The Problem with AI

AGEOD started selling their games on Steam earlier this year and there are a number of very strong complaints from potential Steam customers about the time the AI takes to resolve its turns. (3-5 minutes for TEAW and PON, the latter being probably one of the best strategy games ever) The AGEOD AI is one of the best there is but it seems that if the player has to wait for more than a few seconds, it will drive players bats.

Extending the time the AI takes to make its turn is one option for making the AI more competitive but I doubt it would please most who want a better AI and lightening fast turn resolution.

 

But even the simple things were not implemented, like a DO NOT SIT YOUR ENTIRE ARMY IN MIASMA UNTIL IT IS DEAD flag. Or a KILL PLAYER WHEN THEY HAVE VICTORY WONDER.

 

Pretending a recourse that exists does not exsist does not fix anything. The market changes, if devs can't handle their game being too easy because online resources they did not design it well enough for the era they designed it in.

GalCiv 2 ai was good because it gave the player the option to give the AI more system resources and more time to plot its moves. In Civ they have made no attempt at giving us the option. If I could choose to play against half as many AIs but give them each twice as long to plot their moves I would so it in a minute but Firaxis does not trust it's customers enough to give them the option.

 

I'm not sure that this is really such a big problem for a determined developer. There are several options I can think of immediately to improve this: Reducing the simulation detail for AI vs AI (abstracting some things, for example), calculating things like builds or strategic decisions in another thread during the player turn, pre-calculate common things so you don't have to go through all the steps in between at runtime, letting the player manage things while the AI plays where there's no conflict. I guess more would come to mind thinking about it more deeply.

That said, I agree with you that many players seem to have an absurdly low acceptance threshold for AI turn times. Perhaps people should be 'educated' to be more patient by playing board games. I and some of my friends often take 10 minutes or so for our turns when playing complex games.

Somebody mentioned AI War. I mean no disrespect to the devs, who did a great job there, but I'd like to note that it's probably a lot easier to write an AI for an asymmetric game like that. You can tailor the game rules to being handled by an AI and focus on its strengths rathet than have it play a complex ruleset meant to be challenging to a human player. Being a real-time game also means the player(s) will make less optimal decisions than in a turn-based game like Civ.

 

I came across this from the GalCivIII page, which is basically what you're talking about here.

"As previously mentioned, Galactic Civilizations was the first game to employ a multithreaded computer AI. Moreover, it made use of the Internet to (optionally) store player strategies so that the AI could then be improved on. If players opt in, we are able to see what technologies they researched and in what order. Same for ship designs, planetary improvements, even starbase configurations. As a result, we constantly improved the Galactic Civilizations II AI based on this data. Internet power has improved a great deal since 2006."

 

If it is indeed easier to do, I hope you are right and that we see more of it in future releases. I would assume it is likely a better approach than attempting to make an AI that can play as if human. I get the feeling that it hasn't been figured out how Arcen was able to do it so well and is a bit unique though. I don't really see being real time v turn based affecting the ability to apply the same logic and incorporate key elements of "emergent AI" * in some form for a TBS game however. I guess we'll find out when "Spectral Empire" ** is released by Arcen next Spring (iirc it will be their take on a traditional style TBS 4X). Actually, Stardock is also taking a similar design approach with Sorcerer King which is in Early Access.

*

Spoiler :

Emergent AI

In a quote from the developer, Chris Park, "'There are a ton of RTS games out there that are fun to play...The key weakness that they almost all share, in my opinion, is AI that falls apart in advanced play." AI War is developed using "Emergent AI", which differs from standard game AI in that it doesn't follow "decision trees", or in other words, the AI does not have an easily predictable pattern that can be exploited as in most games. Instead, the AI makes decisions on an individual unit level, a concept the developer coins as "Decentralized Intelligence".

Some other key factors that make the AI War "Emergent AI" unique are:
Strategic Tiers – The 3 levels of 'thinking' in AI War: Strategic, sub-level, and individual unit. The critical decisions made by the Emergent AI are based on a combination of these 3 factors.
Sub-Commanders – The way the individual AI units work together on a group level, in which they will "Do what's best for themselves, but also take into account what the group is doing." The developer compares this design principle to flocking behavior in the animal kingdom. He also points out that because of this design mechanic, the AI is "fairly untrickable", meaning they won't often fall victim to the same tactic.
Fuzzy Logic – This mechanic is designed to prevent players from "knowing" based on standard logic what the AI will do. The general concept of this design strategy is to prevent the AI from being predictable by programming them to not always make the "best" choice, but instead choosing from a larger selection of possibly sub-optimal choices, to prevent the player from being able to easily understand and therefore counter them.
Intelligent Mistakes – The intentional programming of the AI to make "mistakes" on lower difficulties, to make the game easier for the player. Since the AI at its full operating capacity is reportedly an extremely tough opponent to face, the developer designs the AI to be less of a challenge on easier settings.
Asymmetrical AI – A design which simplifies the AI reinforcement/attacking process by simply giving them reinforcements of units they need in important locations (on a timer), rather than over-complicating the design with the build process that humans use.

Source: http://en.wikipedia.org/wiki/AI_War:_Fleet_Command

**
[EDIT: now being called "Stars Beyond Reach" apparently http://www.arcengames.com/forums/index.php/topic,16512.0.html]

http://www.pcgamesn.com/stars-beyon...ars-beyond-reach-a-bonkers-sounding-sci-fi-4x

 

Except for the part where I don't cheat by using exploits. Nor perused the forums for help on crackign civ:be, I was quite confident that it would be just like civ5, all my tactics brought over to civbe just fine. Basically, find good locations, settle them, spam melee and shoot the AI.

I'm not that crazy good at civ series, I just dislike research cheats and so I stay at King difficulty and gotten very good at knowing what to expect. AI always sends insufficient forces for a war. Over 1500+ hours, only the AI has managed to fight me to a stalemate once and that was an runaway hiawatha in gods n kings.

And that was only because hiawatha went in for an war to end all wars instead of sending 10+ units to attack a random city and get its butt kicked.

AOW3 on other hand.... generally put up enough of a resistance unless I catch the AI by surprise or vice versa. Yeah to me too xD

 

it's not too expensive

games don't need 30 artists and 1 incompetent AI programmer
it just needs to be prioritized and staffed properly

for example, 1 college kid could write a better AI than anything firaxis has ever put out

 

There's many facets of AI in a game like civ. And there are some which IMO could be improved without too much effort. Not sure if these things still exist in BE (haven't played enough yet to experience it), but for example in Civ5 you had / have:
- AI which never expands (after losing a settler), although there's enough land nearby
- AI which does not improve their land for pretty much the entire game
- AI which consistently settles in bad locations (off the coast, ..)

This is pretty much all domestic AI, and improvements here could help to reduce the need for AI buffs.

Also trade and diplomacy AI should be possible to be improved without huge efforts.

Obviously the tactical and strategic AI is an entirely different topic.

 

It's actually a fairly simple problem, guys and girls at uni do it as their 4th year honours project, we have our first go at it in second year and do fairly well? The problem is not our code, it's the physical robot which has to deal with a world that can't be accurately described by a mathematical model

Regarding the AI, I couldn't care less if the AI cheats. But despite its cheating, it's still incompetent. I can't even be bothered.


Also a logic based AI is probably not the way to go. There are other tools to use and take advantage of. Logic based programming will just make the problem much harder than it is.

Incidentally, if you were an AI programmer, there's a much larger chance you would work in Academia or other industry, such as finance or the myriad of engineering and science applications, rather than earning comparatively less making video games.

 

GalCiv2's AI is decent (not fantastic - decent) because nearly everything about the game was tweaked and designed so that the AI could maneuver around it. It's less player-focused. It's less about creating player atmosphere.

That's why your weapon techs are Laser 1, Laser 2, Laser 3, and Laser 4. And yes, they are literally named that way.

 

Interesting. I more quickly got bored with GalCiv2, then Civ4, since GalvCiv2 was a less challenge in higher difficulty levels then Civ4.

If anying, I think that GalCiv2 great AI is just gamer myth, supported by boasting of developers themselves. Including that GalCiv2 AI does not cheat, which is blatantly false.

 

If you want an example of how good AI can be created for games, go look at Age of Empires II custom AIs. There's a small but dedicated scripting community, and they've made AIs which will destroy anyone who isn't very good at the game, without bonuses. Add the bonuses in and they become a real challenge for even good players.

It's only possible because the devs left the AI open to player scripting, and players actually learn how to play the game well. I'm pretty sure the Civ5 team suck balls at their own game. Which is why I think it's super important to leave as much of the AI code exposed to modding as possible.

 

To be fair to the developers, well, it's testament to the quality of the game that the developers aren't experts at it, right?

But yes I agree, opening up the AI to modding without having to manually reverse engineer the thing would be right swell.

 

The Civ5 AI can be pretty extensively modded since the DLL source code was made available. The same will be true for BE

 

To be fair, the devs are busy actually making the game. They don't have the time that gamers have to just sit around and play the game non-stop. Gamers can put all their focus into playing the game non-stop and studying every trick and strategy.