by tactical advance, defense and spys. Meaning: Adapting to the enemy moving routines which the enemy can't, getting shredded and drained of ressources.



ASHBERY76 -> RE: The problem of AIs in strategy games (10/31/2011 1:15:55 PM)

I bloody hate omniscence A.I cheats in strategy games where sensors are a factor.I remember playing alpha centauri back in the day getting my stealth sub fleets ready just to find out at a later date the A.I knew where they were all along.



Erik Rutins -> RE: The problem of AIs in strategy games (10/31/2011 2:11:57 PM)


quote:

ORIGINAL: MartialDoctor
In the game Go, they still haven't invented an AI opponent good enough to beat a good, pro player. The best one invented yet lost to a pro player even when given a 9 handicap (the highest available).


Beating a pro player is far different from beating an average player though. It's absolutely true that on a level playing field, an AI can't compete with a human, but we allow you to make the playing field a lot more uneven than a 9 handicap in Go (I wish I had more time to play Go, great game).

Regards,

- Erik



Erik Rutins -> RE: The problem of AIs in strategy games (10/31/2011 2:13:37 PM)


quote:

ORIGINAL: martok
Assuming we have the original DW plus both expansions, at what difficulty level is the AI going to play its best game, minus any major cheats/bonuses?


I think one notch up from the default will be a good place to start for experience players. A lot depends on the rest of your galaxy setup options as well. The AI in Distant Worlds does not cheat at all, but you need to give it some bonuses through the various settings that are under your control if you are an experienced player and want a challenging game.

Regards,

- Erik



MartialDoctor -> RE: The problem of AIs in strategy games (10/31/2011 2:40:05 PM)


quote:

ORIGINAL: ASHBERY76


quote:

ORIGINAL: MartialDoctor
I won't beat this bush any longer. But if you think an AI opponent can be designed to beat a good human opponent, in a game like Distant Worlds (or just about any computer game out there), you are very mistaken.


Really? I consider myself a good human player who has been beaten many times by the A.I.SOTS1 can be pretty dam tough if you fall behind early.

Human players tend to restart failed early games a lot if early colonization is poor,ect,same with ingame events like a bad battle defeat,reload,reload.The A.I cannot do that.I think we would lose a lot more without this option.


I have only played SOTS once and so I can't comment on the AI there. My impression was it was better than average, though.

You are probably correct with the amount of people who reload their failed battles / games; I've been guilty of it before. But would you lose without making the other races larger and more tech advanced than you? My guess is not.

Anyways, I'd recommend not doing the loading and restarting. You'll find you have a much more challenging and fulfilling game if you just play the game as it turns out. You will also have to be much more careful about how you play. I find it much more interesting, at least [:)]



Page: [1]

Valid CSS!




Forum Software © ASPPlayground.NET Advanced Edition 2.4.5 ANSI
0.6870117