The Problem with AI

[Blackw3]

I took a few years off from playing strategy games and only got back into it with the release of Rome 2. After having been disappointed with Rome 2's AI, then BE's AI, and now EL's AI, I can't help but question whether this is an inherit problem with the genre of strategy games (or maybe 4x more specifically) rather than a conscious decision from developers. I even played through 2 campaigns of CiV BNW leading up to BE's release, and even then AI was whatever. Note, I haven't played EUIV so I'd be happy to hear of people experience with that game.

So my question is, is it too expensive to develop an AI that actually enhances the gaming experience (not talking about artificial buffs)? I know AI as a whole has its limitations, but there are other industries that have developed very advance AI (financial / military come to mind). I also don't recall having this much of a problem with AI back in the day, but that could just be me approached games a lot more casually or because the game mechanics were so much simpler that the AI didn't have many things to figure out.

But if this is the case, if the limitations are really there, why are developers not looking to further develop the multiplayer aspect of these games? I fully understand that 4x games present a bigger challenge mostly due to them being turn based and the length of time required to finish a game, but its getting very little attention. I love strategy games, but for a long time I was playing LoL just because it was so easy to hop online w/friends and play together. You don't need to look much further then the success of LoL, or Dota, SC2, or even Clash of Clans to understand the value multiplayer can bring to a game. I'd love to see 4x games get even a fraction of the funding these games get.

Yes, of course most 4x games have a multiplayer option, but they are developed a single player games primarily, and multiplayer is only a secondary concern. One needs to scour the net to find groups and join communities, and schedule play times. There is no ranking system, no organized platform for players to meet. No way to keep track of a players performance or behavior.

I'd love to hear the communities thoughts on the matter.

 

[terraslayer]

Well to be fair, games like starcraft and Warcraft didn't have great AIs either. They depended on similar buffs as in this game (giving them cheap bonuses and a head start) to make them fairly competent. The problem with making a very good AI is that it's difficult. A human player has the option to change its strategy depending on the situation. To make an AI do that, the coders would have to pre-know what situation, and then code the AI how to handle those changes.

I'm not much of an AI coder, but I assume it's one of the hardest parts of coding a game.

 

[Winsling]

The short version is that games have gotten complicated faster than AI has gotten better, because AI doesn't sell. Consider the tech web. It adds depth that isn't present in Civ 5, but the AI is completely incapable of navigating it. It's much harder to teach the AI to use a feature than it is to add it. So publishers choose between lots of novel features, or a feature-poor game with excellent AI. The latter appeals to a small minority.

This is unlikely to change from large publishers. You're more likely to find something to your taste from a smaller shop.

 

[screamingpalm]

I don't think it has anything to do with being expensive, more likely to do with talent and/or passion. AI War by Arcen has an outstanding AI, and they are a teeny tiny indie dev with a mere fraction of the resources of a AAA studio and TWO programmers to code it. It is also a very complex game. It takes an unconventional design approach, and I suppose if other devs could figure out how they did it, we'd probably have better AI's in general.

 

[Ghost_Nappa]

the more complex (and situational) a game gets and the more options there are, the more difficult it becomes for the AI.

Take AI siege battles in RTWII f.e. There are tons of factors to consider laying a siege. And there are tons of different maps, too. You can standardize a couple of things but when it comes to decisions in different situations, the AI usually fails.

Esp. when it comes to cheese and how to react to it.

I guess CiV AI must be one of the harder AIs to make when it comes to games. There's so much a player can optimize (min-max) to get the best results. With new features (f.e. tech-web), new AI problems arise. The tech web is great for the players as I now can react to any kind of situation. But it makes things worse for the AI, which might simply develop 5 times the wrong things in a row. Or not utilize correctly what it got. And then I can easily beat it.

Switching to mp in turn-based strategy games... well. hard thing to do. From what I know, the mp communities are rather small and minor. As you said, it's hard to schedule a game. Esp. when it can take a couple of days. The success of most mp games lies therein that you can just hop in whenever you want and play a couple of hours and log off again. Whereas a turn based strategy game may take weeks to finish (depending on everyone's schedule).

Giving the AI bonuses is usually the way to go. It's been this way ever since and I doubt we'll see a change there soon. Somewhat a learning AI is needed, which doesn't always make the same mistakes. Maybe a system which tracks AI decisions and evaluates them, whether the chosen path was good given the circumstances or not. And using this to base future decisions on it. This would make the AI better with time. But I doubt we'll see something like that anytime soon. Could take years to program and would probably need ressources, a simple gaming studio wouldn't ever dream to have.

this article I just read today, comes to mind:

http://www.thecreep.info/2014/10/google-reveals-it-is-developing.html

 

[screamingpalm]

the more complex (and situational) a game gets and the more options there are, the more difficult it becomes for the AI.

Take AI siege battles in RTVII f.e. There are tons of factors to consider laying a siege. And there are tons of different maps, too. You can standardize a couple of things but when it comes to decisions in different situations, the AI usually fails.

Esp. when it comes to cheese and how to react to it.

I guess CiV AI must be one of the harder AIs to make when it comes to games. There's so much a player can optimize (min-max) to get the best results. With new features (f.e. tech-web), new AI problems arise.

I don't really agree with you. Again, if you look at AI War, where you can have battles with thousands of ships and a myriad of different types of static defenses and dozens of AI "personalities" all wrapped up in an extremely complex and deep game that can make Civ look like child's play... well, it sounds like a lame excuse to me.

 

[SupremacyKing2]

I think some players have unrealistic expectations towards the AI. The fact is that developing an AI for games like civ is extremely hard. Just look at how hard it is to program a small robot to navigate a room with obstacles that a 3 year old could navigate with no problem. We take a lot for granted because it seems to come so easily for us.

 

[kirbdog]

LoL actually worked to improve the AI of their bots, and they can provide a challenge to the player for many hours until the player learns to get better.

Obviously there has been a huge push in computer science to develop superb AI for complex, classic games such as Go, Chess, and Bridge.

(Don't believe those telling you these games are not complex... they are. Even if the branching factor is lower, you need to go a lot deeper into the decision tree to make good choices, while a game like Civ lends itself to shallow, this-game-turn analysis.)

The simple fact of it is that a lot more resources have gone into Chess AI than into Civ AI. Period. End of story. More resources (i.e., cash thrown at AI developers) = Better AI.

 

[Blackw3]

I think some players have unrealistic expectations towards the AI. The fact is that developing an AI for games like civ is extremely hard. Just look at how hard it is to program a small robot to navigate a room with obstacles that a 3 year old could navigate with no problem. We take a lot for granted because it seems to come so easily for us.

Its not a matter of expectation, its a simple matter of the type of games we love continuing to receive the attention and funds necessary to stay profitable. I don't have the exact figures, but this last decade has seen a huge surge of gamers from demographics that we would have never suspected 10-15 years go. The same cannot be said for games like Civ. And believe me, I'm not one of those that thinks washing down a game and making it appealing to a large audience is the way to go. It just needs to look at games like Clash and Clans and figure out what this simple free to play game did to make it a billion dollar game (being available on a cell phone put aside).

Most of the recent successful games provide the player with the opportunity to interact and compete against live players. Those of us that put in hundred of hours playing single player and hopping from one mod to another to extend the life of the game are the minority. So to me its either about improving the AI to the point where its actually Artificial Intelligence (as someone said, learning) or shifting gears to make multiplayer more playable.

 

[Janskey]

Apparently it takes some MIT scientists to create a better AI:
http://people.csail.mit.edu/regina/my_papers/civ11.pdf

 

[Callonia]

Oh please, AOW3 have better AI than Civ series do.

Firaxis thinks an army of 15 units will topple my empire and salt the earth which is such a joke.
Meanwhile If an AI dow on me, I use every single resource I have to stop them.

Firaxis haven't bothered to put in the effort required and it shows. It is with great sadness that my chapter with Civ series has ended in Civ BE as for I'm reasonably certain that Civ 6 would be same shoddy military AI that never gets upgraded even when they go back to boring stacks of doom. I was even more invincible in Civ 4.

While the AI in aow3 made me retreat and abandon an city for the first time in years.

At this point, it just feels like people is making excuses that AI is too hard and collect our money while doing nothing instead of making experiments to see what works and what doesn't work.

 

[johnnybaseball]

I think that people on this forum vastly overestimate how easy these games are for most players. Most players probably don't think there's anything wrong with the AI, and the AI does not detract from those players' enjoyment of the game. It is tough to say that developers should spend resources developing something that only appeals to the people who are already most likely to buy the game. "Better AI" isn't a selling point to anyone who isn't already heavily involved in games like these. IMO, that's probably why they don't focus on it as much.

 

[Blackw3]

I think that people on this forum vastly overestimate how easy these games are for most players. Most players probably don't think there's anything wrong with the AI, and the AI does not detract from those players' enjoyment of the game. It is tough to say that developers should spend resources developing something that only appeals to the people who are already most likely to buy the game. "Better AI" isn't a selling point to anyone who isn't already heavily involved in games like these. IMO, that's probably why they don't focus on it as much.

That's probably very accurate thinking about it actually. I beat BE twice under Apollo, got angry over how easy it was. Got over the game put it aside in about 25 hours, and the other day I asked a friend how he felt about the game and his response was, "Its a pretty good game, my only problem is AI just keeps kicking my $&%." I was blown away. I thought wish I was in your shoes buddy.

 

[edingess]

That's probably very accurate thinking about it actually. I beat BE twice under Apollo, got angry over how easy it was. Got over the game put it aside in about 25 hours, and the other day I asked a friend how he felt about the game and his response was, "Its a pretty good game, my only problem is AI just keeps kicking my $&%." I was blown away. I thought wish I was in your shoes buddy.

Well I'm practically like your friend but most likely because my first three games were about exploring affinities/techs. I wasn't on my A-game. Just wish the AI could have a wider range of maneuvers and actually take a tiny bit of advantages of bonuses. I never realized that in a Harmony vs. Harmony scenario, how annoying scatter could actually be.

 

[atc_chief]

Interesting how people focus on the AI 'cheating' (buffs etc) in the game, while most of us who come to the forums are 'cheating' a bit ourselves. We've got the collective conscious of hundreds of gamers who have played for thousands of hours. Exploits (aka highly refined build paths, slingshots, etc) are uncovered and documented, and then others copy those successful routines.

The internet barely existed when CIV I came out, and I had to work on the game using my own simple mind to figure it out. Exploits that have been documented on forums countless times may only have been uncovered and used by a single player in uncommon conditions or after tens or hundreds of hours of game play.

Additionally, thought the game has evolved, the formula is much the same. We've all learned by now how to specialize cities, maximize builds, 'play CIV war', work the tech tree, etc. Unless there are significant changes in the dynamics of the game, then it's pretty much wash/rinse/repeat.

I challenge you when CIV 6 comes out to play for 100+ hours before peeking on a forum to see what others are up to. Even something as simple as, "how long can a trade route be?" could take hours of game time to figure out vs. popping onto the forum for an answer based on someone else's research.

If you want to try a game with an AI that's recognized as being very solid, then check out Galactic Civilizations II. GalCiv III will be releasing within the year too if I'm not mistaken.

 

[Rellin]

I think that people on this forum vastly overestimate how easy these games are for most players. Most players probably don't think there's anything wrong with the AI, and the AI does not detract from those players' enjoyment of the game. It is tough to say that developers should spend resources developing something that only appeals to the people who are already most likely to buy the game. "Better AI" isn't a selling point to anyone who isn't already heavily involved in games like these. IMO, that's probably why they don't focus on it as much.

Except the number of people playing this game has nose dived since most people are unhappy and waiting for a patch to finish the game and correct some of the serious AI flaws.

 

[terraslayer]

For most people who haven't played many civ games, or are veterans of strategy games, they may find the AI a big hard, until they learn good strategies. However, for people who's been playing at Diety or other such high levels in previous Civ games, they will find the game very easy. The key is to try to make a balance for new players, and veterans alike.

Me personally, I never could go higher than monarch level in previous Civ games. However, I do agree that the AI for this game is a bit too easy compared to previous games (Aside from Vanilla unpatched Civ 5, which was pretty much exactly the same AI).

I doubt many games make really great AIs. Most times they follow the same thing like Civ games, giving the AIs bonuses to make them SEEM like their really good. In games like starcraft and Warcraft (I know these are old, but their the ones I can remember most at the top of my head), if you tried to make a multiplayer map, with AI controlled units, you had to give them a big head start to even be competent. If you just gave them the basic start of the command center and 5 gather units, they AI doesn't do anything at all, just sits there.

I don't know about other games, but it's much easier to program an AI in war games, where you tell it to shoot and kill anything in sight, than it is in strategy games, where there are multiple variables at to consider. Look at chess games, the higher in difficulty you go, the longer it even takes the CPU to make any moves.

Then there's obviously Human error, where the coders might try to make a competent AI, but make some mistakes, or leave some key things out ,and not notice it until post release, and bug reports.

 

[Pietato]

I do not think they even tried to fix the AI for this game. I watched as two AI armies moved backwards and forwards through miasma, until every unit was dead without even fighting.

 

[m15a]

I do not think they even tried to fix the AI for this game. I watched as two AI armies moved backwards and forwards through miasma, until every unit was dead without even fighting.

Conserving bullets. Very advanced tactics. (j/k)

 

[Bezurn]

There is a trade-off that you face when making an AI better. If you want the AI to take the same amount of time as you during its turn, then sure, I bet they can make an AI that could pose a challenge. Unlike a game of Chess, where the board is always the same geometry, the pieces all start in the same position, and each player can see exactly where all pieces are on the board, this is not true for Civ.

Adding degrees of arbitrary information forces the computer to have to think more like a human, and less about the most efficient way to get to attack a target from point A to point B. Chess has the advantage of having numerous opening book plays often programmed into a AI's database and odds of winning based on two different book openings against each other. Then there is information hiding, where the AI does not know true unit composition, and only sees the units you reveal to it. It would help if the AI was allowed to cheat in this department, and was able to "memorize" the board from previous sightings it had of your units, but I do not think it keeps a list like that. What does the AI do with this information, just use it to measure your force strength in a zone, or try to exploit any weakness in your formation.

To make an AI that could just play THAT type of game well would be a challenge.

But wait, there is more. Now the AI has to manage it's cities, manage its tech progression, its culture choices, diplomacy with other opponents, manage it's workers. There is a lot going on there, so if you want to AI to take 30 secs to 1 minute to manage its turn, then the AI will always be substandard.

Is there room for improvement? Sure there is. They have added new systems, and tweaked old ones from Civ 5. It will take time to identify where the AI is failing in using those systems and to nudge it in a more appropriate behavior.